2 The Quenching and Partitioning Process-Background and Recent Progress

SK5弹簧钢薄带的工艺及热处理研究摘要：回火通常可分为三种，即低温回火、中温回火和高温回火，三种回火温度、回火后组织及其性能、主要适用对象如下表所示。

一般弹簧钢的热处理工艺是淬火+中温回火处理,可以获得回火屈氏体,在保证获得良好的弹性极限基础上,还具有良好的韧性和塑性,如85钢,其规范的回火工艺是380℃~440℃。

回火保温时间的选择应尽量保证工件各部分温度均匀，同时保证组织转变充分，并尽可能降低或消除内应力。

回火过程一般在最初0.5h内硬度变化很剧烈，超过2h后，硬度变化很小，因此，一般弹簧钢回火时间均为1~2h。

[1]将淬火后的试样分别在150℃、200℃、250℃、300℃、350℃下进行回火处理，比对金相进行分析；测量硬度值；进行拉伸试验，记录断面收缩率、拉伸率、最大拉伸力等数据；进行试验，记录最大力、最大弯曲强度、弯曲弹性模量等数据。

相关研究发现，抗拉强度与硬度随着回火温度的上升而下降,不同温度阶段的下降趋势不同;断面收缩率在不同的温度范围里有不同的变化趋势。

弹簧钢最重要的性能要求是良好的弹性极限和硬度，通常硬度很高时，断面收缩率会较低，此时的回火温度并不合适，应该选取回火时强度和韧性都较好的回火温度。

关键词：SK5弹簧；热处理；回火温度引言弹簧在生产中的主要功能是吸收冲击能量及缓和振动。

弹簧钢在工业设备、机械零件、铁路、汽车、发动机等行业当中应用非常广泛。

[1]随着我国现代工业能力的增强、汽车、铁路工业稳步发展，工业设备向着载荷量大、运行速度高的方向发展，机械零件承受这越来越高的载荷和频率。

这些现实的需求，要求弹簧钢具备良好的力学性能（高弹性极限σE、屈服极限σs和屈强比σ0.2/σb）、极佳的抗疲劳性能、淬透性，以及优异的物理和化学性能(耐热性、耐腐蚀性、抗氧化性等)。

目前，国内外弹簧钢的研究集中在提升弹簧钢的强度和使用寿命上，主要研究合金成分、组织形态、生产工艺、热处理等问题[2]。

1.OracleDBA面试题之一解释冷备份和热备份的不同点以及各自的优点解答：热备份针对归档模式的数据库，在数据库仍旧处于工作状态时进行备份。

而冷备份指在数据库关闭后，进行备份，适用于所有模式的数据库。

热备份的优点在于当备份时，数据库仍旧可以被使用并且可以将数据库恢复到任意一个时间点。

冷备份的优点在于它的备份和恢复操作相当简单，并且由于冷备份的数据库可以工作在非归档模式下,数据库性能会比归档模式稍好。

（因为不必将archive log写入硬盘）2. 你必须利用备份恢复数据库，但是你没有控制文件，该如何解决问题呢？解答：重建控制文件，用带backup control file 子句的recover 命令恢复数据库。

3. 如何转换init.ora到spfile?解答：使用create spfile from pfile 命令4. OracleDBA面试题：解释data block , extent 和 segment的区别（这里建议用英文术语）解答：data block是数据库中最小的逻辑存储单元。

当数据库的对象需要更多的物理存储空间时，连续的data block就组成了extent . 一个数据库对象拥有的所有extents 被称为该对象的segment.5. 给出两个检查表结构的方法解答：1、DESCRIBE命令2、DBMS_METADATA.GET_DDL 包6. 怎样查看数据库引擎的报错解答：alert log.7. 比较truncate和delete 命令解答：两者都可以用来删除表中所有的记录。

区别在于：truncate是DDL操作，它移动HWK，不需要 rollback segment .而Delete是DML操作需要rollback segment 且花费较长时间.8. 使用索引的理由解答：快速访问表中的data block9. 给出在STAR SCHEMA中的两种表及它们分别含有的数据解答：Fact tables 和dimension tables. fact table 包含大量的主要的信息而dimension tables 存放对fact table 某些属性描述的信息10. FACT Table上需要建立何种索引？解答：位图索引（bitmap index）11.OracleDBA面试题：给出两种相关约束?解答：主键和外键12. 如何在不影响子表的前提下，重建一个母表解答：子表的外键强制实效，重建母表，激活外键13. 解释归档和非归档模式之间的不同和它们各自的优缺点解答：归档模式是指你可以备份所有的数据库 transactions并恢复到任意一个时间点。

When it comes to doing homework after school,the timing can vary greatly depending on individual schedules,school hours,and personal preferences.Here are some general guidelines and tips for managing homework effectively:1.Immediate After School:Some students find it helpful to start their homework as soon as they get home from school.This can be a good strategy if youre still fresh and can focus well immediately after a day of learning.2.After a Break:If you feel drained after school,it might be beneficial to take a short break before starting your homework.A break could involve a snack,a short walk,or a relaxation activity to recharge.3.Dedicated Homework Time:Establish a routine by setting aside a specific time each day for homework.This could be right after school,after dinner,or before bedtime, depending on what works best for you.4.Prioritize Tasks:If you have multiple assignments,prioritize them based on due dates and difficulty.Tackle the most challenging or timeconsuming tasks first when your energy levels are highest.5.Break It Down:Large assignments can be overwhelming.Break them down into smaller,manageable tasks and tackle them one at a time.6.Create a Study Environment:Find a quiet,comfortable place to do your homework where you can focus without distractions.e a Planner or Calendar:Keep track of assignments and due dates using a planner or digital calendar.This can help you stay organized and manage your time effectively.8.Avoid Procrastination:Its easy to put off homework,but this can lead to stress and poor performance.Try to start your work as soon as possible to avoid lastminute rushes.9.Ask for Help:If youre struggling with a particular subject or assignment,dont hesitate to ask your teacher,a tutor,or a classmate for help.10.Take Regular Breaks:Studies show that taking short breaks can improve focus and e techniques like the Pomodoro Technique,where you work for25 minutes and then take a5minute break.11.Stay Healthy:Make sure youre getting enough sleep,eating well,and exercisingregularly.These factors can greatly affect your ability to concentrate and complete homework efficiently.12.Reflect on Your Progress:At the end of each week,take some time to review what youve accomplished and what you can improve on for the next week.Remember,the best time to do homework is when you can focus and be productive.Its important to find a routine that fits your lifestyle and helps you manage your schoolwork effectively.。

CCF推荐的国际学术会议和期刊目录修订版发布CCF(China Computer Federation中国计算机学会)于2010年8月发布了第一版推荐的国际学术会议和期刊目录，一年来，经过业内专家的反馈和修订，于日前推出了修订版，现将修订版予以发布。

本次修订对上一版内容进行了充实，一些会议和期刊的分类排行进行了调整，目录包括：计算机科学理论、计算机体系结构与高性能计算、计算机图形学与多媒体、计算机网络、交叉学科、人工智能与模式识别、软件工程/系统软件/程序设计语言、数据库/数据挖掘/内容检索、网络与信息安全、综合刊物等方向的国际学术会议及期刊目录，供国内高校和科研单位作为学术评价的参考依据。

目录中，刊物和会议分为A、B、C三档。

A类表示国际上极少数的顶级刊物和会议，鼓励我国学者去突破；B类是指国际上著名和非常重要的会议、刊物，代表该领域的较高水平，鼓励国内同行投稿；C类指国际上重要、为国际学术界所认可的会议和刊物。

