Submitted to the ApJ Boron Abundances in Early-B Stars Results from the B III Resonance Lin

拜登国情咨询文2023英文As President Biden delivers his 2023 State of the Union Address, he presents a vision for the future of Americathat is both ambitious and challenging. In his address, Biden outlines his administration's priorities for the coming year, focusing on areas such as domestic investment, national security, and the ongoing pandemic.Domestically, Biden emphasizes the need for continued investment in infrastructure, education, and research and development. He calls for Congress to pass his proposed American Jobs Plan, which aims to create millions of new jobs and boost economic growth through investments in areas like transportation, broadband internet, and clean energy. Biden also emphasizes the importance of education, calling for an increase in federal funding for public schools and expanded access to higher education.National security remains a top priority for Biden, who vows to continue to work with allies and partners to address global challenges. He calls for Congress to approve his proposed budget for national defense, which includes increases in funding for the military and intelligenceagencies. Biden also emphasizes the need to address cybersecurity threats, calling for increased investment in cyber infrastructure and stronger protections for personal data.The ongoing pandemic remains a significant challenge for Biden and his administration. He calls for Congress to pass his proposed pandemic response plan, which includes expanded access to testing, vaccines, and therapeutics. Biden also emphasizes the need for continued investment in public health infrastructure, such as hospitals and community health centers, to ensure that the country is prepared to respond to future outbreaks.In his address, Biden also calls for unity and bipartisanship in the face of these challenges. He emphasizes the need for Congress to work with his administration to pass legislation that will benefit all Americans, regardless of party affiliation. Biden recognizes that achieving these goals will require compromise and concessions from all sides, but he believes that it is necessary to move forward as a nation.The 2023 State of the Union Address is a testament to President Biden's commitment to leading America through a period of transition and uncertainty. His vision for the future is both bold and achievable, and his call for unity and bipartisanship is a necessary reminder that we are stronger when we work together as a nation. As we move forward into the new year, it is important that we remember the challenges that lie ahead and the opportunities that await us if we are willing to work together and invest in our future.**拜登总统2023年国情咨文：引领未来的愿景与挑战** 当拜登总统发表2023年国情咨文时，他展现了一个既雄心勃勃又充满挑战的美国未来愿景。

美参议院批准马克斯·鲍卡斯为新任驻华大使The Senate easily confirmed longtime Sen. Max Baucus on Thursday to become ambassador to China, handing the job to a lawmaker well-versed in U.S. trade policy but with little expertise about military and other issues that have raised tensions with Beijing.美国参议院2月6日以96票支持、0票反对的表决最终批准马克斯·鲍卡斯为新任驻华大使。

结果宣布前后，鲍卡斯接受了同事的祝贺。

Senators gave final approval to the nomination of the moderate Baucus, D-Mont., by 96-0. Trim and youthful looking for his 72 years, Baucus accepted colleagues' congratulations before and during the roll call and voted "present" for his own nomination.Chairman of the Senate Finance Committee, Baucus has worked with Republicans during his 35 years in the Senate on issues ranging from taxes to health care reform -- an independent streak that has vexed Democratic colleagues.Baucus supported GOP President George W. Bush's sweeping 2001 tax cuts and his 2003 creation of Medicare prescription drug benefits, despite opposition by most Democrats.He also helped write President Barack Obama's 2010 health care overhaul. Congress approved that measure solely with Democratic votes after Baucus spent months trying to craft a bipartisan（两党连立的） version with Republicans."I'm proud to stand up for it, because it is helping millions of Americans," Baucussaid in farewell remarks on the Senate floor after Thursday's vote, defending a law that Republicans are making a top issue this election year.Most recently, he'd been working with Republicans to craft bills revamping the income tax system and streamlining congressional votes on trade treaties Obama is seeking with Asia and Europe."His passion is well-known to all of us, his decades of experience here in Congress. He's an excellent choice that President Obama made to represent America's interests in China," said Senate Majority Leader Harry Reid, D-Nev.Sen. Orrin Hatch, R-Utah, a long-time Finance panel colleague, lauded the Montanan for "his willingness to put partisan differences aside for the greater good of all."In 2001, Baucus championed China's inclusion in the World Trade Organization, which has helped Beijing conduct business with other countries. He has sponsored legislation -- not enacted -- punishing China for undervaluing its currency and has criticized Beijing for blocking U.S. imports of beef, a big business in his rural state.Baucus had already announced he would not seek re-election this fall when Obama tapped him last December for the ambassadorship.His early departure from the Senate -- spokesman Sean Neary said Baucus planned to formally