下载文档原格式
张培基英译现代散文选11. Phrasal Expression & Word…amid(st) thunderous applausea bare subsistence勉强糊口Tom earn a bare subsistence wage.汤姆挣得工资勉强糊口.A be characteristic of B=B be characterize d by AA bend in a river / mountainA blind alleyA brass drum 小铜鼓A bygoneage过去的时代a clot of bloodA mass confused of…庞杂的A faint scent ofA forgon e conclusion 预料中的必然结局A hired hand on contractA jumbleof…一大堆A keen sense ofA long-timer of BeijingA loose community of smaller familyA man of profound learningA mere drop in the oceanA niche in the templeof fameA passing glanceA philosophical approach to lifea positive outcom eA scene of poeticcharmA sensation of blissfulnessA speck of mudA stone’s throwA trace / shade / tint/ sprinkle of…A Treasury of Best Chines e Prose 古文观止A vast tract of landA virtuous man / a man of suprem e virtue/ moral integrity A widening expans e of waterabandon …to fateAche/agoniz e with painAdjoining roomadmiresb for sthaffected 做作aim highamiabl e by natureAmuse onesel f by…= do …for funAmusing episod ean enlightened kingAn odd-jobberAn opportune moment合适时机Ancestral homeApproach senilityaptly 恰如其分地Art troupe文工团At a stretch/sittingAt one’s comman dat the present momentAvaricious desires 贪念Bark up the wrong tree 攻击错了目标bashfulBask in the sunshineBe ablaze/aglow with lightbe advanced in yearsbe an encumbrance to…Be beset/troubled with/byBe blurre d by…Be boggeddown = be trappe dBe bound up with…密切关联be central/indispensabl e to sb.Be coopedup = be cagedbe deeplygrieve d to learn of…be distinguishe d byBe engrossed in…Be exquisite and nicelyarrangedbe forever cherished / treasuredBe hale and heartyBe havene d frombe humanly impossibleBe imprinted/carved/engraved/ingrained on/upon…be in a fixBe instrumentalinBe interwoven with 交织着be irrelevant / foreign toBe keenlyaware ofbe keyed up紧张Be led by the nosebe of southern breedBe off and onbe on an equal footing with…Be on the lips of…Be on the minds of..be out to do…be overgrown with wild woodsbe packedwith…Be plague d = annoye d = upsetBe possessedBe possessed ofbe reconciled to…be reduce d tobe saddle d withbe sent to gallowsBe short of/devoidofBe shrouded in = be covere d in= be enveloped inBe sloppyin thinkingBe strewn/covered/festooned with…Be stumpe d by = baffle dBe tantamount to…= equalBe tinted/ colore d by…Be troubled / seizedwith…Be tuckedaway in…Be weaned断奶Be weighe d down/ troubled withbe wet with perspiration/…Be wide of / far from the mark 离谱Be wild with excitement / joyBe worthyof…无愧于Beam = a big smile on faceBear a thin coating of…begin by degrees 逐渐开始Beguil e = while / idle / fritter away BibliomaniabickerBirds of a feather flock togetherBlackout: (战时)灯火管制Blurt 脱口而出bookish / pedantic / impractical view Bordersth on the westboudoir 闺房Bountiful free giftsBow down toBrazenly claim / credit厚颜无耻邀功Break into uncontrolledsobsBrilliant talentburst with vitalityBury the hatchetButtonup clothesBy dint of 凭借CaponCarcass: slaughtered animalfor food Cavernous mouthChant ancient Chines e books 诵读古籍chicken-and-egg 因果难断的Chit chatClick away the secondsCome dimly into sightCome out exceedingly wellCome to pass 出现，发生Come up againstCome upon a windfallCome/be of ageConfirmed = habitualConfirmed = habitualCongenial disposition 天性convulsive sobcool one’s heelCordiallycotton-paddedgown 棉衣Court ladies= palacemaidCrackling: sound made by burning wood Crane one’s necksCredit A with B = attribute B to ACrumbl e into dustCurl upcurtly= rudelyCut down on = economize ondecadent 颓废的Decorated archway 牌楼degenerationof publicmoralityDeliberatelymake a mystery of…departone’s lifeDevelop a likingfor…diehar d 顽固分子Dignitarydingy = dark + dirtyDiscontinuation of heartbeatdisheveled hairDispeldarknessdissipate one’s fatigu eDizzy = giddyDo … a / an + adj + justic eDo one’s bitDo sb a good turnDo.. by instinctdog-tireddoublesure = fully convincedDrab = dull, boring, monotonousDrag on : (neg.) lastDrain one’s teardrops / bloodDraw an analogy between A and BDrawn-out = prolongeddream = longing = aspiration 憧憬Dwell ondwindl e awayEat into…Eat with gusto 津津有味地吃edge away 慢慢离开eke out a living勉强度日Elegia c Address to My NephewShi’erlang祭祀十二郎文Emaciated = fragil eEmbellishment 点缀Empty into 注入，流入Enormously magnificent / generousEntertain ambitionEvery hook and cranny每个角落Exemplary 模范的Expanses of vacancyExtol 称赞Fall through = bubble= be disillusione d Falteringly 结巴地Familyrules of good behaviorFare likewiseFascinating=imposing=peerless=breath-takingfawn on / flatter / toady toFeeblyFickleman 薄情男子Fierce-browedFighting sceneFirst-hand experienceFlagstoneflowercultivationFollowone’s bent/inclination/willfor one’s part 就某人而言Forsaken rosesFrame/state of mindFrolic嬉闹，玩耍Frostbittenfruitsof laborGain accesstoGentlegracefulnessGesticulate = gestur eGet up to mischiefGet wind of…GloomysternnessgloomysternnessGnaw at sb = afflict / tortur e sbgo out of one’s way 特地Go out to…(be emotionally drawn to)Go soldiering 军旅生涯go through the wringer 历尽幸苦God of LongevityGrove 树林，果园Grow by/in leaps and boundsgrow in luxurianceGrow in rich abundanceGuileless 老实的，不奸诈Hang on to one’s presence 不肯离开hardier speciesHarp onHarp on 唠叨have …to one’s name= be in one’s possessionHave attachment for…Have blind faith in…Have designs on…Have no claim to…对…无拥有权Have nothing to live onhawk one’s waresHeart-rending = grievousHeavenly abodeHerculean strengthHoary-headed头发花白Hoodwink = cajolehousehold choreshumor editors 应付编辑I find them all. 尽收眼底Impassioned speech/ essayIn all earnest 再三，尽力，竭力In anticipationofIn defens e of…= in one’s defens e in every way 彻头彻尾in line with 符合In spite of onesel f 不由自主In the boom / prime of youthIn the morning hazeIn the prime of lifeIn unisonInadvertently 无意间indomitable = steadfastingratiate sb with…= toady to sbingratiate 讨好Inkling = hintIn-lawsInspiration gushes/comes gushing to one’s mind. Intermittently = continually InwardlyIrreparable lossJeer at 奚落jeering voiceJujubetree 枣树Jump down one’s throat猛地回击keep a wary eyeKeep early hoursKeep one’s body and soul together Keep up appearancesKitchen-maidKnack of farming 劳作技能LabyrinthLack of propriety = impropriety Lackluster 单调的Lamentwith a deep sighLangui d = lethargic = sluggish Laud sb to the skylavishattention onlazybonesLie wasteLie/remainunknownLiken / compar e sb to …LimpidbrookListless 无精打采Literati 文人学士Locusttree 槐树Longing note 杜鹃啼血look askanc e at 斜视Look over one’s shoulderloving/fond memoriesMake a clean break with…Make a fanfar eMake for…有助于Make light of 无视malicious squintmalingerman of lettersmandarin jacketmean everythingMeritorious servic emist-shroudedModel …after…Money-shopmoral excellenceMore dead than aliveMorning glory 牵牛花Moult one’s shellMyriads of fireflies荧光千点Mysterious lorenameless fear/atrocity/loneliness Negotiate gorgesNext to noneNot sleep a winknuptial chamber 洞房Obsequious = flatteringOctogenarian八十岁的人Old homeOmenon sentryduty 站哨On the pretext of…以…为借口/理由OstrichlikeOverflow with material desirepack the hallPalpable = noticeable = tangible Parting sorrowPass for = be regarded asPass out of existencepeaceful calmnesspendant 坠饰Peopleof all strataPerennially youngPersonal inclinationPersonal liking= preferenceperspire / blood profuselyPerusalPet phrasepetticoat influence = nepotismpigstyPit – a –pat = pitter–patteringPitch-darknessPlay up to = fawn on bigwigspleasure of lifePly sb with sth 不断供应Pounceon…猛地扑向Prostrate onesel f 顺重Prowl 潜行Puddles of rainwater雨水坑Put sb in the mind of…Quite a few varieties of…Reddening glow of setting sunReedRefrain from…Reign suprem e 占优势Render a servic e