这些分类目录每年将学术界的反馈和意见，进行修订，并逐步增加研究方向。

中国计算机学会推荐国际学术刊物（网络/信息安全）一、 A类序号刊物简称刊物全称出版社网址1. TIFS IEEE Transactions on Information Forensics andSecurity IEEE /organizations/society/sp/tifs.html2. TDSC IEEE Transactions on Dependable and Secure ComputingIEEE /tdsc/3. TISSEC ACM Transactions on Information and SystemSecurity ACM /二、 B类序号刊物简称刊物全称出版社网址1. Journal of Cryptology Springer /jofc/jofc.html2. Journal of Computer SecurityIOS Press /jcs/3. IEEE Security & Privacy IEEE/security/4. Computers &Security Elsevier http://www.elsevier.nl/inca/publications/store/4/0/5/8/7/7/5. JISecJournal of Internet Security NahumGoldmann. /JiSec/index.asp6. Designs, Codes andCryptography Springer /east/home/math/numbers?SGWID=5 -10048-70-35730330-07. IET Information Security IET /IET-IFS8. EURASIP Journal on InformationSecurity Hindawi /journals/is三、C类序号刊物简称刊物全称出版社网址1. CISDA Computational Intelligence for Security and DefenseApplications IEEE /2. CLSR Computer Law and SecurityReports Elsevier /science/journal/026736493. Information Management & Computer Security MCB UniversityPress /info/journals/imcs/imcs.jsp4. Information Security TechnicalReport Elsevier /locate/istr中国计算机学会推荐国际学术会议（网络/信息安全方向）一、A类序号会议简称会议全称出版社网址1. S&PIEEE Symposium on Security and Privacy IEEE /TC/SP-Index.html2. CCSACM Conference on Computer and Communications Security ACM /sigs/sigsac/ccs/3. CRYPTO International Cryptology Conference Springer-Verlag /conferences/二、B类序号会议简称会议全称出版社网址1. SecurityUSENIX Security Symposium USENIX /events/2. NDSSISOC Network and Distributed System Security Symposium Internet Society /isoc/conferences/ndss/3. EurocryptAnnual International Conference on the Theory and Applications of Cryptographic Techniques Springer /conferences/eurocrypt2009/4. IH Workshop on Information Hiding Springer-Verlag /~rja14/ihws.html5. ESORICSEuropean Symposium on Research in Computer Security Springer-Verlag as.fr/%7Eesorics/6. RAIDInternational Symposium on Recent Advances in Intrusion Detection Springer-Verlag /7. ACSACAnnual Computer Security Applications ConferenceIEEE /8. DSNThe International Conference on Dependable Systems and Networks IEEE/IFIP /9. CSFWIEEE Computer Security Foundations Workshop /CSFWweb/10. TCC Theory of Cryptography Conference Springer-Verlag /~tcc08/11. ASIACRYPT Annual International Conference on the Theory and Application of Cryptology and Information Security Springer-Verlag /conferences/ 12. PKC International Workshop on Practice and Theory in Public Key Cryptography Springer-Verlag /workshops/pkc2008/三、 C类序号会议简称会议全称出版社网址1. SecureCommInternational Conference on Security and Privacy in Communication Networks ACM /2. ASIACCSACM Symposium on Information, Computer and Communications Security ACM .tw/asiaccs/3. ACNSApplied Cryptography and Network Security Springer-Verlag /acns_home/4. NSPWNew Security Paradigms Workshop ACM /current/5. FC Financial Cryptography Springer-Verlag http://fc08.ifca.ai/6. SACACM Symposium on Applied Computing ACM /conferences/sac/ 7. ICICS International Conference on Information and Communications Security Springer /ICICS06/8. ISC Information Security Conference Springer /9. ICISCInternational Conference on Information Security and Cryptology Springer /10. FSE Fast Software Encryption Springer http://fse2008.epfl.ch/11. WiSe ACM Workshop on Wireless Security ACM /~adrian/wise2004/12. SASN ACM Workshop on Security of Ad-Hoc and Sensor Networks ACM /~szhu/SASN2006/13. WORM ACM Workshop on Rapid Malcode ACM /~farnam/worm2006.html14. DRM ACM Workshop on Digital Rights Management ACM /~drm2007/15. SEC IFIP International Information Security Conference Springer http://sec2008.dti.unimi.it/16. IWIAIEEE International Information Assurance Workshop IEEE /17. IAWIEEE SMC Information Assurance Workshop IEEE /workshop18. SACMATACM Symposium on Access Control Models and Technologies ACM /19. CHESWorkshop on Cryptographic Hardware and Embedded Systems Springer /20. CT-RSA RSA Conference, Cryptographers' Track Springer /21. DIMVA SIG SIDAR Conference on Detection of Intrusions and Malware and Vulnerability Assessment IEEE /dimva200622. SRUTI Steps to Reducing Unwanted Traffic on the Internet USENIX /events/23. HotSecUSENIX Workshop on Hot Topics in Security USENIX /events/ 24. HotBots USENIX Workshop on Hot Topics in Understanding Botnets USENIX /event/hotbots07/tech/25. ACM MM&SEC ACM Multimedia and Security Workshop ACM。