resign late Thursday -- could turn what looked like a likely GOP Senate seat pickup into a more competitive race.Democratic Montana Gov. Steve Bullock planned a Friday news conference at which he was expected to name a temporary replacement for Baucus, perhaps Democratic Lt. Gov. John Walsh. That could let the appointee build a Senate record and gain visibility to defend the seat against expected GOP candidate Rep. Steve Daines.Baucus' confirmation comes as China has emerged as a leading global economic and military power, at times causing strains between Beijing and Washington and its allies in the region.China's economy is second in size only to that of the United States. The U.S. trade deficit with China hit $318 billion last year, far larger than it is with any other country, and the $1.3 trillion in Treasury securities China owns make it the largest foreign holder of U.S. debt.Ties have also frayed as the U.S. has accused China of cybertheft of American intellectual property and of holding down the value of its currency to give a price advantage to its companies trading overseas.U.S. officials have clashed with China over its human rights record, accusing it of mistreating political dissidents and many minority groups. On Thursday, Obama said at the annual National Prayer Breakfast that he has told leaders in Beijing that "realizing China's potential rests on upholding universal rights."Beijing has been modernizing its military forces and engaging in territorial disputes in the seas off East Asia with Japan, South Korea and the Philippines.Baucus will replace Gary Locke, the former Commerce secretary and first Chinese-American to serve as U.S. ambassador to China.。

Utilization-Based Techniques for Statically Mapping Heterogeneous Applications onto the HiPer-D Heterogeneous Computing System Shoukat Ali†,Jong-Kook Kim†,Yang Yu∗,Shriram B.Gundala∗,Sethavidh Gertphol∗,Howard Jay Siegel‡§,Anthony A.Maciejewski‡,and Viktor Prasanna∗†Purdue University∗University of Southern California School of Electrical and Computer Engineering Department of Electrical Engineering West Lafayette,IN47907-1285USA Los Angeles,CA90089-2560USA {alis,jongkook}@{yangyu,gundala,gertphol,prasanna}@Colorado State University‡Department of Electrical and Computer Engineering§Department of Computer ScienceFort Collins,CO80523-1373USA{hj,aam}@Originally submitted May2002Revised November2002Submitted to PDCPSpecial Issue on Algorithms,Systems and Tools for High Performance ComputingAbstractThis research investigates the problem of allocating a set of heterogeneous applications to a set of heterogeneous machines connected together by a high-speed network.The proposed resource allocation heuristics were implemented on the High Performance Distributed Computing Program’s (HiPer-D)Naval Surface Warfare Center testbed.The goal of this study is to design static resource allocation heuristics that balance the utilization of the computation and network resources while ensuring very low failure rates.A failure occurs if no allocation is found that allows the system to meet its resource and quality of service constraints.The broader goal is to determine an initial resource allocation that maximizes the time before run-time re-allocation is required for managing an increased workload.This study proposes two heuristics that perform well with respect to the load-balancing and failure rates.These heuristics are,therefore,very desirable for HiPer-D like systems where low failure rates can be a critical requirement.This research was supported by the DARPA/ITO Quorum Program through the Office of Naval Research under Grant No.N00014-00-1-0599,and by the Colorado State University George T.Abell Endowment.Some of the equipment used was donated by Intel and Microsoft.1.IntroductionWith the widespread use of increasingly powerful commercial off-the-shelf(COTS)products, some system designers have started a shift from custom development to COTS-based systems to achieve lower costs and moreflexible systems[GiK99,HaZ95].However,to use COTS-based systems effectively as parts of a larger system,one needs to exploit the heterogeneity in processor speeds, memory structures,specialized hardware capabilities,etc.,that most likely will be present in different COTS products.Heterogeneous computing(HC)is the coordinated use of different types of machines, networks,and of widely varying application mixtures and to maximize the combined performance or cost-effectiveness,e.g.,[BrS01b,Esh96,FrS93,MaB99].An important research problem in HC is mapping,i.e.,how to assign resources to applications so as to maximize some performance criterion violating any resource and quality of service constraints.This research addresses the problem of mapping in a High Performance Distributed Computing Program(HiPer-D)[HaZ95]like HC system consisting of heterogeneous sets of sensors, applications,Each sensor produces data sets periodically at afixed rate (the rate may be different for different sensors),and these data sets are input into applications. The applications process the data sets and send the output to other applications or to actuators. Each application is required tofinish processing a given data set before the next one arrives(this is the throughput constraint on each application).Each sensor can be characterized by the amount of “load”constituted by a given