to…respect = reverent = well-mannered Reverberating bellRhapsody on E’pang Palace阿房宫赋rosary念珠Royal edictRubble碎石Rules and regulationssafe haven = soft / peaceful shelter Sallow蜡黄，灰黄Sapphire mountains and emeral d riverssave up forsb grow out of …长大Scarlet cloudschemeagainst each otherScrapealong/by艰难度日Scurry/hastenhomewardssedan-chairSee the light of day 出版问世Seek a livingseek solaceSentimental value 纪念价值Servant girlSet one’s mind on sth/ to do sthShadowboxingshed tears over…shoulder pole 扁担Shout / yell one’s head offShrug off one’s words = distrustShy away from…= avoid…Sickento seesidelong glancesideway growthSign with regretSit bolt uprightSize sb upSleep on/oversmart alecksnatchawaySocialstrataSodden浸湿的Sound traile d off.Sparkling starSpeak volubly 口若悬河Spell disasterSpell disasterspin cottoninto yarn 纺纱Springdrizzl eSpur onesel f onStand to benefit in …可能/将会有利于Starry= star-studde d= a constellation of performers and artistsStereotyped essaySth be ingrained 根深蒂固sth be written all over one’s faceStrengthen/toughen/brace up one’s spirits strideover / brush / whiz / flit pastStrokethe scar / chinSuperbattractionsupercilious高傲的supplicate 恳求Suprem e duty 最大责任swagger 昂首阔步走sweat streams down face.Tadpol eTake …at face value 只看表面take …very much to heartTake sb to task= severely reproach/reprimand sbtake to sth/doingtaste of = look likeTattered clothesTear itselfaway fromTemple太阳穴，鬓角That’s final.the askingprice 要价The hooting of an owlThe long night wore onamid(st) its dripping sound.The sea of mortals = the livingThere’s no limit to learning.Thinning hairTime and money he had noneTo complicate matters= to worsenthe situationTo one’s heart’s contentto the effectthat…大意是To the neglect of meals and sleepTo the tune of…touch off = sparkTower to the skyTram 有轨电车Treat sb/sth to…让某人享受…Tricolor 法国三色国旗Tumultuous = turbulentTurbid, sordid, corpulentTwig 树枝twine and climb 盘旋而上Twinkl e / sparkl e with…Twist and turnUmbilical cord 脐带unbosom onesel fUnder one’s wing / protectionUndiminished= remainstrong= not receding UnflinchinglyUnkempt = untidyunruffled = calm = placidunsettled lifeUnshirkable responsibilityUsurp the throneUtter/let out a soundVenerable Guovenomous = maliciousverand aVermin= harmful people/ objectVie in doing sthwane = pine away = wear awayWaysid e flower路边野花Waywardness 任性Wear an air of casualindifferenceWear and tear of timeWeeping willowWell upWheat in the ear 芒种whetstone 磨刀石Whimper = sobWhims and desires 奇思幻想white poplar白杨Wide erudition 博学Wind up staying…With no strings attachedWitheraway in solitude 在孤寂中凋零Without a soul in sightwryly = ironically2. SentenceA feeling of forlornness will begin to creep up on you.A glimmer of light filters into …A keen / biting/ stinging / cutting / savagesatireA life free from worries and caresA matterof survival or extinctionA soldier knows no compromise.As fresh as ever in my memoryAs smoothas a mirror道滑如拭Be besideonesel f / overwhelmed with joy / rageBe entranced / attracted / enthralled / entice dBe fraught with (dreams/concerns)Be more than eager to do…Be reserved and content to live in obscurityBe worthyof remembranceBring…into closercontact with…Conveyone’s regards to…Cries gradually recedefrom hearing.discredit / disgrace sb in publicDrink nonstop to…= propos e repeated toastsEndurecountless hardshipsGraduate with honorsHarbor/have an enmityagainst…Hardlybe able to escapecensur eHave sth to recommend itself总是好的He never made enoughmoney.His debts mounte d with each passing year.how many days I’ve got = I’m entitled toI feel duty-bound to…= I feel it incumbent on me to…It filledme with much concern to learn of your indisposition. Lapse into sympathetic silenc eMake a still betterplace of…Merge into a harmonious wholemuffle dMy heart was tramping / racing/ pounding.My heart’s throbbing with gratitude.nocturnal merry-makingunder candle-lightsRegister/recordthe days of youthreliveold days/timesSee through the vanityof human societyThe choiceis yours / lies in you.There reignspeace and quietness.There’s no denying the fact that…There’s nothing but stillness there.A strip of water shouldhave becomeso vast a distance=keep us poles apartBe representative/typical/symbolic/emblematic/quintessential of…Climb (up) the socialladder= rise to power and positionFulfill the task impose d on you and your fatherby historyIf I were the sea-tide, I’d marshal rolling waves to cleans e the beach of all accumulated filth.Lots of thingsare apt to fade away as one’s life experience accumulatesPlunge/throw/thrust/thrashthe house into deep darknessthe greatest / best …that ever breathed since time immemorialWith/have a slighttouch/portion/shade/trace/tint of regret微叹。
The Dell EMC S5148 switch is an innovative, future-ready T op-of-Rack (T oR) open networking switch providing excellent capabilities and cost-effectiveness for the enterprise, mid-market, Tier2 cloud and NFV service providers with demanding compute and storage traffic environments.The S5148F-ON 25GbE switch is Dell EMC’s latest disaggregated hardware and software data center networking solution that provides state-of-the-art data plane programmability, backward compatible 25GbE server port connections, 100GbE uplinks, storage optimized architecture, and a broad range of functionality to meet the growing demands of today’s data center environment now and in the future.The compact S5148F-ON model design provides industry-leading density with up to 72 ports of 25GbE or up to 48 ports of 25GbE and 6 ports of 100GbE in a 1RU form factor.Using industry-leading hardware and a choice of Dell EMC’s OS10 or select 3rd party network operating systems and tools, the S5148F-ON Series offers flexibility by provision of configuration profiles and delivers non-blocking performance for workloads sensitive to packet loss. The compact S5148F-ON model provides multi rate speedenabling denser footprints and simplifying migration to 25GbE server connections and 100GbE fabrics.Data plane programmability allows the S5148F-ON to meet thedemands of the converged software defined data center by offering support for any future or emerging protocols, including hardware-based VXLAN (Layer 2 and Layer 3 gateway) support. Priority-based flow control (PFC), data center bridge exchange (DCBX) and enhanced transmission selection (ETS) make the S5148F-ON an excellent choice for DCB environments.The Dell EMC S5148F-ON model supports the open source Open Network Install Environment (ONIE) for zero touch installation of alternate network operating systems.Maximum performance and functionalityThe Dell EMC Networking S-Series S5148F-ON is a high-performance, multi-function, 10/25/40/50/100 GbE T oR switch purpose-built for applications in high-performance data center, cloud and computing environments.In addition, the S5148F-ON incorporates multiple architectural features that optimize data center network flexibility, efficiency, and availability, including IO panel to PSU airflow or PSU to IO panel airflow for hot/Key applications •Organizations looking to enter the software-defined data center era with a choice of networking technologies designed to deliver the flexibility they need• Use cases that require customization to any packet processing steps or supporting new protocols• Native high-density 25 GbE T oR server access in high- performance data center environments• 25 GbE backward compatible to 10G and 1G for future proofing and data center server migration to faster uplink speeds. • Capability to support mixed 