Energy Conversion and Management Heating systems with PLC and frequency controlSalah Abdallah, Riyad Abu-MallouhAbstract: In this work, medium capacity controlled heating system is designed and constructed. The programming method of control of heating process is achieved by means of integrated programmable logic controller (PLC) and frequency inverter (FI). The PLC main function is to determine the required temperatures levels and the related time intervals of the heating hold time in the furnace. FI is used to control the dynamicChange of temperature between various operating points. The designed system show the capability for full control of temperature from zero to maximum for any required range of time in case of increasing or decreasing the temperature. All variables of the system will be changed gradually until reaching their needed working points. An experimental study was performed to investigate the effect of tempering temperature and temperingtime on hardness and fatigue resistance of 0.4% carbon steel. It was found than creasing tempering temperature above 550 _C or tempering time decreases the hardness of the material. It was also found that there is a maximum number of cycles to which the specimen can survive what ever the applied load was.Keywords:Heat treatment;PLC control;Frequency control;Tempering1. IntroductionHeat treatment is the controlled heating and cooling of metals to alter their physical and mechanical properties without changing the product shape. Heat treatment is often associated with increasing the strength of material, but it can also be used to alter certain manufacturability objectives such as improve machining, improve formability, and restore ductility after a cold working operation.Thus it is a very enabling process that can not only help other manufacturing process, but can also improve product performance by increasing strength or other desirable characteristics [1,2].A new single-switch parallel resonant converter for induction heating was introduced in [3]. The circuit consists of an input LC filter, a bridge rectifier and only one controlled power switch.The switch operates in a soft communication mode and serves as a high frequency generator.A voltage-fed resonant LCL inverter with phase shift control was presented in [4]. It was seen that the control strategy offered advantages in the megahertz operating region, where a constant switching frequency is required. The inverter steady state operation is analyzed using fundamental frequency analyses.A cost-effective high efficiency inverter with phase-shifted pulse modulation scheme was proposed for medium power (5–30) kW induction heating applications isdiscussed in [5]. The proposed inverter accomplishes soft switching operation over a wide power regulation range. The actual power conversion efficiency reached was96.7%.A control method of reducing the size of the dc-link capacitors of a converter–inverter system was presented in [6]. The main idea is to utilize the inverter operation status in the current control of the converter. This control strategy is effective in regulating the dc-voltage level. Even the dc-link capacitor is arbitrarily small and the load varies abruptly. In [7] a method was proposed to accurately predict the minimum required temperature recovery considering repeatability and accuracy of the leak detector by investigating the relation between temperature recovery time and theoretical thermal time constant for various test volumes and applied pressures using PLC system. In [8] a methodology was demonstrated to design a PLC program that organized the relation between the physical inputs and outputs of the pumping tools in manufacturing systems. In [9] an experimental study was performed to investigate the effect of using two axes tracking with PLC control on the solar energy collected. The two axes tracking surface showed better performance with an increase in the collected energy up to 41% compared to the fixed surface. The PLC main function is to control the required temperature levels and the related time intervals of the heating hold time in the furnace [10]. Frequency inverter is used to control the dynamic change of temperature between various operating points [11,12]. This integration of the PLC and frequency control shows the capability for full control of temperature from zero to maximum in dynamic and static conditions, in case of increasing or decreasing the temperature .The properties and microstructure as a function of tempering time at intercritical temperatures in HY-80 steel castings were evaluated by [13]. They varied the time for which the steel was held in the inter critical temperature range. An important finding of this study is that, contrary to normal behavior during tempering HY-80 steel tempered in the inter critical range demonstrates a severe loss of toughness; which can be exaggerated for longer hold times and higher temperatures.A fractography survey on high cycle fatigue failure in Fe–C–Cr–Mo–X alloys was made by [14]. They found that various parameters are likely to influence high cycle fatigue failures, the most significant one dealing with the nature and location of embedded precipitates and the forging reduction ratio. The ageing effect on cyclic plasticity of a tempered martensitic steel was studied by [15]. They carried out specific isothermal cyclic deformation tests on a tempered martensitic steel 55NiCrMoV7 at four hardness levels in the temperature range 20–600 _C. They found that the cyclic stress response generally shows an initial exponential softening for the first few cycles, followed by a gradual softening without saturation, hardness dramatically decreases when the specimen is simultaneously subjected to ageing and fatigue at elevated temperature, cyclic softening intensity increases with testing temperature from 300 to 600 _C, but the maximal softening intensity occurs at room temperature.Back-propagation neural networks were used by [16] to optimize the heat treatment technique of high-vanadium high-speed steel including predictions of retainedaustenite content, hardness and wear resistance according to quenching and tempering A novel concept for the heat treatment of, different to customary quenching and tempering was proposed by [17]. This novel treatment has been termed ‘quenching an d partitioning’ (Q&P), to distinguish it from quenching and tempering and can be used to generate microstructures with austenite combinations giving attractive properties.Reversible transformation, ageing and low-temperature tempering of iron–carbon were studied by [18]. In this study, reversible transformation was observed in Fe-based high-carbon alloys at temperatures when none lattice defects (including ) could diffuse. Formation and behaviorof carbon-vacancy clusters was studied and discussed. Chemical composition of hexagonal e-carbide was determined as Fe3C. The effect of particle size and e-carbide/ orientation relationship on the transformation was discussed.1.The heating system design and controlIn this work, the design of PLC and frequency controlled heating system were performed using an open loop and programming method of control in which stored instructions in memory of PLC was used to control the actuation of heating process. The block diagram of the hardware components of the automatically controlled heating system is shown in Fig. 1.Personal computer is used to write the control program then download it to the PLC [19,10] through communication cable. The PLC is S7-200 type, which has 12 inputs, 8 outputs and 220 VAC supply S7-200 uses ladder logic diagram programming language describes in Refs.. The PLC main function is to instruct the analog unit to go on or off and e the required percentage output and the related hold time intervals. The digital output of PLC is ranging from zero to 32,700 quantization levels. The analog unit function is to transfer the digital output value at the output of PLC into analog value, which ranging from zero to 10 VDC at the output of analog unit. In the program a different percentages of output voltage is supplied to the furnace by the frequency inverter which is originally stated by the analog unit output where 0 DVC equals 0% at the output of the frequencyinverter and 10 VDC equals 100% at the output of frequency inverter. Frequency inverter is a single phase input, three phase output, 220 VAC rated output voltage, 50 Hz rated output frequency and 3 kW power.In this work a frequency inverter of type SINAMICS G110 is used [11]. Frequency inverter according to the different incoming instructions of PLC through analog unit operates the three phase heater with the required percentage of voltage and frequency. Parameter unit is a type of programmers used to program the ramp up and ramp down time between each two controlled levels. So, frequency inverter has two types of commands: (1) Type of commands which is supplied by the PLC to the analog unit then to the frequency inverter to state the hold time intervals. (2) Type of commands which is supplied by parameter unit td control the ramp up and ramp down time to make a soft transition conditions between various operating levels [20].The furnace consists of a three phase heater in a room which is 3 kW rated power, with 220 V AC rated input voltage, 50 Hz rated frequency and star connection.3. Mathematical description of frequency controlled heating SystemThe main important parts of the PLC and frequency controlled three phase heating system are the frequency inverter and the three phase furnace.3.1. Modeling of the frequency inverterFrom all types of frequency inverters for regulation of AC motors, the more important is frequency inverter with clear dc part as shown in Fig. 2. This type of frequency inverter has high technical and economical specifications.The equation of the frequency control channel can be written asF=kpup(1)where kp is the inverter amplification coefficient by frequency control channel and up the input voltage of frequency generator. The equation of the voltage control channel will beU=kvuu / (2)where uu and u is the output and input voltages of the inverter.kv = 2/（π3）.The dc circuit of thyristor frequency inverter contains LdC-filter, where Ld is the inductance of the filter and C is the capacitance of the filter condenser.The equation of the dc circuit will beUu=Ed-Rdid-Ld(did/dt) (3)WhereId=iu-icIc=C(duu/dt) (4)where Ed is the rectified voltage at the output of converter, Rd is the active resistance of coil and id is the coil current, iu is the input current of the inverter and ic the current of the condenser.3.2. Modeling of the furnaceThe right hand side of Fig. 2 illustrates a furnace consisting of a