data set(e.g.,number of objects in the data set to be processed).The system is expected to operate at a given value of initial workload(i.e.,the set of sensor load values). However,unpredictable changes in the initial sensor loads are likely over time as the content of data sets generated by the sensors changes.This causes unpredictable increases in the execution and communication times for different applications.This means that,at a certain value of the increased workload,some applications may not be able to process a given data set before the next one arrives, causing a throughput violation.Such increases in the workload could require a run-time re-mappingthe increased workload.The goals of this research are to design a mapping heuristic that(1)produces an initial mapping that maximizes the time before a run-time re-mapping is required,and(2)has a very small“failure rate.”It is possible that when a given heuristic is executed for a particular scenario,the heuristic might not be able tofind a mapping(e.g.,if the heuristic is left with an application that needs more CPU utilization than is available on any machine).This may happen because no feasible mapping exists or because the heuristic made some“bad”assignments early on in its operation so as to leave not enough capacity on the machines for the remaining applications.The failure rate for a given heuristic is defined in this research as the ratio of the number of instances in which the heuristic fails tofind a resource allocation to the total number of times the heuristic was executed in different scenarios.In an ideal situation,one would know the dependence of the computation and communication times of different applications on changes in the workload from its initial state,and use that infor-mation to determine a mapping that could meet thefirst goal.Instead,the information that was provided was,for each application,the worst-case values of the minimum CPU,input network link, and output network link utilizations that the application requires on a given machine to satisfy the throughput constraint at the initial workload.This research uses minimizing U max,the utilization of the most utilized computation or network resource,as an approximate approach to reaching the first goal.This approach is reasonable because,generally,as the load from a given sensor increases,the applications that receive data sets from that sensor will utilize more resources.If the system is minimally utilized initially,it can better“absorb”such increases in the workload.The contribution of this research is the design of two mapping heuristics,MIP*and HRA Max-min,for the initial allocation of resources to applications in the HiPer-D environment.MIP*and HRA Max-min are compared with several other heuristics from the related literature by using simulation experiments,and are seen to outperform the other heuristics,particularly for HC systems with high heterogeneity.The heuristics perform well in the sense that they have very small failure rates,and generate mappings with low values of U max(as compared to other heuristics).These heuristics are, therefore,very desirable for HiPer-D like systems where increases in workload are likely(assuming that systems that are minimally utilized initially can“absorb”such increases)and low failure rates can be a critical requirement.Note that even though HiPer-D is a real-time system,the contribution of this research is not in guaranteeing the real-time operation of the system,but in designing an initial mapping(to be used when the system isfirst started)that delays the need for real-time, dynamic,on-line re-mapping due to changes in the workload that will cause a throughput violation.The heuristics discussed in this research are“static”mapping heuristics.Static mapping is performed when the applications are mapped in an off-line planning phase[BrS01a],e.g.,when a system isfirst started up and a mapping is needed to ensure that all quality of service(QoS) constraints will be met for a given initial workload.The mapping problem has been shown,in general,to be NP-complete[Cof76,Fer89,IbK77].Thus,the development of heuristic techniques tofind near-optimal mappings is an active area of research,e.g.,[AlK02,BaV01,BrS01a,BrS01b, Esh96,FoK99,GeY02,MaA99,MiF00,WuS00].MSHN(Management System for Heterogeneous Networks)is a collaborative research effort State University,Purdue University,the University of Southern California,NOEMIX, and the Naval Postgraduate School[HeK99].It was supported by the DARPA/ITO Quorum Pro-gram.One objective of MSHN is to design and evaluate mapping heuristics for different types of HC environments,including the COTS-based HiPer-D environment at the Naval Surface Warfare Center (NSWC)[HaZ95].A specific example of a HiPer-D subsystem is shown in Figure1.Fire Sim21, OTH(Over the Horizon)Data Server,and ALT(Air Engagement Control Local Track)Data Server send periodic data to the applications.Tacfire,CFF(Call for Fire)Broker,Land Attack Engagement Server,Deconflict Server,Gun Control,and Display Components are the applications to be mapped. The arrows denote communications,and the labels next to them denote the network protocols used for communications.The labels in parentheses next to the applications denote the types of machines on which those applications can execute.The HiPer-D system consists of a large number of such