25G and 10G servers on front panel ports without any limitations• iSCSI storage deployment including DCB converged lossless transactions• Suitable as a T oR or Leaf switch in 100G Active Fabric implementations• As a high speed VXLAN L2/L3 gateway that connects the hypervisor-based overlay networks with non-virtualized • infrastructure•Emerging applications requiring hardware support for new protocolsKey features •1RU high-density 25/10/1 GbE T oR switch with up to forty eight ports of native 25 GbE (SFP28) ports supporting 25 GbE without breakout cables• Multi-rate 100GbE ports support 10/25/40/50 GbE• 3.6 Tbps (full-duplex) non-blocking, cut-through switching fabric delivers line-rate performance under full load**• Programmable packet modification and forwarding • Programmable packet mirroring and multi-pathing • Converged network support for DCB and ECN capability • IO panel to PSU airflow or PSU to IO panel airflow • Redundant, hot-swappable power supplies and fans • IEEE 1588v2 PTP hardware supportDELL EMC NETWORKING S5148F-ON SERIES SWITCHProgrammable high-performance open networking top-of-rack switch with native 25Gserver ports and 100G network fabric connectivity• FCoE transit (FIP Snooping)• Full data center bridging (DCB) support for lossless iSCSI SANs, RoCE and converged network.• Redundant, hot-swappable power supplies and fans• I/O panel to PSU airflow or PSU to I/O panel airflow(reversable airflow)• VRF-lite enables sharing of networking infrastructure and provides L3 traffic isolation across tenants• 16, 28, 40, 52, 64 10GbE ports availableKey features with Dell EMC Networking OS10• Consistent DevOps framework across compute, storage and networking elements• Standard networking features, interfaces and scripting functions for legacy network operations integration• Standards-based switching hardware abstraction via Switch Abstraction Interface (SAI)• Pervasive, unrestricted developer environment via Control Plane Services (CPS)• Open and programmatic management interface via Common Management Services (CMS)• OS10 Premium Edition software enables Dell EMC layer 2 and 3 switching and routing protocols with integrated IP Services,Quality of Service, Manageability and Automation features• Platform agnostic via standard hardware abstraction layer (OCP-SAI)• Unmodified Linux kernel and unmodified Linux distribution• OS10 Open Edition software decoupled from L2/L3 protocol stack and services• Leverage common open source tools and best-practices (data models, commit rollbacks)• Increase VM Mobility region by stretching L2 VLAN within or across two DCs with unique VLT capabilities• Scalable L2 and L3 Ethernet Switching with QoS, ACL and a full complement of standards based IPv4 and IPv6 features including OSPF, BGP and PBR• Enhanced mirroring capabilities including local mirroring, Remote Port Mirroring (RPM), and Encapsulated Remote Port Mirroring(ERPM).• Converged network support for DCB, with priority flow control (802.1Qbb), ETS (802.1Qaz), DCBx and iSCSI TLV• Rogue NIC control provides hardware-based protection from NICS sending out excessive pause frames48 line-rate 25 Gigabit Ethernet SFP28 ports6 line-rate 100 Gigabit Ethernet QSFP28 ports1 RJ45 console/management port with RS232signaling1 Micro-USB type B optional console port1 10/100/1000 Base-T Ethernet port used asmanagement port1 USB type A port for the external mass storage Size: 1 RU, 1.72 h x 17.1 w x 18.1” d (4.4 h x 43.4 w x46 cm d)Weight: 22lbs (9.97kg)ISO 7779 A-weighted sound pressure level: 59.6 dBA at 73.4°F (23°C)Power supply: 100–240 VAC 50/60 HzMax. thermal output: 1956 BTU/hMax. current draw per system:5.73A/4.8A at 100/120V AC2.87A/2.4A at 200/240V ACMax. power consumption: 516 Watts (AC)T yp. power consumption: 421 Watts (AC) with all optics loadedMax. operating specifications:Operating temperature: 32° to 113°F (0° to 45°C) Operating humidity: 5 to 90% (RH), non-condensingFresh Air Compliant to 45CMax. non-operating specifications:Storage temperature: –40° to 158°F (–40° to70°C)Storage humidity: 5 to 95% (RH), non-condensingRedundancyHot swappable redundant power suppliesHot swappable redundant fansPerformanceSwitch fabric capacity: 3.6TbpsPacket buffer memory: 16MBCPU memory: 16GBMAC addresses: Up to 512KARP table: Up to 256KIPv4 routes: Up to 128KIPv6 routes: Up to 64KMulticast hosts: Up to 64KLink aggregation: Unlimited links per group, up to 36 groupsLayer 2 VLANs: 4KMSTP: 64 instancesLAG Load Balancing: User Configurable (MAC, IP, TCP/UDPport)IEEE Compliance802.1AB LLDPTIA-1057 LLDP-MED802.1s MSTP802.1w RSTP 802.3ad Link Aggregation with LACP802.3ae 10 Gigabit Ethernet (10GBase-X)802.3ba 40 Gigabit Ethernet (40GBase-X)802.3i Ethernet (10Base-T)802.3u Fast Ethernet (100Base-TX)802.3z Gigabit Ethernet (1000BaseX)802.1D Bridging, STP802.1p L2 Prioritization802.1Q VLAN T agging, Double VLAN T agging,GVRP802.1Qbb PFC802.1Qaz ETS802.1s MSTP802.1w RSTPPVST+802.1X Network Access Control802.3ab Gigabit Ethernet (1000BASE-T) orbreakout802.3ac Frame Extensions for VLAN T agging802.3ad Link Aggregation with LACP802.3ae 10 Gigabit Ethernet (10GBase-X)802.3ba 40 Gigabit Ethernet (40GBase-SR4,40GBase-CR4, 40GBase-LR4, 100GBase-SR10,100GBase-LR4, 100GBase-ER4) on optical ports802.3bj 100 Gigabit Ethernet802.3u Fast Ethernet (100Base-TX) on mgmtports802.3x Flow Control802.3z Gigabit Ethernet (1000Base-X) with QSAANSI/TIA-1057 LLDP-MEDJumbo MTU support 9,416 bytesLayer2 Protocols4301 Security Architecture for IPSec*4302 I PSec Authentication Header*4303 E SP Protocol*802.1D Compatible802.1p L2 Prioritization802.1Q VLAN T agging802.1s MSTP802.1w RSTP802.1t RPVST+802.3ad Link Aggregation with LACPVLT Virtual Link TrunkingRFC Compliance768 UDP793 TCP854 T elnet959 FTP1321 MD51350 TFTP2474 Differentiated Services2698 T wo Rate Three Color Marker3164 Syslog4254 SSHv2791 I Pv4792 ICMP826 ARP1027 Proxy ARP1035 DNS (client)1042 Ethernet Transmission1191 Path MTU Discovery1305 NTPv41519 CIDR1812 Routers1858 IP Fragment Filtering2131 DHCP (server and relay)5798 VRRP3021 31-bit Prefixes3046 DHCP Option 82 (Relay)1812 Requirements for IPv4 Routers1918 Address Allocation for Private Internets2474 Diffserv Field in IPv4 and Ipv6 Headers2596 Assured Forwarding PHB Group3195 Reliable Delivery for Syslog3246 Expedited Assured Forwarding4364 VRF-lite (IPv4 VRF with OSPF andBGP)*General IPv6 Protocols1981 Path MTU Discovery*2460 I Pv62461 Neighbor Discovery*2462 Stateless Address AutoConfig2463 I CMPv62464 Ethernet Transmission2675 Jumbo grams3587 Global Unicast Address Format4291 IPv6 Addressing2464 Transmission of IPv6 Packets overEthernet Networks2711 IPv6 Router Alert Option4007 IPv6 Scoped Address Architecture4213 Basic Transition Mechanisms for IPv6Hosts and Routers4291 IPv6 Addressing Architecture5095 Deprecation of T ype 0 Routing Headers inI Pv6IPv6 Management support (telnet, FTP, TACACS,RADIUS, SSH, NTP)OSPF (v2/v3)1587 NSSA1745 OSPF/BGP interaction1765 OSPF Database overflow2154 MD52328 OSPFv22370 Opaque LSA3101 OSPF NSSA3623 OSPF Graceful Restart (Helper mode)*BGP 1997 Communities 2385 MD52439 Route Flap Damping 2796 Route Reflection 2842 Capabilities 2918 Route Refresh 3065 Confederations 4271 BGP-44360 Extended Communities 4893 4-byte ASN5396 4-byte ASN Representation 5492Capabilities AdvertisementLinux Distribution Debian Linux version 8.4Linux Kernel 3.16MIBSIP MIB– Net SNMPIP Forward MIB– Net SNMPHost Resources MIB– Net SNMP IF MIB – Net SNMP LLDP MIB Entity MIB LAG MIBDell-Vendor