three phase heater in a room. The three phase heater emits heat at a rate of q1 and the room loses heat at a rate of q2.q1=3khiu(5)where Kh is the heater coefficient.Assuming that the air in the room is at a uniform temperature T and that there is no heat storage in the walls of the room, we can derive an equation describing the time rate of change in room temperature.Q1-Q2=Co(dT/dt) (6)where Co is the thermal capacity of the air in the room.If the temperature inside the room is T and that outside the room is To, thenq2=(T-To)/Ro (7)where Ro is the of the walls. Substituting for q2 givesq1-(T-To)/Ro=codT/dtHenceRoCodT/dt+T=Roq1+To (8)4. Experimentation and results4.1. Testing the control system performanceIn order to test the ability of the designed control system to perform the desiredtemperature control properly, successive heat treatment experiments were performed and their results are described as follows. Fig. 3 shows the input frequency vs. time and Fig. 4 shows the temperature vs. time for the two specimens 1 and 2 where this process represent a heat-treatment process carried out with ramping function only (i.e. the furnace and frequency inverter only are used), the holding time in this case was counted off from the moment the furnace is actuated (form room temperature). That is why the graph starts at 20 _C and then temperature rise up using the ramp up function in the frequency inverter on a range of time for1 h ramp up, then temperature reaches 400 _C and maintained for 2 h. Then a ramp down time for 1 h is carried out. Fig. 5 shows the input frequency vs. time and Fig. 6 shows the temperature vs. time for the second two specimens 3 and 4 where this process represent a controlled heat-treatment process (i.e. the furnace and frequency inverter integrated with the PLC and an analog unit are used), and as referred before the holding time in this case was counted off from the moment the furnace is actuated (form room temperature). That is why the graph starts at 20 _C and then temperature rise up using the ramp up function and in steps sequence depending on the PLC program, the first ramp up process takes approx. 2 h then temperature decrease to 200 _C in half hour and ramp ups again in steps for 1 h to reaches 400 _C again and finally decreases to room temperature in a process that takes 0.5 h.4.2. Heat treatment experimentsSixty specimens were machined according to the standard dimensions suitable for the fatigue testing machine. These dimensions are shown in Fig. 7. These specimens were made of 0.4%C steel.Four of these specimens were left without any further heat treatment. The rest were heat treated. The heat treatment process of typical steel involves heating the object (austenizing) and then causing a quick and sharp drop in its temperature (quenching). Together, these two processes produce an extremely hard microstructure in medium-carbon or high-carbon steels, which can then be ‘‘tempered” to prevent the material from shattering. As mentioned above, the first stage of the heat treatment is water quenching. This was done by heating the specimens to 860 _C for a period of 1 h. That results in transforming the microstructure to austenite (face centered cubic). This heating process was followed by water quenching the specimens to room temperature. This means that the microstructure was transformed to martensite (body centered tetragonal). These quenched specimens were dried and kept in plastic bags to prevent corrosion of the specimens. The second heat treatment stage was tempering. Chemically, the process of tempering is a transformation from metastable martensite to ferrite and cementite. This change is accomplished by annealing at a temperature below the austenizing temperature, but high enough that nucleation of cementite particles can occur. The formation of cementite draws carbon from the surrounding alloy, allowing it to transform to ferrite. Cooling the object ends the annealing process, stopping cementite formation by slowing down the diffusion of carbon.In this paper, the effect of tempering temperature and tempering time on hardness and fatigue resistance of 0.4% carbon steel was studied. Specimens were divided into four categories accordingAccording to tempering temperature (400, 500, 550 and 600 _C), and tothree subcategories according to tempering time (1, 2 and 3 h).The hardness of the specimens was then tested using standard brinell testing machine and the results are recorded and plotted in Fig. 8.The specimens were then tested using standard fatigue testing machine and the results are reported and plotted in Figs. 9–11.5. DiscussionIn this paper, the effect of tempering temperature and tempering time on hardness and fatigue resistance of 0.4% carbon steel was studied. Fig. 8, shows that, for temperatures less than 550 _C,the effect of tempering temperature on hardness of the tested specimens is small. But it becomes significant for temperatures greater than 550 _C (4%–8% decrease in hardness). Accordingly, for tempering to reduce the hardness effectively, the tempering temperature should be greater than 550 _C.For tempering temperatures greater than 550 _C, we note that hardness decreases linearly with increasing temperature. In regard to the tempering time, this figure shows that increasing the tempering time decreases the resulting hardness (4%–8% decrease in hardness).For now, if we have a look at fatigue test results, we can see that the number of cycles to failure increases with decreasing the applied cyclic load. We can also note that the number of cycles to which the specimen can survive have a maximum value over which the specimen will fail whatever the applied load was. This can be clearly seen from the sudden sharp increase of the slope of the curve as number of cycles increases.It can also be seen in the figure that this maximum possible life of the specimenincreases with increasing temperature. To demonstrate this behavior clearly, a plot of Log(N) as a function of temperature is shown in Fig. 12 for a force of 8N. (5.5 _ 103 cycles for T = 400 _C to 1 _ 106 cycles for T = 600 _C). To investigate the effect of the tempering time on the fatigue behavior of 0.4% carbon steel, the same results shown in Figs. 9– 11 is redrawn in Figs. 13–16 shown.These figures show that for 8N load and T = 500 _C, the number of cycles is 66,958 for 3 h tempering period and 38,974 for 1 h tempering period. That means that the number of cycles increased by 72% due to increasing tempering time.6. ConclusionIn this work, medium capacity controlled heating system is designed and constructed using PLC and frequency control. The designed system shows the capability for full control of temperature from zero to maximum and from maximum to zero for any required range of time. As results for the experimentation of the controlled heating system it was found for significant decrease in the hardness of 0.4% carbon steel to be achieved, tempering temperature should be more than 550 _C. Increasing the reheat time decreases the hardness of 0.4% carbon steel. There is a maximum number of cycles to which the specimen can survive whatever the applied load was. This maximum number of cycles increases with increasing tempering temperature. Increasing the tempering time increases fatigue life of 0.4% carbon steel. References[1] Rajan TV, Sharma CP. Heat treatment principle and techniques. New Delhi: Prentice Hall of India; 1992.[2] Arafat A, Abu Hamid E, Abdel Fattah AK. Royal Scientific Society, introduction of heat treatment technology in designing and manufacturing of spare parts of local industries in Jordan. Higher Council Sci Technol Amman – Jordan, 1999.[3] Shenkman A, Axelrod B, Berkovich Y. Improved modification of the si gleswitch AC–AC converter for induction heating applications. IEE Proc –Electr Power Appl 2004;151(1):1–4.[4] Mollov SV, Theodoridis M, Forsyth AJ. High frequency voltage-fed inverter with phase-shift control for induction heating. IEEE 2003:12–8.[5] Kifune H, Hatanaka Y, Nakaoka M. Cost effective phase shifted pulse modulation soft switching high frequency inverter for induction heating applications. IEE Proc – Electr Power Appl 2004;151(1):19–25.[6] Jung J, Lim S, Nam K. A feedback linearizing control scheme for a PWM converter–inverter having a very small DC-link capacitor. IEEE Trans Ind Appl 1999;35(5).[7] Harus LG, Cai M, Kawahima K, Kagawa T. Determination of temperature recovery time in differential-pressure-based air leak detector. Measur Sci Technol 2006;17:411–8.[8] Abdallah S, Al-Dahwi A. A method to design an automated pump plants with PLC control. Issued in the publications of the international engineering conference ‘‘Mutah 2004”, Jordan, in April 2004.[9] Abdalah S, Nijmeh S. Two axis sun tracking system with PLC control. Issued Energy Conversion Manag 2004;45:1931–9.[10] Siemens Sinamics G110 operating instruction manual. User documentation issued 04/02.[11] Siemens PLC S7-200 operating instruction manual. Issued 06/02.[12] Bolton W. Mechatronics electronic control systems in mechanical engineering. Addison Wesley Longman Limited; 1999.[13] Stephen D Funni, Michelle G Koul, Angela L Moran. Evaluation of prosperities and microstructure as a function of tempering time at intercritical temperatures in HY-80 ‘‘steel” castings. Mechanical Engineer ing Department, US Naval Academy; 2007.[14] Tchoufang J, Beckers J. Fractography survey on high cycle fatigue failure: crack origin characterization and correlations between mechanical tests and microstructure in Fe-C-Cr-Mo–X alloys. Universite’ de Lie’ge; September 2006. [15] Zhang Z, Delagnes D, Bernhart G. Ageing effect on cyclic plasticity of a tempered martensitic steel. China: Dalian Maritime University; 2006.[16] Xu L, Xing J, Wei S, Zhang Y. Ruilong. Optimization of heat treatment technique of high-vanadium high-speed steel based on back-propagation neural networks. Henan University of Science and Tech; 2005.[17] Edmonds D, He K, Rizzo F, Cooman B, Matlock D, Speer J. Quenching and partitioning martensite – a novel steel heat treatment. University of Leeds; 2006. [18] Kaputkin D. Reversible martensitic transformation ageing and lowtemperature tempering of iron-carbon martensite. Technology University; 2006.[19] Mandado E, Macros J, Perze S. Programmable logic devices and logic controllers. Prentice-Hall; 1996.[20] Abdallah S. The regulation of starting and breaking dynamic characteristics in water pumping systems with open-loop control. In: Proceedings of the 8th Cairo University conference on mechanical design and production. Cairo, Egypt; January 2004. p. 4–6.采用PLC和变频控制的加热系统Salah Abu mallouh *阿卜杜拉，利雅得机械与工业工程系，应用科学大学，安曼11931，乔丹关键词：热处理PLC控制频率控制回火摘要：在这项工作中，中等容量控制的加热系统的设计与构建。