subsystems.The rest of the paper is organized as follows.The system model is described in Section2. Section3presents the static mapping heuristics.The details of the simulation experiments are given in Section4.Section5outlines how this work is related to the previous work in this area.Section6 concludes the paper.2.ModelThe system consists of heterogeneous sets of sensors,applications,machines,and actuators. Each machine on the network has a full-duplex communication link to a non-blocking switch.The sensors and actuators have unidirectional connections(Figure2).Each sensor produces data sets periodically,and these data streams are fed into applications.The applications process the data setsFigure1:An example HiPer-D subsystem composed of heterogeneous COTS machines and net-works.and send the output to other applications or to actuators(Figure1).Figure2:The hardware model.All machines have dedicated,full duplex communication links toa non-blocking switch.Let M be the set of machines in the system.Each machine in M has some“background loads”on its CPU and input/output network links.The background loads on a machine are the utilizations of the CPU,input link,and output link before mapped on the system.Let A be the set of applications that need to be mapped.All applications in A execute con-tinuously to process the periodic inputs that arrive from the sensors or to process input data from predecessor applications.An application can start processing input data as soon as the data is avail-able and the application hasfinished processing prior data inputs.Recall that each application is required tofinish processing a given data set before the next one arrives.An application is characterized by the worst-case values of the minimum CPU,input network link,and output network link utilizations that it requires on a given machine to satisfy the through-put constraint at the initial workload.That is,each utilization is a worst-case value for the fraction of the resource required by the given application on a given machine to process a data set(based onits initial load value)within its allowed time period.The utilization of resources information is the required amount of resource to ensure that an application does not violate its throughput constraint. As an example of the relation between the utilization of resource and throughput constraint consider the following.Assume that application A needs20%of machine1for the given initial workload. If application A gets20%of the CPU cycles on machine1,then it can meet the throughput con-straint.The20%is the percentage of machine cycles needed(as specified by the sponsor,NSWC) irrespective of the OS multitasking scheduling method within a given machine.This information is the experimental data for the given application,given machine,sensor output rate,and workload.Let C(a i,m j)be the CPU utilization that application a i requires on machine m j,with the initial workload,to process a data set within the required time period(based on the rates of its associated sensors).Analogously,let I(a i,m j)and O(a i,m j)be the input network link and output network link utilizations,respectively,that application a i requires on machine m j.Note that the values of C(a i,m j),I(a i,m j),and O(a i,m j)are associated with the initial workload.Let C bgj ,I bgj,and O bgjbe the background utilizations on the CPU,input network link,andoutput network link,respectively,of machine m j.Let A j be the set of applications already mapped on machine m j.Let C j,I j,and O j be the total utilizations on the CPU,input network link,and output network link,respectively,of machine m j.Then,C j=C bgj +a i∈A jC(a i,m j),I j=I bgj +a i∈A jI(a i,m j),andO j=O bgj +a i∈A jO(a i,m j).The total utilization of the most heavily loaded resource(the CPU,input link,or output link)of machine m j is given by U j=max(C j,I j,O j).Ensuring that no resource is more than100%utilized implies that,for1≤j≤|M|,U j≤100%.Recall that the performance objective for the mappings in this study is the minimization of U max, the utilization of the most heavily loaded resource,from now on called the maximum utilization.Note that U max=max1≤j≤|M|(U j).Including the background load on resources,the utilization of any resource on any machine cannot exceed100%.This condition ensures that no application violates its throughput constraint. If the resource utilization exceeds100%,this means that some or all applications on that machine will violate their throughput constraint.If the total utilization of a resource is less than100%,all applications mapped onto the resource do not violate their throughput constraint.3.Mapping HeuristicsThis study examinesfive mapping heuristics,namely:(i)the Min-min heuristic,(ii)the Max-min heuristic,(iii)the host-restriction-aware(HRA)Min-min heuristic,(iv)the host-restriction-aware (HRA)Max-min heuristic,and(v)the Mixed-Integer-Programming-based heuristic(referred to as MIP*in the following text).The HRA Min-min,HRA Max-min,and MIP*heuristics are the three heuristics proposed in this research;Min-min and Max-min are used here for comparison purposes.The Min-min heuristic is based on[IbK77],and is one of the heuristics implemented in SmartNet [FrG98].Some variants of the Min-min heuristic were studied in,e.g.,[Arh98,BrS01a,BrS02,MaA99, WuS00],and were seen to perform well in many different