MIBTCP MIB – Net SNMP UDP MIB – Net SNMP SNMPv2 MIB – Net SNMP Network Management SNMPv1/2SSHv2FTP, TFTP, SCP SyslogPort Mirroring RADIUS 802.1XSupport Assist (Phone Home)Netconf APIs XML SchemaCLI Commit (Scratchpad)AutomationControl Plane Services APIs Linux Utilities and Scripting Tools Quality of Service Access Control Lists Prefix List Route-MapRate Shaping (Egress)Rate Policing (Ingress)Scheduling Algorithms Round RobinWeighted Round Robin Deficit Round Robin Strict PriorityWeighted Random Early Detect Security 2865 RADIUS 3162 Radius and IPv64250, 4251, 4252, 4253, 4254 SSHv2Data center bridging802.1QbbPriority-Based Flow Control802.1Qaz Enhanced Transmission Selection (ETS)*Data Center Bridging eXchange(DCBx) DCBx Application TLV (iSCSI, FCoE*)Regulatory compliance SafetyUL/CSA 60950-1, Second Edition EN 60950-1, Second EditionIEC 60950-1, Second Edition Including All National Deviations and Group DifferencesEN 60825-1 Safety of Laser Products Part 1: EquipmentClassification Requirements and User’s GuideEN 60825-2 Safety of Laser Products Part 2: Safety of Optical Fibre Communication Systems FDA Regulation 21 CFR 1040.10 and 1040.11Emissions & Immunity EMC complianceFCC Part 15 (CFR 47) (USA) Class A ICES-003 (Canada) Class AEN55032: 2015 (Europe) Class A CISPR32 (International) Class AAS/NZS CISPR32 (Australia and New Zealand) Class AVCCI (Japan) Class A KN32 (Korea) Class ACNS13438 (T aiwan) Class A CISPR22EN55022EN61000-3-2EN61000-3-3EN61000-6-1EN300 386EN 61000-4-2 ESDEN 61000-4-3 Radiated Immunity EN 61000-4-4 EFT EN 61000-4-5 SurgeEN 61000-4-6 Low Frequency Conducted Immunity NEBSGR-63-Core GR-1089-Core ATT -TP-76200VZ.TPR.9305RoHSRoHS 6 and China RoHS compliantCertificationsJapan: VCCI V3/2009 Class AUSA: FCC CFR 47 Part 15, Subpart B:2009, Class A Warranty1 Year Return to DepotLearn more at /Networking*Future release**Packet sizes over 147 BytesIT Lifecycle Services for NetworkingExperts, insights and easeOur highly trained experts, withinnovative tools and proven processes, help you transform your IT investments into strategic advantages.Plan & Design Let us analyze yourmultivendor environment and deliver a comprehensive report and action plan to build upon the existing network and improve performance.Deploy & IntegrateGet new wired or wireless network technology installed and configured with ProDeploy. Reduce costs, save time, and get up and running cateEnsure your staff builds the right skills for long-termsuccess. Get certified on Dell EMC Networking technology and learn how to increase performance and optimize infrastructure.Manage & SupportGain access to technical experts and quickly resolve multivendor networking challenges with ProSupport. Spend less time resolving network issues and more time innovating.OptimizeMaximize performance for dynamic IT environments with Dell EMC Optimize. Benefit from in-depth predictive analysis, remote monitoring and a dedicated systems analyst for your network.RetireWe can help you resell or retire excess hardware while meeting local regulatory guidelines and acting in an environmentally responsible way.Learn more at/Services。
score based generative modelScore based generative model is a type of artificial intelligence that simulates creative processes and allows for the creation of new and innovative ideas. It is based on a multi-score architecture which uses different metrics to generate and evaluate new ideas or solutions. This model is believed to be a more efficient and less time-consuming approach to problem-solving and idea generation.Score based generative models use numerical scores to assess individual ideas or a group of ideas. Each score is then combined to create a cumulative score that can be usedto rank the idea or group of ideas. This evaluation system allows AI models to identify which ideas have the greatest potential for success and also rate and eliminate ideas that may not be worth pursuing.Score based generative models have applications in a wide range of areas, such as machine learning and research. For example, they can be used to assess images or text to support image or language recognition, or to generate novel and creative designs in engineering, music, or art. It can also be used to develop predictive models for understanding customer preferences or for generating product recommendations.Score based generative models can be used to help foster creative ideas in a variety of ways. First, they can help identify the best ideas, making it easier to move forwardwith a project. Second, it can help generate better ideas by allowing for more experimentation. Finally, it can provideunique insights that could not have been discovered through traditional methods.Overall, Score based generative models have the potential to revolutionize both how we think and how we create. By using scores to evaluate and prioritize ideas, AI models can reduce the time and cost associated with discovering great ideas and uncovering new insights.。
base models 和instruction models -回复什么是base models 和instruction models，以及它们在机器学习中的作用和区别。
在机器学习中，模型是一种可用来对数据进行预测或分类的数学表示。
引入“base models”（基本模型）和“instruction models”（指导模型）的概念，有助于更好地理解机器学习中的模型类型和功能。
首先，让我们探索“base models”（基本模型）。
基本模型也被称为“原始模型”或“基础模型”，这些模型通常是最简单的模型形式，能够学习数据的基本模式和关系。
基本模型可以分为两大类：线性模型和非线性模型。
线性模型是最简单的基础模型之一。
它基于线性方程和线性变换的理论，通过对输入特征进行加权求和的方式来预测输出结果。
线性模型在实际应用中有着广泛的应用，尤其在解决分类和回归问题时非常有效。
常见的线性模型包括线性回归、逻辑回归和线性支持向量机（SVM）。
非线性模型是另一类基础模型。
与线性模型不同，非线性模型使用非线性函数来建模数据之间的复杂关系。
非线性模型通常在数据具有高度的非线性结构时表现更好。
常见的非线性模型包括决策树、随机森林、AdaBoost和多层感知机（MLP）等。
尽管基本模型可以对数据进行预测和分类，但它们的处理能力有限。
这就引出了“instruction models”（指导模型）的概念。
指导模型是用来进一步提高模型性能和/或解决复杂任务的模型形式。
指导模型通常使用基本模型的预测结果作为输入，并通过学习这些基本模型的预测误差或残差来建模。
通过使用指导模型，我们可以利用基本模型的优势，并消除其缺点。
指导模型可以分为两大类：集成学习模型和深度学习模型。
集成学习模型通过组合多个基本模型来进行预测。
常见的集成学习方法包括Bagging、Boosting和随机森林等。
这些方法利用了基本模型之间的差异和多样性，从而提高了模型的性能和鲁棒性。
base models 和instruction models -回复什么是base models 和instruction models，它们在机器学习中的作用是什么？在机器学习中，模型是用来对数据进行建模和预测的工具。
基础模型（base models）和指令模型（instruction models）是常见的两种模型类型。
它们在机器学习任务中有不同的使用方式和作用。
首先，让我们来了解基础模型（base models）。
基础模型是机器学习中最基本和最简单的模型类型，通常是在没有任何先验知识或指导下构建的。
基础模型不依赖任何特定任务中的规则或指示，而是依靠大量的数据进行模型训练和学习。
这种模型的主要任务是从数据中学习模式、规律和关联性，并根据这些学习到的模式生成预测结果。
基础模型可以采用多种机器学习方法，包括传统的统计学习方法（例如线性回归、逻辑回归、决策树等）和现代的深度学习方法（例如神经网络）。
无论采用哪种方法，基础模型通常具有较低的复杂度和灵活性，但其训练和预测过程较为简单和高效。
基础模型在机器学习中扮演着重要的角色。
首先，它们经常被用作基准（baseline）模型，用于评估和比较其他更复杂或改进的模型。
通过建立一个基本模型并对其结果进行评估，我们可以确定其他模型相对于基准模型的性能改进程度。
其次，基础模型可以作为整个机器学习流程中的一个组成部分，用来构建更复杂的模型或进行模型集成。
然而，基础模型也有其局限性。
由于其对数据的学习是基于数据本身而没有任何先验知识，它们可能无法捕捉到一些特定任务中的关键规则和指示。
这就引出了指令模型（instruction models）的概念。
指令模型在机器学习中是一类特殊的模型，其使用指示、规则或先验知识来帮助模型进行训练和预测。
这些指令可以是人工设计的规则，也可以是由领域专家提供的先验知识。
指令模型的核心思想是在模型中引入外部知识，从而改善模型的学习和预测能力。
base models 和instruction models -回复Base Models 和Instruction Models：机器学习的两个重要概念在机器学习领域中，Base Models 和Instruction Models 是两个非常重要的概念。
它们分别代表了机器学习模型的训练和预测阶段。
本文将一步一步地回答关于这两个概念的问题，以加深对它们的理解。
什么是Base Models？Base Models 是指机器学习模型的训练阶段中使用的模型。
它们通常是一个简单的模型，作为训练的起点。
在训练过程中，通过不断迭代和优化，Base Models 逐渐提升自己的性能。
Base Models 的选择通常根据问题的复杂度和数据集的大小来决定。
对于简单的问题和小规模数据集，可以使用简单的模型作为Base Model，如线性回归、决策树等。
而对于复杂的问题和大规模数据集，需要使用更强大的模型作为Base Model，如深度神经网络、支持向量机等。
Base Models 的主要作用是捕捉数据中的一些基本模式和关系。
它们提供了一个初始的预测能力，但往往还不够准确。
因此，在训练过程中，需要通过引入Instruction Models 来进一步提升模型的性能。
什么是Instruction Models？Instruction Models 是指在机器学习模型的训练阶段中使用的模型，用于指导和优化Base Models。
它们通过分析和学习数据中的复杂模式和关系，提供更精确的指导，帮助Base Models 不断改进自己的性能。
Instruction Models 的选择往往取决于问题的复杂度和数据的特征。
常用的Instruction Models 包括深度神经网络、集成学习模型等。
在训练过程中，Instruction Models 会根据数据的特征和目标函数，生成一些关于如何优化Base Models 的指示。