英语作文批改评语大全英语作文1When it comes to correcting and providing comments on English compositions, there are several common types that teachers and reviewers often use. Firstly, in terms of grammar, if there is a subject-verb disagreement or a tense error, the comment might be like, "Pay attention to the agreement between the subject and the verb in this sentence. Here, the verb form should be changed to match the subject." Or, "The tense you used is incorrect. This context requires the past tense instead of the present tense." Secondly, regarding vocabulary, if the words used are not rich or accurate, a suggestion could be, "Try to expand your vocabulary and use more precise words to express your ideas. For example, instead of using 'good', you could use 'excellent', 'wonderful' or 'marvelous' to make your writing more vivid." When it comes to the logical structure, if the paragraphs are not divided properly or the theme is not prominent, the comment might be, "The organization of your composition is a bit chaotic. Each paragraph should have a clear focus and contribute to the overall theme. Consider re-arranging your ideas to make the logic more coherent and the flow more smooth." All in all, these types of comments aim to help students improve their English writing skills and make their compositions more refined and compelling.2English composition correction comments play a crucial role in students' writing improvement. Firstly, they help students recognize their own problems. For instance, if a student consistently makes grammar mistakes, the comments can point them out precisely, allowing the student to be aware of the issue and take steps to correct it. Secondly, these comments guide students to form good writing habits. By providing suggestions on organization, vocabulary usage, and sentence structure, students can gradually develop a more logical and coherent writing style. Moreover, positive comments act as a powerful incentive for students to strive forward. When they receive praise for their creativity or clear expression, it boosts their confidence and encourages them to continue working hard and improving. In conclusion, the importance of English composition correction comments cannot be overstated. They are like guiding lights that illuminate the path for students to become better writers, enabling them to express themselves more accurately and effectively in the English language.3When it comes to providing effective feedback on English compositions, several key points should be kept in mind. Firstly, it is essential to precisely point out the problems existing in the essay. Forinstance, if there are grammar mistakes such as incorrect verb tenses or wrong word choices, they should be clearly identified. Secondly, clear directions for improvement need to be given. Instead of simply saying "It's not good", it's better to suggest specific ways like "You could expand your description in this paragraph by adding more details". Encouraging language also plays a vital role in boosting students' confidence. Phrases like "You have made great progress in organizing your thoughts" or "Your creativity in this story is impressive" can inspire them to keep writing. Moreover, combining real examples to illustrate the comments makes them more persuasive. For example, if a student has difficulty structuring an argumentative essay, you can refer to a well-structured sample and explain how it works. In conclusion, giving useful and motivating feedback not only helps students improve their writing skills but also nurtures their passion for English writing.4When it comes to correcting English compositions, different levels of writing require distinct approaches and comments. For beginner-level compositions, the main focus lies in correcting basic grammar and vocabulary mistakes. For instance, simple verb tenses like the present simple and past simple might be frequently misused, and common words might be spelled wrongly. The comments should clearly point out these errors and provide correct forms to help students build a solid foundation.For intermediate-level compositions, the attention shifts towards the logic and completeness of the content. It's important to check if the paragraphs are well-organized and if the ideas flow smoothly. Are there enough supporting details to explain the main points? The comments should offer suggestions on how to improve the structure and coherence of the essay.When it comes to advanced-level compositions, the emphasis is on refining the language and exploring the depth of thought. Are the expressions sophisticated and precise? Does the composition present unique and profound insights? The comments at this level should encourage students to think critically and express themselves more elegantly.In conclusion, the key to effective composition correction lies in tailoring the comments to the specific level of the student's writing, guiding them to continuously improve and progress.5When it comes to the assessment of English compositions, the comments provided play a crucial role in enhancing the quality of teaching. Through effective feedback in the form of comments, teachers can gain insights into students' writing abilities and areas that need improvement. This enables them to optimize their teaching methods accordingly. For instance, if a majority of students struggle with grammar, the teacher canfocus more on grammar explanations and exercises in subsequent lessons.The comments also facilitate communication between teachers and students. They serve as a bridge, allowing teachers to express their expectations and guidance, while students can better understand their strengths and weaknesses. This two-way interaction helps build a positive learning atmosphere and strengthens the bond between them.Furthermore, the accumulation and analysis of these comments can drive the reform and development of English writing teaching. By identifying common problems and trends, educators can update teaching materials and approaches to better meet the needs of students. In conclusion, the comments on English compositions are not just words on paper but powerful tools that have a significant impact on improving the quality of teaching and promoting the progress of students in English writing.。