NVIDIA RTX 5000 Ada Generation Performance for endless possibilities. DatasheetPowering the Next Era of InnovationIndustries are embracing accelerated computing and AI to tackle powerful dynamics and unlock transformative possibilities. Generative AI is reshapingthe way professionals create and innovate across various domains, from design and engineering to entertainment and healthcare. The NVIDIA RTX™ 5000 Ada Generation GPU, with third-generation RTX technology, unlocks breakthroughsin generative AI, revolutionizing productivity and offering unprecedented creative possibilities.The NVIDIA RTX 5000 Ada Generation GPU is purpose-built for today’s professional workflows. Built on the NVIDIA Ada Lovelace architecture, it combines 100 third-generation RT Cores, 400 fourth-generation Tensor Cores, and 12,800 CUDA® cores with 32 gigabytes (GB) of graphics memory to deliver the next generation of AI graphics and petaFLOPS inferencing performance, accelerating rendering, AI, graphics, and compute workloads. RTX 5000-powered workstations equip you for success in today’s demanding business landscape.NVIDIA RTX professional graphics cards are certified for a broad range of professional applications, tested by leading independent software vendors (ISVs) and workstation manufacturers, and backed by a global team of support specialists. Get the peace of mind to focus on what matters with the premier visual computing solution for mission-critical business.Key Features>PCIe Gen4>Four DisplayPort 1.4a connectors >AV1 encode and decode support >DisplayPort with audio>3D stereo support with stereo connector>NVIDIA® GPUDirect® for Video support>NVIDIA GPUDirect remote direct memory access (RDMA) support >NVIDIA Quadro® Sync II¹ compatibility>NVIDIA RTX Experience>NVIDIA RTX Desktop Manager software>NVIDIA RTX IO support>HDCP 2.2 support>NVIDIA Mosaic² technologyRendering**********************************************(5.2GHzTurbo), 64GB RAM, Windows 11 Enterprise x64, Chaos V-Ray v5.0,NVIDIA Driver 536.15. Relative speedup for 1920x1080 resolution,scene 12 pipeline subtest render time (seconds). Performance basedon pre-released build, subject to change.Omniverse**********************************************(5.2GHzTurbo), 64GB RAM, Windows 11 Enterprise x64, NVIDIA Driver528.49. CAD application performance based on internal testing ofNVIDIA Omniverse Create with several models of varying size andrender complexity. Performance is measured as frames renderedper second. NVIDIA DLSS 3 is enabled for NVIDIA RTX 5000 AdaGeneration GPUs, DLSS 2 enabled for non-Ada generation GPUs.Performance based on pre-released build, subject to change. Training**********************************************(5.2GHzTurbo), 64GB RAM, Windows 11 Enterprise x64, PyTorch v2.1.0,NVIDIA Driver 528.86. Relative speedup for JASPER TrainingPhase, precision = Mixed, batch size = 64. Performance based onpre-released build, subject to change.NVIDIA RTX 5000 Ada Generation | Datasheet | 1Ready to Get Started?To learn more about NVIDIA RTX 5000, visit:/rtx-50001 Quadro Sync II card sold separately. I2 Windows 10 and Linux. I3 Peak rates based on GPU boost clock. I4 Effective FP8teraFLOPS (TFLOPS) using sparsity. I 5 Display ports are on by default for RTX 5000. Display ports aren’t active when usingvGPU software. | 6 Virtualization support for the RTX 5000 Ada Generation GPU will be available in an upcoming NVIDIA vGPUrelease, anticipated in Q3, 2023. | 7 Product is based on a published Khronos specification and is expected to pass the Khronosconformance testing process when available. Current conformance status can be found at /conformance© 2023 NVIDIA Corporation. All rights reserved. NVIDIA, the NVIDIA logo, CUDA, GPUDirect, NVLink, Quadro, and RTX aretrademarks and/or registered trademarks of NVIDIA Corporation in the U.S. and other countries. Other company and productnames may be trademarks of the respective companies with which they are associated. All other trademarks are the propertyof their respective owners. 2788511. JUL23PNY Part Numbers VCNRTX5000ADA-PBYVCNRTX5000ADA-PBVCNRTX5000ADA-EDUVCNRTX5000ADA-BLKVCNRTX5000ADASYNC-PBGPU Memory32GB GDDR6Memory Interface256 bitMemory Bandwidth576GB/sError Correcting Code (ECC)YesNVIDIA Ada LovelaceArchitecture-Based CUDA Cores12,800NVIDIA Fourth-GenerationTensor Cores400NVIDIA Third-Generation RT Cores100Single-Precision Performance65.3 TFLOPS³RT Core Performance151.0 TFLOPS³Tensor Performance1044.4 TFLOPS4System Interface PCIe 4.0 x16Power Consumption Total board power: 250WThermal Solution ActiveForm Factor 4.4” H x 10.5” L, single slotDisplay Connectors4x DisplayPort 1.4a5Max Simultaneous Displays4x 4096 x 2160 @ 120Hz4x 5120 x 2880 @ 60Hz2x 7680 x 4320 @ 60HzEncode/Decode Engines2x encode, 2x decode (+AV1 encodeand decode)VR Ready YesvGPU Software Support6>NVIDIA vPC/vApps>NVIDIA RTX Virtual WorkstationvGPU Profiles Supported See the Virtual GPU licensing guide.Graphics APIs DirectX 12, Shader Model 6.7,OpenGL 4.67, Vulkan 1.37Compute APIs CUDA 12.2, OpenCL 3.0,DirectComputeNVIDIA NVLink ®NoGraphics**********************************************(5.2GHzTurbo), 64GB RAM, Windows 11 Enterprise x64, SPECviewperf2020, NVIDIA Driver 528.49. Relative speedup for 4K SiemensNX composite score. Performance based on pre-released build,subject to change.HPC**********************************************(5.2GHzTurbo), 64GB RAM, Windows 11 Enterprise x64, CUDA 11.8(cuBLAS performance), NVIDIA Driver 525.85. Relative speedupfor GFLOPS, precision = INT8, input = zero. Performance based onpre-released build, subject to change.Generative AI**********************************************(5.2GHzTurbo), 64GB RAM, Windows 11 Enterprise x64, Stable DiffusionWebUI v1.3.1, NVIDIA Driver 536.15. Relative speedup for512x512 image generation. Performance based on pre-releasedbuild, subject to change.。
DatasheetProduct DescriptionAs technologies such as cloud computing, big data, and artificial intelligence continue to develop and grow in popularity, enterprises are deepening their digital transformation, covering aspects such as office, production, and testing. Traditional data centers can no longer keep pace with development, and cloud-based transformation has become an inevitable trend. However, the current data center cloudification solutions currently available in the industry focus on virtualizing resources, improving resource utilization, automating deployment, and implementing cloud-based strategies, but overlook network management and service operation challenges brought by the growing scale and traffic of data centers. Traditional manual O&M cannot effectively deal with complex application migration policies, unstable service experience quality, difficult fault locating, and large-scale security policy management.Huawei iMaster NCE-FabricInsight — a data center network analyzer — eschews the traditional resource status-based monitoring mode. Instead, it detects network health status in real time and monitors networks from the perspective of applications, helping customers detect exceptions in a timely manner while also ensuring continuous and stable application running.Key ComponentsiMaster NCE-FabricInsight (FabricInsight for short) uses Telemetry to collect network-wide traffic and metrics within seconds, analyzes and displays network data through intelligent big data algorithms, and provides northbound APIs to interconnect with upper-layer application systems.HighlightsNetwork-wide health evaluation●Establishes a network health evaluation system based on the five-layer model, implementing 24/7 real-time networkmonitoring.