ANNEX CDESCRIPTIONS OF MAIN PROCESSESThis Annex contains the descriptions of the main processes determining POP environmental behavior used in the participating models.C.1. Gas/particle partitioning EVN-BETR and UK-MODELThe gas-particle partitioning is described with the help of the Finizio Aerosol Partition coefficient K QA . It ’s dependence on the octanol-air partition coefficient K oa is depicted by the following formula:K QA = 3.5 · K oaThe fugacity capacity of the bulk air compartment can then be written as the sum of the gaseous and particle-bound chemical fraction:(1 – particles in air volume fraction) · Z air + (particles in air volume fraction) · K QA · Z airwhere Z air - 1/ (R ·T ) is the fugacity capacity in air;T - corrected environmental temperature for annual mean of 90C; R - gas constant = 8.314 Pa ·m 3/mol K; Particles in air volume fraction - 2·10-11;K oa = K ow / K aw - 51616649 for PCB 153 at the averaged ambient temperature T; Averaged ambient temperature = 9︒C (base temperature).CliMoChemcited from [Scheringer et al ., 2003]The gas/particle partitioning is calculated as follows [Finizio et al ., 1997]:238550.lg .-⎪⎪⎭⎫⎝⎛⋅=h owpartair KK K This equation is used to calculate the fraction Phi, which indicates the particle-bound fraction of the substance. Phi-values range from 0-1.lg 60.55lg 2.23K ow k K partair partair K h ⎛⎫⎪=+=⋅- ⎪⎝⎭in (m 3/g)()tsp k tspk Phi partair partair ⋅+⋅=1G-CIEMSWhen K oa is not available as input:K qa = 6 · 106/P ls ,where K qa is dimensionless particle/gas partition coefficient and P ls is liquid vapour pressure. Final partitioning is calculated with TSP and density of aerosol particles in fugacity format. Vapour pressure is temperature corrected when the temperature is different from 25 ︒C. When K oa is available as input:K qa = y · K oa / (1000),where, K qa is dimensionless particle/gas partition coefficient, y is organic matter mass fraction, and isthe density of aerosol particles.(Note: G-CIEMS model can calculate V/P partitioning from only molecular weight (for preliminary assessment purpose) or from K oa . Two output 1 and 2 is presented in Chapter 4 as G-CIEMS-1 and G-SIEMS-2).DEHM-POPThe gas-particle partitioning is calculated using the absorption model:,)(1+=φTSP K TSP K p pwhere is the fraction of compound sorbed to particles, K p is gas-particle partitioning coefficient, andTSP is the total suspended particulate matter [e.g. Falconer and Harner , 2000]. K p is calculated using theK oa approach:log ,91.11log log -+=om oa r p f K m Kwhere m r is a constant expected to have a value close to +1 for equilibrium partitioning, K oa is the octanol-air partitioning coefficient, f om is the fraction of organic matter in the particles, and 11.91 is a constant determined by the intercept br = log f om – 11.91 [Finizio et al ., 1997, Falconer and Harner , 2000]. The temperature dependent K oa is calculated using the expression:)),11(exp()()(TT R U T K T K ref oa ref oa oa -∆= where ∆U oa is the internal energy of phase transfer, R is the universal gas constant, T is the temperature and K oa (T ref ) is the value of K oa at the reference temperature T ref [Beyer et al., 2002].SimpleBoxcited from[Brandes et al., 1996]Air-aerosol partition coefficients are usually not known. However, some information is frequently available on the fraction of the chemical that occurs in association with the aerosol phase. SimpleBox uses this information for the computations. A value for the fraction of the chemical that is associated with the aerosol phase, FRass aerosol, can be entered directly, or estimated on the basis of the chemical's vapor pressure, according to Junge[1977]. In this equation, the sub-cooled liquid vapour pressure should be used. For solids, a correction is applied according to Mackay [1991]:If MELTINGPOINT < TEMPERATURE [S] (substance is liquid): θθCONST.+T URE VAPORPRESS CONST.= FRass S aerosol )(][ If MELTINGPOINT > TEMPERATURE [S] (substance is solid):θθ-CONST.+e T URE VAPORPRESS CONST.= FRass S E TEMPERATUR NTMELTINGPOI S aerosol ).(.][][).(1796with FRassa erosol[S] - fraction of the chemical in air that is associated with aerosol particles at scale S [-] (A);VAPOR PRESSURE(T) - vapor pressure of the chemical at temperature T at scale S [Pa] (A);MELTINGPOINT - melting point of the chemical [K] (A);CONST- constant [Pa ·m] (C);- surface area of aerosol phase [m aerosol 2/m air 3] (C);TEMPERATURE [S ]- temperature at the air-water interface at scale S [K] (A). with the product CONSTset equal to 10-4 Pa.CAM/POPsIn the CAM/POPs model, the process of POP partitioning between the gas and particulate phases in atmosphere is based on Junge-Pankow adsorption model [Junge , 1977; Pankow , 1987] POP fraction Φ adsorbed on the atmospheric aerosol particles is given by:Θ⋅+Θ⋅=Φc P c L 0where, Φ - fraction of POPs adsorbed on aerosol particles;Θ - aerosol surface area available for adsorption, m 2 aerosol/m 3 air;P 0L - liquid-phase saturation vapour pressure of pure compound, Pa;c - parameter that depends on the thermodynamics of the adsorption process and surfaceproperties of the aerosol (Pa · cm).Junge ’s proposed value of the parameter c is 17.2 Pa · cm [P ankow , 1987; Falconer et al., 1994;Bidleman et al., 1998].The liquid vapour pressure, P 0L , are derived from:b TmP L +=010log ,where the slope (m) and the intercept (b) are estimated to calculate liquid vapour pressure of POPs with changing air temperature [Falconer et al., 1995; Harner et al., 1996]. Temperature dependence of P0L for each congener can be seen in Table C.1.Table C.1. Liquid vapour pressure of PCBs as a function of air temperatureAerosol surface area, , is calculated by multiplying aerosol number density by its wet surface area.MSCE-POPIn the current model version (MSCE-POP 1) the characterization of POP partitioning between the gas and particulate phase of a pollutant is performed using subcooled liquid vapour pressure p ol (Pa). According to the Junge-Pankow adsorption model [Junge , 1977; Pankow , 1987] POP fraction ϕ adsorbed on the atmospheric aerosol particles equals to:θθϕ⋅+⋅=c p c olwhere c is the constant depending on thermodynamic parameters of adsorption process and onproperties of aerosol particle surface. It assumed c = 0.17 Pa ·m [Junge , 1977] for background aerosol;is the specific surface of aerosol particles, m 2/m 3. Assumed= 1.5·10-4 [Whitby , 1978].The temperature dependence of p ol (Pa) is parameterized in the model by:⎪⎭⎫ ⎝⎛--⋅=0110T T a olol P e p p ,where T 0 = 283.15 K is the reference temperature, T (K) is the ambient temperature, p 0ol is p ol value at reference temperature, and a P is the coefficient of temperature dependence. The values of p 0ol and a P for considered PCB congeners used in the model are presented in Table C.2.Table C.2. Coefficients of p ol temperature dependence for three PCB congeners used in MSCE-POP modelCongener p 0ol a P PCB-28 6.43·10–3 9383 PCB-153 9.69·10–5 10995 PCB-1801.67·10–511610At present the work on modification of the description of gas aerosol partitioning within MSCE-POP model is ongoing. The approach using the octanol-air partitioning coefficient absorption model presented in [Falconer and Harner, 2000] is tested. Under this approach POP fraction ϕ adsorbed on the atmospheric aerosol particles is calculated as:TSPK TSP K p p ⋅+⋅=ϕ1where K p is the particle-gas partitioning coefficient and TSP is the total suspended particle concentration (μg ⋅m -3). The constant K p is calculated for PCBs via K oa by the following regression equations [Falconerand Harner, 2000]:log K p = log K oa + log f om – 11.91,(experimental version - MSCE-POP 2)where K oa is the octanol/air partitioning coefficient and f om is the fraction of organic matter in the atmospheric aerosol involved in partitioning.The temperature dependence of K oa is parameterized in the model by:⎪⎭⎫ ⎝⎛--⋅=0110T T a oaoa K eK K ,where K 0oa is K oa value at reference temperature, and a K is the coefficient of temperature dependence. The values of K 0oa and a for considered PCB congeners used in the model are presented in Table C.3.Table C.3. Coefficients of K oa temperature dependence for three PCB congeners used in MSCE-POP modelEVN-BETR and UK-MODELThe intermedia transport of chemicals is described using D-values (mol/Pa ·h), which represent how fast advective and reactive/degradation processes are occurring. In the case of the air to surface exchange, the D-value defining dry particle deposition is:Dair-surface = Surface Area · Particles in air volume fraction · V q · Z air · K QAKnowing these values can help calculate the flux of a chemical entering a region and, thus, it ’s amount in the compartment under study.Surface area - compartment specific;Particles in air volume fraction - 2 10-11; Vq - dry deposition velocity = 10.8 m/h.CliMoChemcited from [Scheringer et al ., 2003]Dry deposition to baresoil a , water, vegetation-covered soil bgas gas i drygas C V A v Phi dtdC ⋅⎪⎪⎭⎫⎝⎛⎪⎪⎭⎫ ⎝⎛⋅-=a If the year consists of exactly four seasons with varying temperatures,v dry for deposition to baresoil is changingtaking into account that in the cold season the atmosphere is more stable and the deposition rate therefore is smaller. The spring and fall values are interpolations between the summer and winter values. v dry changes as follows [Waniaand McLachlan , 2001]:bIf the year consists of exactly four seasons with varying temperatures, v dry for deposition to vegetation-covered soil ischanging taking into account that in the cold season the atmosphere is more stable and the deposition rate therefore is smaller. The spring and fall values are interpolations between the summer and winter values. v dry changes as follows [Wania and McLachlan , 2001]: The vegetation cover consists of three types: Grass, Coniferous Forest and Deciduous Forest. The variables VegGrass, VegCon and VegDec describe the fraction of the vegetation-covered soil occupied by the different cover types. Their numeric value is between 0-1 and depends on the climatic zone.Dry deposition to vegetationgas gas veg gas C V A vdryPhi dtdC ⋅⎪⎪⎭⎫⎝⎛⎪⎪⎭⎫⎝⎛⋅-=Paramete rDescriptionNumeric ValueC gas concentration of substance in gaseous phasePhi particle-bound fraction of the substance (see C.1.) between 0-1v drydeposition ratevariable, depending on climatic zone*A veg Area of vegetation (identical with Area of vegetation-covered soil)variableV gas Volume of gaseous phase variable* - the model contains three types of vegetation. For each type, the deposition rate (v dry ) is different (see table below). Depending on the composition of a climatic zone, v dry is calculated as follows:vdry fractiongrass vdry fractiondec vdry fractioncon vdry i i grass i dec i con =⋅+⋅+⋅ParameterDescriptionNumeric Value Referencevdry i deposition rate in climatic zone ivdry grass deposition rate to grass55.92 m/d Horstmann and McLachlan [1998] Möller [2002]vdry dec deposition rate to deciduous forest 447.6 m/d vdry condeposition rate to coniferous forest43.2 m/d fractiongrass ifraction of grass of total vegetation in climatic zone ivariablefractiondec i fraction of deciduous forest of total vegetation in climatic zone ivariablefractioncon i fraction of coniferous forest of total vegetation in climatic zone ivariableBecause of increased stability of the atmosphere in the spring, fall and winterseason, the deposition rates vdrygrass, vdrydec and vdrycon are divided by 3 for the winterseason and by 2 for the spring and fall seasons (given that the year consists of exactly four seasons with varying temperature [Wania andMcLachlan , 2001]).G-CIEMSF = v Dep · (TSP/) · C particleWhere F is mass flux of compound for this chemical, v Dep is dry deposition velocity of particles, TSP is particulate concentration of weight/volume dimensionis density of aerosol particles, C particleis compound volumetric concentration in particles. Same value is used on all land and water surfaces.DEHM-POPThe dry deposition of particulate phase is calculated as a flux given by the atmospheric concentration times a deposition velocity [Christensen , 1997]. The deposition velocity is highly dependent on the meteorological conditions and the surface properties. The size of the particles is assumed to be 1m[Christensen , 1997].For unstable conditions in the atmosphere (when L < 0, i.e. at day time with clear sky), the deposition velocity is calculated using:),)300(1()3/2(*La u v d -+=where u* is the surface friction velocity, a is a constant depending on the surface properties, and L is the Monin-Obukhov length.For stable conditions in the atmosphere (when L > 0, i.e. at night time with clear sky), the deposition velocity is calculated using:.*au v d =The surface friction velocity is calculated using:,*)10ln(*35.0*zUu =where U is the wind speed and z is the roughness length which depends on the properties of the surface, and varies seasonally.The Monin-Obukhov length is calculated using:,)(.)