●Intelligently predicts potential risks such as network reliability deterioration and capacity risks, detecting and resolvingproblems in advance.Minute-level fault locating●Builds the network knowledge graph to proactively identify typical faults within 1 minute, automatically locate them within 3minutes, and rectify them within 5 minutes.●Performs network path analysis based on real service flows and demarcates and locates faults in minutes after they arereported.Full network service openness●Provides full network data service openness, drag-and-drop orchestration, and generation of scenario-specific APIs in one-click mode, ensuring integration with third-party systems in days.●Supports integration with service analysis systems, implementing integrated O&M of applications and networks.Key FeaturesTelemetry-Powered Network Visibility in All ScenariosFabricInsight collects statistics on metrics such as devices, boards, queues, interfaces, and entries through Google Remote Procedure Call (gRPC) and displays the dynamic baseline range of each metric using machine learning algorithms. This enables FabricInsight to quickly detect the time point when a baseline exception occurs and proactively identify issues before they interrupt services. In addition, it automatically associates each abnormal time point with the affected service flows, allowing users to view the flow behavior data that passes through the device at the time point when an exception, such as a connection setup failure, occurs.iMaster NCE-FabricInsight Data SheetComprehensive Network Health EvaluationNetwork health check in traditional O&M is inefficient and cannot accurately reflect the network status in real time because it must be performed manually on devices one by one during off-peak hours. FabricInsight takes a different approach. It performs network-wide modeling based on the knowledge graph, constructs a five-layer evaluation system (device, network, protocol, overlay, and service), and intuitively displays the 24/7 network quality. In addition, it dynamically detects key metrics and proactively identifies potential risks such as reliability deterioration and capacity risks.By providing network health evaluation reports in real time or periodically, FabricInsight helps network administrators gain insights into networks and improve O&M efficiency and service experience quality."1-3-5" TroubleshootingData centers are not only service support centers, but also value creation centers. For 98% of enterprises, they will lose more than US$100,000 per hour if their services are interrupted, which is why customers have zero tolerance for network interruptions. Traditional network O&M is mainly performed manually, making it difficult and time-consuming to locate network faults, and severely affecting service continuity.Leveraging Telemetry, FabricInsight collects data on the management, forwarding, and data planes of the entire network in all scenarios, and detects exceptions within 1 minute. In addition, FabricInsight uses the knowledge graph to automatically identify the root causes of faults and potential risks within 3 minutes and provide effective rectification suggestions. Furthermore, FabricInsight collaborates with Huawei iMaster NCE-Fabric to recommend fault handling plans, enabling typical faults to be quickly rectified within 5 minutes.Network Intent VerificationAs the network infrastructure becomes more complex and the network scale becomes larger, checking whether service changes achieve the desired result is critically important. According to a survey conducted by Dimensional Research, 69% of O&M teams manually check network connectivity, resulting in inefficient and incomplete verification.FabricInsight provides service intent verification on the data plane. In key service assurance scenarios such as service changes, FabricInsight delivers 24/7 automatic verification of whether the network intent meets expectations and identifies full-path connectivity. It also detects service and underlay interconnection exceptions within seconds, automatically analyzes root causes for abnormal paths, and notifies users of promptly handling the exceptions.iMaster NCE-FabricInsight Data SheetNetwork Change VisibilityAs data center networks are subject to frequent network changes, traditional manual O&M faces pressing challenges in terms of detecting thousands of device configuration changes and learning tens of thousands of entries per device.With network snapshot management, FabricInsight supports automatic and manual synchronization of network snapshots from dimensions of device configuration, entry, topology, capacity, and performance. In addition, it automatically analyzes differences before and after changes, and clearly displays the detection results.IP 360 ManagementWhen production systems are migrated to the cloud, the VMM automatically completes VM deployment and migration. However, information such as VM node location, VM migration or offline time, and VM distribution cannot be quickly found, meaning that only passive O&M can be performed on the network side.FabricInsight provides IP 360 analysis to quickly learn the number of online VMs and the distribution of top N switches connected to VMs, helping network administrators effectively plan resources in advance. FabricInsight supports full lifecycle management of VMs on the entire network, displays VM logout, migration, and login records in real time, and provides network-wide IP snapshot analysis. It also compares all IP address changes before and after network changes, and checks whether exceptions such as VM logout occur.Intelligent Analysis of Network-Wide LogsAfter a network fault occurs, a large number of logs are generated and 95% of them are invalid logs. In traditional O&M, the manual check of logs one by one is time-consuming and inefficient.FabricInsight visualizes network-wide log events, including the multi-dimensional trends, distribution statistics, and details from Layer 0 to Layer 4. In addition, more than 200 default rules are preset in the system or user-defined rules can be customized to aggregate and clear abnormal logs, improving log analysis efficiency.Network Path AnalysisOnce a service fault is reported, the network department needs to collaborate with the service department to demarcate and locate the fault. Traditional O&M relies on manual analysis of nodes one by one and cannot identify network forwarding paths.FabricInsight can search for the forwarding path of real TCP service flows on the network in one-click mode based on source and destination IP addresses and identify the status of devices, interfaces, and links along the path to quickly demarcate faults. In addition, it automatically recommends key information for fault locating, performs one-click intelligent diagnosis of possible root causes, and locates root causes of service connectivity and poor-QoE issues in minutes.iMaster NCE-FabricInsight Data SheetIntegrated O&M of Applications and NetworksWith the development of digital transformation, as well as the explosive growth of data, application and network systems are separated. Once a service fault occurs, multiple departments need to collaborate and communicate with each other to locate the fault, which is inefficient and