(*ρ-⋅⋅⋅=p dc H g Tu L 3503 where T is the temperature, g is the gravitational constant (g = 9.82), H d is the heat flux, c p is the specific heat at constant pressure, andis the air density. L is positive (stable atmosphere) when the heat flux isnegative (night time clear sky) and negative (unstable atmosphere) when the heat flux is positive (day time clear sky).SimpleBoxcited from [Brandes et al ., 1996]Value for the deposition mass transfer coefficients DRYDEP aerosol may be obtained by means of:FRass .RATE AEROSOLDEP = DRYDEP S aerosol S S aerosol ][][][with DRYDEP aerosol [S ]: mass transfer coefficient for dry deposition of aerosol-associated chemical at scaleS [m air /s] (D);AEROSOLDEPRATE [S ] - deposition velocity of the aerosol particles at scale S with which the chemicalis associated [m/s] (A);FRassaerosol [S ]: fraction of the chemical in air that is associated with aerosol particles at scale S [-](A).Deposition mass flows of the chemical depend on the rate of dry aerosol deposition. Deposition velocities of aerosols vary greatly with the size of the particles. As chemicals may be associated with particles of a specific size, the deposition velocities depend also on the chemical. The values given are typical values, to be used as a starting point:scm = RATE AEROSOLDEP S /1.0][with AEROSOLDEPRATE [S ]: deposition velocity of the aerosol particles with which the chemical isassociated at scale S [m/s] (A)CAM/POPsRemoval of POPs particles is coupled with the dry deposition of aerosols in the CAM model. The dry deposition flux can be written as:,p d d C v F ⋅-=where V d is the deposition velocity and C p is the particle concentration [G ong et al. 2003]. Dry deposition flux of POPs gas is written as:,g d d C v F ⋅-=where the dry deposition velocity of gas, V d , is calculated in CAM model [G ong et al. 2003].MSCE-POPDry deposition to grass and bare soilAccording to model of [Sehmel , 1980], deposition flux over grass is calculated as(),*soil C soil soil p z B u A C F 02+⋅=where as above *u , is the friction velocity;z 0 is the surface roughness, mm;A soil = 0.02,B soil = 0.01,C soil = 0.33.Dry deposition to forestAccording to model of [Ruijgrok et al., 1997], deposition flux over forest is calculated ashp u u E C F 2*⋅=,where u h is the wind speed at forest height h = z b ;*u is the friction velocity, m/s;()()()20/80exp 1*-+=RH u E γαβ, α = 0.048, β = 0.3 and γ = 0.25, RH =80.Dry deposition to seawaterAccording to model of [Lindfors et al., 1991], deposition flux over seawater is calculated as:)(*sea sea p B u A C F +⋅=2,where *u is the friction velocity, m/s;A sea = 0.15,B sea = 0.013Similar to DEHM-POP model values of deposition flux depend on meteorological conditions (friction velocity u *). Therefore, in the results of calculation experiments we present the range of obtained values of the flux together with its value at average environmental parameters.C.3. Wet deposition EVN-BETR and UK-MODELIn a similar way as for the dry particle deposition, wet scavenging is defined as the result of:D air-surface = Q · Surface Area · Particles in air fraction · K QA · U R · Z airIn the case of deposition in vegetation, the percentage of rain interception due to vegetation is taken into account.U R - rain rate = 8.84 x 10-5 m/h; Q - Rain Scavenging ratio = 200000;Or Snow Scavenging Ratio = 1000000CliMoChemcited from [Scheringer et al ., 2003]Wet deposition from gaseous phase to baresoil and water()gas h gas i rain gas C K V A v Phi dtdC ⋅⎪⎪⎭⎫⎝⎛⋅⎪⎪⎭⎫ ⎝⎛⋅--=11Wet deposition from gaseous phase to vegetation-covered soil()()gas rt h gas vsoil rain gas C f K V A v Phi dtdC ⋅-⋅⎪⎪⎭⎫ ⎝⎛⋅⎪⎪⎭⎫ ⎝⎛⋅--=111* - for deciduous and coniferous forest, the f rf -value is 0.35 for the summer season, for grass, the f rf -value is 0.12 forthe summer season. In the winter season, the value is at 10% of the summer season, in spring and fall season, the value is at the linear interpolation value between summer and winter season. Because the composition of the vegetation varies with the climatic zones, the contributions of grass, coniferous and deciduous forest to the overall f rf -value of a specific climate zone differ and are proportional to the fraction of the respective vegetation type in a climatic zone.Wet deposition from gaseous phase to vegetation()gas rt h gas vsoil rain gas C f K V A v Phi dtdC ⋅⋅⎪⎪⎭⎫⎝⎛⋅⎪⎪⎭⎫ ⎝⎛⋅--=11* - for deciduous and coniferous forest, the f rf -value is 0.35 for the summer season, for grass, the f rf -value is 0.12 for the summer season. In the winter season, the value is at 10% of the summer season, in spring and fall season, the value is at the linear interpolation value between summer and winter season. Because the composition of the vegetation varies with the climatic zones, the contributions of grass, coniferous and deciduous forest to the overall f rf -value of a specific climate zone differ and are proportional to the fraction of the respective vegetation type in a climatic zone.Wet deposition from particulate phase to bare soil and watergas gas i ratio rain gas C VA scav v Phi dtdC ⋅⎪⎪⎭⎫⎝⎛⎪⎪⎭⎫ ⎝⎛⋅⋅-=Wet deposition from particulate phase to vegetation-covered soil()gas rt gas vsoil ratio rain gas C f V A scav v Phi dtdC ⋅-⋅⎪⎪⎭⎫⎝⎛⎪⎪⎭⎫ ⎝⎛⋅⋅-=1* - for deciduous and coniferous forest, the f rf -value is 0.35 for the summer season, for grass, the f rf -value is 0.12 for the summer season. In the winter season, the value is at 10% of the summer season, in spring and fall season, the value is at the linear interpolation value between summer and winter season. Because the composition of the vegetation varies with the climatic zones, the contributions of grass, coniferous and deciduous forest to the overall f rf -value of a specific climate zone differ and are proportional to the fraction of the respective vegetation type in a climatic zone.Wet deposition from particulate phase to vegetationgas rt gas vsoil ratio rain gas C f V A scav v Phi dtdC ⋅⋅⎪⎪⎭⎫⎝⎛⎪⎪⎭⎫⎝⎛⋅⋅-=*- for deciduous and coniferous forest, the f rf -value is 0.35 for the summer season, for grass, the f rf -value is 0.12 forthe summer season. In the winter season, the value is at 10% of the summer season, in spring and fall season, the value is at the linear interpolation value between summer and winter season. Because the composition of the vegetation varies with the climatic zones, the contributions of grass, coniferous and deciduous forest to the overall f rf-value of a specific climate zone differ and are proportional to the fraction of the respective vegetation type in a climatic zone.G-CIEMSF = R rain · C air / H +(TSP/) · Q · C particle ,where F is total mass flux by this process, R rain is rain rate, C air is gaseous concentration of chemical, H is Henry ’s law constant, TSP is particulate concentration,is density of particles, Q is scavenging ratio ofparticles, and C particle is volumetric chemical concentration in particles. Same Q value is assumed over all land and water surfaces.DEHM-POPThe wet deposition is given as a flux calculated as the product between the air concentration and a scavenging ratio [Christensen , 1997]. Different scavenging ratios are used for in-cloud scavenging and below-cloud scavenging. It is assumed that air pollution is scavenged more efficient in the clouds than below the clouds. The below cloud scavenging rate at a given heightis given by:,)()(wa bc H P W ρσΛ=σwhere bc is the below cloud scavenging coefficient, P a is the total precipitation at the level, H is aneffective thickness for scavenging (H = 1000 m), and w is the density of water. The in cloud scavengingrate at a given height is given by:,)()(wc H P W ρσΛ=σ where c is the below cloud scavenging coefficient, P a is the is the total precipitation created inside thecloud layer. The used scavenging ratios are:bc = 100000 andc = 700000.The total amount of pollutant scavenged by the wet deposition is then dependent on the actual height of the formation of the precipitation, and on the vertical concentration profile.SimpleBoxcited from [Brandes et al ., 1996]Value for the deposition mass transfer coefficient WASHOUT may be obtained by means of: ][][][S S S .SCAVratio RAINRATE = WASHOUTwithWASHOUT [S ] - mass transfer coefficient for wet atmospheric deposition at scale S [m air ·s -1] (D);RAINRATE [S ]- rate of wet precipitation at scale S [m rain .s -1] (A);SCAVratio [S ] - scavenging ratio (quotient of the total concentration in rainwater and the totalconcentration in air) of the chemical at scale S [-] (A).The scavenging ratio may be known from measurements or estimated:][][][][][1S S aerosol S w ater air S aerosol S COLLECTeff FRass K FRass SCAVratio +-=-with SCAVratio [S ] - scavenging ratio (quotient of the total concentration in rainwater and the totalconcentration in air) of the chemical at scale S [-] (A);FRass aerosol [S ] - fraction of the chemical in air that is associated with aerosol particles at scale S [-](A);K air-water [S ] - air-water equilibrium distribution constant at scale S [mol ·m air -3/mol ·m water -3] (A);COLLECTeff [S ] - aerosol collection efficiency at scale S [-] (A),The first term represents an estimate of the (equilibrium) distribution between rain water in air and the gas phase of air. The second term represents the scavenging of aerosol particles by rain droplets. The proportionality constant of 2⋅105 is taken from Mackay [1991].Deposition mass flows of the chemical depend on the rate of wet precipitation. Collection efficiencies of aerosols vary greatly with the size of the particles. As chemicals may be associated with particles of a specific size, the collection efficiencies depends also on the chemical. The values given are typical values, to be used as a starting point:5][10.2 = COLLECTeff SwithCOLLECTeff [S ] - aerosol collection efficiency at scale S [-] (A ).Table C4. RAINRATE of the scales* - from Wania and Mackay [1995]CAM/POPsThe precipitation scavenging of POPs particles by falling rain or snow is coupled with the wet removal of aerosols in the CAM model.The particulate wet deposition flux, F p , can be written as:p snow or rain p C h F )(ψ-=,where h is the falling distance, C p is the particulate phase POPs concentration, ψ is the scavenging rate for rain or snow [Gong et al. 1997].The gas phase POPs are assumed to be in quasi-steady equilibrium with the rain drop. The air-water equilibrium coefficient, K aw , is a dimensionless partition coefficient that can be derived from Henry ’s Law constant, H (Pa ·m 3/mol) [Seinfeld , 1986].K aw = H / (R ·T)and, H = H 0 exp (a(1/T 0 - 1/T)),where T is the temperature (K) and R is the gas constant (8.314 Pa ·m 3/mol ·K or J/mol/K), H 0 is the value at the reference temperature T 0, and a is a parameter of temperature dependence.The net wet deposition flux, F w , is then written as:G AW w C K p F ⋅-=)/(where p is the precipitation rate, usually reported in mm/h and C G is the gas phase POPs concentration.MSCE-POPWet deposition of particulate phaseThe values of concentration in precipitation are given by the formula:p a p s w C W C =,where pa C - the particle bound phase concentration in the air surface layer, ng/m 3;sw C - the suspended phase concentration in precipitation water, ng/m 3;W p =1.5 ⋅ 105 - the dimensionless washout ratio for the particulate phase.Wet deposition of gaseous phaseg a g d w C W C =,where dw C - the dissolved phase concentration in precipitation water, ng/m 3;gaC - the gaseous phase concentration in air, ng/m 3; W g = 1/K H - the dimensionless washout ratio for the gaseous phase;K H - the dimensionless Henry ’s law constant.C.4. Gaseous exchange between atmosphere and soil EVN-BETR and UK-MODELAir-soil diffusionD air-soil = (Soil Area · Z air ) / [(Z air / (MTC as · Z air + MTC sw · Z water )) + 1 / MTC sabl ]where Average soil depth = 10 cm;Soil Area = 8.36 1012 m 2;Z water = Z air / K AW = 543 mol/m 3 Pa;MTC as - soil air-phase diffusion transport velocity = 0.04 m/h; MTC sw - soil water-phase diffusion transport velocity = 1 x 10-5 m/h; MTC sabl - soil air boundary layer transport velocity = 1 m/h.Air-soil rain dissolutionD air-soil = Soil Area · U R · Z waterCliMoChemcited from [Scheringer et al ., 2003]Diffusion from atmosphere to baresoil and vegetation-covered soil()gas gas i K foc rhoom regc K h gasS gas C V A vS K vL vG v Phi dtdC ow h⋅⋅⎪⎪⎭⎪⎪⎬⎫⎪⎪⎩⎪⎪⎨⎧+++--=-⋅⋅⋅1111Analogous to the procedure in subsection C.3, the parameter v gasS is changed if the year consists of exactly four seasons with varying temperatures (compare tables below).baresoil:vegetation-covered soil:The vegetation cover consists of three types: Grass, Coniferous Forest and Deciduous Forest. The variables VegGrass, VegCon and VegDec describe the fraction of the vegetation-covered soil occupied by the different cover types. Their numeric value is between 0-1 and depend on the climatic zone. Calculation of vG, vL, vS [Jury et al., 1983].When calculating the v i -value, the corresponding Di -value must be used (eg. for calculating vG, the DG-value is used):()⎥⎦⎤⎢⎣⎡⋅++--⋅⋅⋅⋅=h airsoil w soil airsoil w soil ow OC h h i i K frac frac frac frac K f rhoom regc K D v soil 12。

A Partitioning Framework for Aggressive Data SkippingLiwen Sun,Sanjay Krishnan,Reynold S.Xin and Michael J.FranklinAMPLab,UC Berkeley{liwen,sanjay,rxin,franklin}@ABSTRACTWe propose to demonstrate afine-grained partitioning frame-work that reorganizes the data tuples into small blocks at data loading time.The goal is to enable queries to maxi-mally skip scanning data blocks.The partition framework consists of four steps:(1)workload analysis,which extracts features from a query workload,(2)augmentation,which augments each data tuple with a feature vector,(3)reduce, which succinctly represents a set of data tuples using a set of feature vectors,and(4)partitioning,which performs a clus-tering algorithm to partition the feature vectors and uses the clustering result to guide the actual data partitioning. Our experiments show that our techniques result in a3-7x query response time improvement over traditional range partitioning due to more e↵ective data skipping.1.INTRODUCTIONA rapidly increasing number of applications require inter-active data analysis on enormous datasets.This necessitates the capability of low-latency query processing for large-scale query engines.Among others,an e↵ective way to improve query latency is to reduce the unnecessary data access.For example,column store techniques have been widely adopted to avoid touching irrelevant columns.To reduce the scan of unwanted tuples,there is an increasing interest in data skip-ping in recent systems[4,5,2].By partitioning the data into small blocks,these systems associate each block with some metadata,such as the min and max values of each column.A query canfirst evaluate itsfilter against these block-level metadata and decide which blocks can be safely skipped. Data skipping can be viewed as a generalization of par-tition pruning.On a horizontally partitioned table,parti-tion pruning allows a query to prune partitions based on their associated partition key ranges.Data skipping ex-tends the idea of partition pruning in several ways.First, storing the min and max values of each column enables skip-ping based on non-partition-key columns,especially on the columns that are naturally clustered with the partition keys. This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivs3.0Unported License.To view a copy of this li-cense,visit /licenses/by-nc-nd/3.0/.Obtain per-mission prior to any use beyond those covered by the license.Contact copyright holder by emailing info@.Articles from this volume were invited to present their results at the40th International Conference on Very Large Data Bases,September1st-5th2014,Hangzhou,China. Proceedings of the VLDB Endowment,V ol.7,No.13Copyright2014VLDB Endowment2150-8097/14/08.Second,for better skipping chances,data skipping is often considered on very small blocks,e.g.,partitions consisting of 1,000’s to10,000’s of tuples[4,2]or simply HDFS blocks[5]. The e↵ectiveness of data skipping depends on the inter-play between queryfilters and the partitioning scheme.In a data warehouse environment,for example,a time-range partitioning scheme o↵ers great opportunities for partition pruning,as most queries are observed to have time-rangefil-ters.While range partitioning is simple and useful for data warehouse operations,it may not be ideal for generating a large number of small blocks for e↵ective data skipping. Specifically,range partitioning lacks of a principled way of: (1)setting thefine-grained value ranges on each column that matches the data skew and workload skew,(2)allocating the number of partitions for di↵erent partitioning columns and (3)capturing inter-column data correlation andfilter corre-lation.We propose to demonstrate afine-grained partitioning framework that,at data loading time,partitions data tu-ples into small,balanced blocks with a goal of maximizing data skipping.We call this framework WARP,based on its a four-step workflow:W orkload analysis,A ugmentation, R educe,and P artitioning.Wefirst analyze a query log o✏ine and extract a set of representative queryfilters as features.Intuitively,we want a small set of features that can subsume as many queries in the workload as possible. Given these features,a set of tuples can be succinctly repre-sented as a(much smaller)set of feature vectors.We then partition these feature vectors by solving an optimization problem.Finally,the partitioning scheme of these feature vectors will be used to guide the partition of actual data tuples.After partitioning the tuples into data blocks,we store the features and some concise metadata for each block in the system catalog.When a query comes,wefirst check if this query can be subsumed by any of the features and then decide which blocks can be skipped.Instead of specifying the partitioning columns and value ranges as in range partitioning,WARP factors in the com-mon interests of the workload as features and,based on these common interests,finds a partitioning scheme by solving an optimization problem.Thefine-grained tuple-level parti-tioning decision output by WARP o↵ers greaterflexibility and better chances for data skipping.Since WARP pro-duces very small blocks,it can be used to further segment traditional range partitions.In fact,as data is often batch inserted in a data warehouse environment,it is a good prac-tice to apply WARP within each individual time-range(e.g., date)partition,instead of moving tuples across di↵erentoutputinput(1)Workload analysisworkload(vector, count)pairs(4)Partitioningfeaturestuples(2)Augmentation(3)Reduce(vector, tuple)pairspartitioned tuplesFigure 1:The W ARP Workflowor a table partition such as a date partition,and a workload represented as a collection of queries.The query workload can be obtained from the query logs.We now walk through the four steps of the workflow,as depicted in Figure 1.2.1Workload AnalysisWorkload analysis is an o ✏ine process that extracts a set of features from the query workload.Its core is a frequent itemset mining problem.We model as an item each predi-cate or each disjunction of predicates and model each query as a set of (conjunctive)items.We run a frequent itemset mining algorithm to find frequent predicate sets .Note that our goal here is to identify a set of features that subsume as many queries as possible.Therefore,in the mining process,we count the number of queries a predicate set subsumes,instead of the number of queries in which a predicate set occurs as in a traditional sense.Some predicate sets may be redundant due to the subsumption relations.For instance,revenue <0may be redundant if revenue <10is also in the result.Thus,we finally select a set of features by removing redundant frequent predicate sets.As shown in Example 1,the output of workload analysis is a set of features,each of which is associated with an integer indicating how many queries it subsumes.Example 1.Features output by workload analyzer:F 1:event =’buy’,60F 2:product =’jeans’,20F 3:publisher =’google’^revenue <0,10example partitioning scheme over the vectors.Note that the partitioning of the vectors governs the partitioning of the tuples.For example,all the tuples with vector (0,1,0)as key will be routed to partition P 1.For each partition of vectors,we derive a union vector ,which is the union of all vectors in it.For example,the union vector of partition P 1is (1,1,0).Since the third bit of this vector is 0,we know that no vector in P 1has the third bit on,and hence,no tuple in P 1satisfies feature F 3.In this case,all the queries subsumed by F 3can safely skip scanning P 1.Intuitively,our objective is to find a partitioning that maximizes the (weighted)sum of zeros in the union vec-tors.This objective is quite di ↵erent from traditional clus-tering objectives such as k-means and distance-based met-rics.We proved that finding an optimal partitioning under this objective is NP-hard [3].We adopt the classic bottom-up clustering framework as a heuristic:every vector starts as a partition by itself,and at each iteration,we merge the two vectors that hurtsa the objective the least.To make the partitions almost balanced,we remove a partition from further merging once its size reaches a parameter minSize .By consulting the partitioning result of vectors,each of the (vector,tuple)-pairs (from the Augmentation step)is routed to its destination partition.The WARP workflow can be executed at data loading time and may be re-executed later to account for workload changes.In the event that the data arrival rate is high or that the new data needs to be queried immediately,WARP can be postponed.System Catalog SELECT&publisher,&sum(revenue)FROM&events&WHERE&product&=&'jeans'&and&event&=&'buy' GROUP&BY&publisher F1:&event&=&'buy'F2:&product&=&'jeans'F3:&publisher&=&'google',&revenue&<&0&&&&&&&&&&&&P1:&(1,&1,&0)&&&&&&&&&&&&&P2:&(0,&1,&1)&&&&&&&&&&&&&P3:&(1,&0,&0)query&vector:&(0,0,1) blocks&to&scan:&P1featuresunion vectors in Query Processingneed a two-step skip checking mechanism,as illustrated in Figure3:(1)Feature matching.Given the features we store in the system catalog.When a query comes,wefirst check which features subsume this query.We then encode this information in a query vector.The i-th bit of the query vector is0if this query can be subsumed by the i-th feature. For example,in Figure3,wefind that feature F1and F2 can subsume the query,but not F3,and thus we constructa query vector(0,0,1).(2)Union vector checking.Given that we store a union vector for each block,we now check the query vector against these union vectors and decide which blocks can be skipped.Specifically,we perform a bitwise OR operation between the query vector and each union vector.For any block,if the result of the OR operation has at least one0 bit in it,then this block can be safely skipped.In Figure3, the OR ed vectors for P1,P2and P3are(1,1,1),(0,1,1)and (1,0,1)respectively.Thus,we know that we only need to scan P1.Finally,this information is passed to the table scan operator.The above procedure happens before the actual query ex-ecution.As we can see,we only need to maintain a set of features used and one bit vector for each block.This meta-data is very small and can be stored in main memory.The skip checking only involves a feature matching step and,for each block,a bit-OR operation.This incurs little overhead to query latency.Note that this skipping mechanism can be used in conjunction with existing skipping mechanisms based on per-column value ranges.4.SYSTEM PROTOTYPEWe prototype our blocking techniques on Shark[5],a fully Apache Hive-compatible data warehousing system us-ing Apache Spark[6]as runtime.Shark Background.Shark uses Hive’s query parser to parse and compiles HiveQL(SQL-like)queries to a query plan,and then translates it to Spark tasks.A Shark table is stored as a Spark data abstraction called Resilient Dis-tributed Dataset(RDD),which is physically stored as a list of data blocks,each of which can be either memory-or disk-resident.Each Spark block is a task processing unit and has a default size of128MB.Shark can skip some of these blocks during a table scan.At data import time,Shark col-lects the data statistics for each block,such as the min and max values of each column.These block-level statistics are maintained in the system catalog.The table scan operator first fetches the queryfilter operators and applies them on these statistics to prune blocks.Only the blocks that are not pruned in this step are actually scanned.We now briefly discuss how our techniques were imple-mented on Shark.First,we collect a query trace from the query logging system of Shark or from an external source. We use Shark’s query parser(through Hive)to convert each query string into a set of conjunctivefilter operators.We implemented a isSubsume(f1,f2)function using a set of rules to check iffilter f1subsumesfilter f2.A workload analysis module was added in Shark to compute frequent itemsets and remove redundant results based on subsump-tion,as discussed in Section2.Given the features,represented asfilter operators,we sim-ply wrote a Spark map function to transform each data tuple into a(vector,tuple)key-value pairs.Then we use a Spark reduce function to group-by these key value pairs.We im-plemented a bottom-up clustering algorithm as an indepen-dent module in Shark.Note that external libraries could be used here for an optimized implementation.We then built a customized partitioner class which implements the Par-titioner interface in Spark.This partitioner stores a(vec-tor,partition-id)map in memory and routes each(vector, tuple)-pair to its corresponding destination block.We then re-partition original table(an RDD)using this partitioner. In the end,we added the metadata described in Section3 to the Shark system catalog.A table scan in Shark with WARP can utilize the conjunction of two block skipping mechanisms:our feature-based skipping(Section3)and the existing skipping based on value ranges.5.PERFORMANCEWe conducted extensive evaluation using TPC-H and a real-world analytical workload.The experiments were run on an Amazon Spark EC2cluster of25m2.4xlarge instances, each with82.66GHz CPU cores,68.4GB of RAM and800 GB of disk.Here we present some performance highlight from TPC-H.For full experimental results,refer to[3]. Dataset.We denormalize the TPC-H tables against the lineitem table.With a scale factor of100,the resulting table has roughly600million rows and is700GB in size.We select eight TPC-H query templates(q3,q5,q6,q8,q10,q12,q14, q19)that involve the lineitem table and have selectivefilters. The FROM clauses in these templates were all changed to be the denormalized ing the TPC-H query generator, we generate800queries as the training workload,100from each template.We then independently generate80queries for testing,10from each template.We compare WARP against Shark’s existing data skip-ping on top of range partitioning for running the80test queries.We manually devise a composite range partitioning scheme on{o orderdate,r name,c mktsegmt,quantity}by identifying the frequently queried columns from the training0.03%!0.20%! 1.80%Selectivity !!Shark !!% tuples !Figure 4:Number of Tuples Scanned2.9!9.6!33!!Shark !!Shark !!time (sec)!Figure 5:End-to-End Query Response Time queries.For WARP,we first partition the data by month on o orderdate and apply WARP on each month partition.We set the number of features as 15and the partition size to be 50k tuples.We used both our feature-based skipping and Shark’s existing skipping for WARP.Figure 4shows the number of tuples scanned by these two approaches.Shark’s data skipping mechanism on range par-titioning is decent,as it only scans 1.8%percent of the table for an average query.WARP brings down this number to be 0.2%.As a reference,we also plot the average selectivity of these queries,which is the minimum number of tuples that have to be scanned for answering these queries.We compare the end-to-end query performance in Fig-ure 5.Note that the table is memory resident.Without any skipping,Shark scans the whole table for every query,which takes 30seconds on average.By switching on the skipping,the average query response time in Shark becomes 9.6seconds.With WARP,it only needs 2.9seconds,a 3⇥improvement.To summarize,by deploying WARP on Shark,we signifi-cantly reduce the number of tuples scanned and this reduc-tion e ↵ectively translates to a significant end-to-end query response time improvement.6.DEMONSTRATION DETAILSWe will demonstrate our WARP prototype on Spark de-ployed on a Amazon EC2cluster [1].We will use the denor-malized TPC-H dataset.6.1Demonstration ScenarioOur demonstration will put the conference attendees in the position of a database administrator (DBA)or a perfor-mance engineer.She would like to consider working with WARP to analyze her workload characteristics,load the data,and observe the query performance improvement over existing partitioning methods.We will walk through the following three steps:Workload Analysis.To make the best use of WARP,we illustrate and visualize the key characteristics of the work-load.The questions we aim to address in this part of demon-stration are:(1)how predictable are future queries from a past-workload analysis?(2)how many features do we need?and (3)how representative are these features?At a web console,the attendees can tune some parameters,such as the number of features,and interactively observe the actualSkipping in Action.At this stage,we have several bers,this part of demonstration helps attendees understand why some queries benefit more from the WARP partitioning than others.For comparison,we also print out block-level min and max values and show how they help queries skip blocks in existing systems.6.2TakeawayThis demonstration illustrates the reason behind WARP’s significant performance benefits and its ease-of-use features,e.g.,only two parameters to ing WARP’s buit-in workload analysis helps users understand their workload characteristics before setting up WARP.To deploy WARP on a query engine requires little e ↵ort,as it needs minimal modification on the table scan operator and is independent of the other parts of query execution.For instance,we mod-ified Shark’s table scan operator for WARP using less than 100lines of Scala code.7.REFERENCES[1]Running Spark on Amazon EC2.https:///docs/0.9.0/ec2-scripts.html.[2]A.Hall,O.Bachmann,R.B¨u ssow,S.G˘a nceanu,andM.Nunkesser.Processing a trillion cells per mouse click.PVLDB ,5(11):1436–1446,2012.[3]L.Sun,M.J.Franklin,S.Krishnan,and R.S.Xin.Fine-grained partitioning for aggressive data skipping.In SIGMOD Conference ,pages 1115–1126,2014.[4]V.Raman et al.DB2with BLU acceleration:So muchmore than just a column store.PVLDB ,6(11):1080–1091,2013.[5]R.S.Xin,J.Rosen,M.Zaharia,M.J.Franklin,S.Shenker,and I.Stoica.Shark:SQL and Rich Analytics at Scale.In SIGMOD ,pages 13–24,2013.[6]M.Zaharia,M.Chowdhury,T.Das,A.Dave,J.Ma,M.McCauley,M.J.Franklin,S.Shenker,and I.Stoica.Resilient distributed datasets:a fault-tolerantabstraction for in-memory cluster computing.In NSDI ,pages 2–2,2012.。