cannot meet the requirements of service innovation and development.To address this problem, networks need to quickly provide data services to applications. In traditional mode, APIs are developed one by one based on hard coding and it takes several months to develop a scenario-specific API.FabricInsight provides full network data service openness and quickly rolls out scenario-specific APIs through drag-and-drop operations, implementing interconnection with upper-layer application systems in days. In addition, FabricInsight can collaborate with Netis Business Performance Center (BPC) to achieve integrated O&M of applications and networks. It also detects service quality deterioration in real time and visualizes application and network paths in an E2E mode to quickly locate faults, implementing collaboration between applications and networks and ensuring stable running of services in a timely manner.Unified Multi-DC and Multi-Cloud Network AnalysisAs the digital transformation accelerates, data centers are evolving from single DC mode to multi-DC multi-cloud mode. Traditionally, multiple tools are used for segment-by-segment O&M, lacking unified inter-DC visualized analysis.FabricInsight provides unified O&M analysis for multi-DC networks from an overall perspective.●Analyzes inter-DC/fabric application access traffic and evaluates network health based on north-south, east-west, andintra-DC traffic.●Performs knowledge graph modeling and analysis on inter-DC networks and identifies inter-DC network issues,demarcating and locating faults within minutes.●Automatically verifies whether inter-DC service access meets expectations before and after changes based on DPV.CompositionThe following table describes the basic and value-added packages of FabricInsight.iMaster NCE-FabricInsight Data SheetiMaster NCE-FabricInsight Data Sheet NetworkingFabricInsight supports the following networks:− Hardware-centralized gateway networking − Hardware-distributed gateway networking − Pure IP (IP fabric) networking − Software SDN network (HCS) −Note: 1. The underlay network is deployed based on IP forwarding.2. IP address overlapping scenarios (for example, multi-tenant and VPC scenarios) are not supported.3. VXLAN mapping is not supported.4. The SVF network is not supported.5.HCS supports only Region Type I-Layer 3 networking.Ordering InformationiMaster NCE-FabricInsight provides a 180-day trial license. T o apply for the trial license, visit the ESDP at /isdp/.iMaster NCE-FabricInsight Data Sheet More InformationFor more information about Huawei iMaster NCE-FabricInsight, visit the following link: Copyright © Huawei Technologies Co., Ltd. 2022. All rights reserved.No part of this document may be reproduced or transmitted in any form or by any means without prior writtenconsent of Huawei Technologies Co., Ltd.Trademarks and Permissionsand other Huawei trademarks are trademarks of Huawei Technologies Co., Ltd.All other trademarks and trade names mentioned in this document are the property of their respective holders.NoticeThe purchased products, services and features are stipulated by the contract made between Huawei and thecustomer. All or part of the products, services and features described in this document may not be within thepurchase scope or the usage scope. Unless otherwise specified in the contract, all statements, information, andrecommendations in this document are provided "AS IS" without warranties, guarantees or representations of anykind, either express or implied.The information in this document is subject to change without notice. Every effort has been made in thepreparation of this document to ensure accuracy of the contents, but all statements, information, andrecommendations in this document do not constitute a warranty of any kind, express or implied. Huawei Technologies Co., Ltd. Address: Huawei Industrial Base Bantian, Longgang Shenzhen 518129 People's Republic of China Website:。
A Time-Frame Based Trust Model for GridsWoodas i and Kam-Wing NgDepartment of Computer Science and Engineering,The Chinese University of Hong Kong,Shatin,N.T.,Hong Kong+852-2609-8440{wklai,kwng}@.hkAbstract.A Grid is a virtual resource framework[1,2]which providesthe ability to access,utilize,and manage a variety of heterogeneousresources among multiple domains.However,selecting appropriate re-sources within such a distributed Grid environment is a complex anddifficult task.In recent years,trust models have been proposed[4,5],butmost of them do not deal with the temporal issue of trust.In this pa-per,we aim to extend the trust model with a novel data structure calledTime-Frame which can store temporal information regard-ing to the trustinvolved.With the Time-Frame structure,we can evaluate the resourcesaccurately by means of the available temporal information.1IntroductionA Grid[1,2]is a large-scale distributed computing system.It aims to encour-age the sharing of computational,storage and other resources among multiple domains.Trust is a sociological topic that exists in the human society.An im-portant practical example of trust and reputation is eBay[6](), most likely the largest online auction site.After each transaction,buyers and sellers can rate each other and the overall reputation of a participant is com-puted as the sum of these ratings over the last six months.The eBay example does illustrate the use of trust.However,one may argue that why eBay only takes the last six months’transaction into account,but not a longer period,say, one year.We will further discuss this issue in the coming sections.In this paper, we try to present a data structure called Time-Frame.Time-Frame can be used to extend the Trust model so as to enhance the capability and accuracy.Due to the limited space,we focus on the Time-Frame structure and its operations only in this paper.If you want to look into the detail of what trust and reputation are,please refer to the reference[3].2Discussion on Trust2.1Temporal IssuesTrust and Reputation are usually based on the past transactions of the entities. History can tell how well a resource has behaved.Analogously,in human society, H.Zhuge and G.C.Fox(Eds.):GCC2005,LNCS3795,pp.190–195,2005.c Springer-Verlag Berlin Heidelberg2005A Time-Frame Based Trust Model for Grids191 we always look into someone’s history in order to evaluate whether we should trust him or not.However,to have a fair view on a particular entity,one should study the whole history pertaining to that entity.In the Grid context,it would mean that each entity should store all the past transactions in order to get a more accurate trust evaluation.But this induces huge storage consumption. However,with reasonable storage consumption,it would be beneficial if we can take as many past transactions as we could into consideration when we have to decide whether to trust an entity or not.Our novel data structure,Time-Frame, uses a logarithmic scale structure in order to store the temporal information of a resource.Meanwhile,as mentioned in[4],trust decays with time.Therefore, in our trust model,more weight will be given to the re-cent history.2.2Temporal Failure LocalityIf a resource fails to fulfill the request submitted,there could be lots of rea-sons behind the failure,like outage,system failure or network failure,etc.In reality,failures would take time to recover.Therefore,if we continue to request for service just after a failure has immediately occurred,most likely our request will lead to another failure again as the problem would probably not have been resolved yet.In this paper,we name this phenomenon“Temporal failure local-ity”.Obviously,Trust models should take this phenomenon into account so as to enhance accuracy.3Time FrameThe existing trust models need further enhancements.In section2,some discus-sions were presented.Unfortunately,the existing trust models cannot address all the mentioned issues and this further motivates the evolution.Each Time-Frame represents a period of time.The design of the Time-Frame structure is based on the fact that people are often interested in recent changes at afine granularity,but long term changes at a coarse granularity.Figure1shows such a Time-Frame structure.The Time-Frame structure is constructed based on a logarithmic time scale.Suppose the current Time-Frame holds the transactions in the current quarter.Then the remaining slots are for the last quarter,the last two quarters,4quarters,8quarters,16quarters,etc.,growing at an exponential rate of2.According to this model,with one year of data and thefinest precision at a quarter,we will need log2(365*24*4)+1=around16Time-Frames.It is very space-efficient.In each time frame,two numbers will be stored with respect to the time period specified.They are the number of successful transactions and the number of unsuccessful transactions in that specified time period.In addition,in a Time-Frame structure,intermediate buffer frames need to be maintained.These intermediate frames will replace or be merged with other Time-Frames when they are full.Figure2shows the Time-Frame structure with intermediate buffer frames which exist at each level except for level0.Moreover,i and K.-W.NgFig.1.Time frame with logarithmic partitionFig.2.Time-Frame structure with buffer Time-Frameas only one Time-Frame buffer is needed at any granularity level,the size of the Time-Frame structure can grow no larger than2 log2(N) +2where N is the number of quarter-based batches seen for the Time-Frame structure.3.1Time-Frame UpdatingFirst of all,the stream of transactions is broken up intofixed sized batches (quarter-based).Besides,one Time-Frame structure will be used to maintain the transaction history of one resource domain.Consider a client domain A continues to interact with a resource domain R.In domain A,a Time-Frame structure will then be maintained for resource domain R.Thefirst quarter transaction history between R and A will be kept in the level-0time frame.After that,when the next quarter transaction history comes, the content of the level-0time frame will be shifted to the level-1time frame and the recent quarter transaction history will be kept in the level-0time frame.A Time-Frame Based Trust Model for Grids193Fig.3.Time-Frame structure after the first quarter transactionLater on,when another next quarter transaction history comes,again,the content of the level-0time frame will always be shifted to the level-1time frame.However,the level-1time frame is already occupied and it needs to be shifted upwards too.For the Time-Frame whose level is greater than zero,if it wants to be shifted to the next level,it should always check whether the buffer Time-Frame at its level is being occupied or not.There will be two different cases.If the buffer time frame at that level is not occupied,thecontent of that Time-Frame should be placed in the buffer Time-Frame.Instead,if the buffertimeFig.4.Final Time-Frame structurei and K.-W.Ngframe at that level is occupied,the content of that Time-Frame will be merged with the buffer time frame and then be shifted to the next level.This process will be propagated to the next level until the highest level Time-Frame.Let us take an example here to illustrate the updating of the Time-Frame structure.Consider a client domain A continues to interact with a resource domain R.In thefirst quarter period,10transactions have been performed and 8are successful.Thefirst quarter information will be kept in the level-0time frame and Figure3shows how the Time-Frame structure looks like.In the next quarter,14transactions have been performed and6of them are successful.After that,12transactions have been performed and8of them are successful.The interaction continues and15transactions have been performed and10of them are successful in next quarter.Figure4shows thefinal Time-Frame structure for the previous1-hour transactions.The Time-Frame structure will be updated continuously according to the above discussion when the next quarter information arrives.4Analysis with Time-FrameWith the Time-Frame structure,the transaction history of a domain will now be stored in different Time-Frame partitions.Each Time-Frame represents a period of time.However,the length of the period is not equal for all the Time-Frames. Instead,they are in logarithmic scale.Logarithmic scale Time-Frame helps to give more weight to the recent history than the ancient history.However,the terms ancient and recent are not absolute,they are somehow related to the computing environment,such as the transaction rate is high or not.In general, as a Grid task can last for a few days to be completed and recorded,therefore,we assume that the recent5-day’s activities are the most important for a resource domain to be evaluated.To evaluate how good our logarithmic Time-Frame structure is in giving bias to the recent5-days transaction history,we have the following analysis.Consider we use16Time-Frames to hold the transaction history for8216hours1year. Out of these16Time-Frames,10of them will be used to hold the time period of length128hours5days.The ratio of5days to365days=5/365=0.014. However,using our logarithmic Time-Frame structure,the ratio of the5-days transaction history to the365-days transaction history is=10/16=0.625.We actually scale the5-days transaction history up by0.625/0.014=44.6times.To conclude,the weight of these recent5-days activities is more than half when we are actually considering one-year transaction history.This feature of Time-Frame can help the existing trust models to give bias to the recent history and further address the issues mentioned in Section2.5ConclusionNowadays,many resources are available through the Internet.To select the best out of these resources,it is not a simple issue but relies on trust and reputation.A Time-Frame Based Trust Model for Grids195 However,the existing trust models do not address the temporal issues of trust as mentioned in Section2.Therefore,Time-Frame,a novel data structure,is introduced in this paper.Time-Frame can be integrated into the existing trust models so as to enhance the capability and accuracy.References1.Ian Foster,The Grid:A New Infrastructure for21st Century Science,Physics Today,Febru-ary2002.2.Ian Foster,Carl Kesselman,and Steve Tuecke,The Anatomy of the Grid,EnablingScalable Virtual Organizations,International Journal of Supercomputer Applica-tions,2001.3.Alfarez Abdul-Rahman,Stephen Hailes Supporting Trust in Virtual Communi-ties.In Proceedings Hawaii International Conference on System Sciences33,Maui, Hawaii,4-7January2000.4.Farag Azzedin and Muthucumaru Maheswaran,“Evolving and Managing Trust inGrid Computing Systems,”Proceedings of the2002IEEE Canadian Conference on Electrical Computer Engineering.5.Farag Azzedin and Muthucumaru Maheswaran,Integrating Trust into Grid Re-source Man-agement Systems,Proceedings of the International Conference on Par-allel Processing(ICPP’02).6.“ebay,”Web Page.[Online].Available:.。