environments.Formally,the version of the Min-min heuristic designed for this system can be defined as follows. Let U r,j be the total utilization of the most heavily loaded resource(the CPU,input link,or output link)of machine m j,if the currently unmapped application a r is mapped on machine m j.That is,U r,j=max(C j+C(a r,m j),I j+I(a r,m j),O j+O(a r,m j)).LetU∗=mini:a i∈(A−Sj:m j∈M A j)(minj:m j∈MU i,j).The outer minimum in the preceding expression is taken over all unmapped applications(i.e.,A−j:m j∈MA j).Figure3shows the pseudo-code used to implement Min-min for this system.Min-minselects the x,y for which U x,y=U∗≤1,assigns a x to m y,adds a x to A y,and updates U y to reflect the assignment.The above process is repeated until all applications are mapped.If U∗>1in some iteration,a mapping cannot be found with this heuristic.The time complexity of Min-min is O(|M||A|2).(1)do until all applications are mapped(2)for each unmapped application a r,findthe machine m k such thatU r,k=minj:m j∈MU r,j and U r,k≤100%(3)if no such machine found,this heuristic cannotfind a mappingstop(4)from the(a r,m k)pairs found in step(2),select the pair(a x,m y)forwhich U x,y=min(ar,m k)U r,k(5)assign the application a x to the machine m y(6)mark the application a x as mapped(7)update C y,I y,and O y(8)end doFigure3:The Min-min heuristic.The Max-min heuristic is similar to the Min-min heuristic,and is also one of the heuristics implemented in SmartNet[FrG98].It differs from the Min-min heuristic in that nowU∗=maxi:a i∈(A−Sj:m j∈M A j)(minj:m j∈MU i,j),and in Figure3,“min(ar,m k)”in step(4)is replaced with“max(ar,m k).”Max-min is likely to dobetter than Min-min in scenarios like the one shown in Figure4.Here,the system consists of threeapplications and two machines.The table in Figure 4shows the C (a i ,m j )values.For all i,j ,I (a i ,m j )=O (a i ,m j )=C bg j =I bg j =O bg j =0.Min-min maps a 0and a 1before mapping a 2,whileMax-min maps a 2first and then a 1and a 0.The preceding example shows one difference between Max-min and Min-min in a hypothetical situation.The real comparative study is in Section 4.Note that the Max-min and Min-min heuristics as described above are directly based on their classical counterparts.This research has adapted the objective function to fit the problem and environment,so that these heuristics could be used here to compare with our proposed heuristics.In addition,note that the time complexity of Max-min is the same as that of Min-min,i.e.,O (|M||A|2).Max-min mapping Min -min mapping2%11%18%15%40%20%m 1m 0a 0 a 1a 2 Figure 4:A scenario where Max-min performs better than Min-min.The host-restriction-aware Min-min heuristic (HRA Min-min)considers the fact that in many systems a given application may not be able to execute on all machines in the system.This may arise because the application is not compiled for all machines or it requires specialized capabilities available only on select machines.In such systems,the Min-min or Max-min heuristics may fail to find “obvious”mappings for some cases.One such case is shown in Table 1,where,for all i,j ,I (a i ,m j )=O (a i ,m j )=C bg j =I bg j =O bg j =0.The symbol ∞for an entry C (a i ,m j )indicates that application a i cannot execute on machine m j .Min-min assigns a 1to m 1in the first iteration,thereby depriving a 0of the only machine on which it could execute.Similarly,Max-min first assigns a 0to m 1,and then assigns a 1to m 0in the second iteration,thereby depriving a 2of the only machine on which it could execute.Hence,both Min-min and Max-min fail to find the obvious mapping (a 0on m 1,a 1on m 2,and a 2on m 0).The HRA Min-min heuristic,described next,does find the obvious mapping.The HRA Min-min heuristic is shown in Figure 5.In each iteration,the heuristic splits the unmapped applications into two sets,and tries to map those sets separately.Let U p 1and U p 2be the two sets of unmapped applications in iteration p .Let S (k )be the set of applications that can map on exactly k machines.In the first iteration,the heuristic splits all applications such thatU 11=S (1)and U 12=A −S (1).Then it attempts to map the applications in U 11onto their respective machines.If that partial mapping is not successful,then no mapping exists.If this partial mappingis successful,the heuristic saves the partial mapping,and then tries to map U 12by using Min-min.Ifthe mapping of U12fails,the heuristic undoes any changes it made to the system while trying tofindthe mapping of U12,and then moves to the second iteration.In the second iteration,U21=S(2)andU2 2=A−U21−U11.In general,before the N-th iteration,S(i)for1≤i<N has been mapped.Atthat time,U N1=S(N)and U N2=A−Nk=1U k1.For all N,HRA Min-min uses Min-min to map U N1and uses Min-min to map U N2.(For the case of U11,performing Min-min is equivalent to assigningeach application in U11to the only machine on which it can execute.)Also note that if S(N)=∅forthe N-th iteration,HRA Min-min simply proceeds to the(N+1)-th iteration without performing either of the two Min-min operations in iteration N.A time complexity analysis for HRA Min-min is now presented.The time complexity of deter-mining the set S(N)for1≤N≤|M|is O(|M||A|).The maximum number of while iterations(Line 2,Figure5)is equal to the total number of machines in the system.If a complete mapping is found in the i-th while iteration,then the running time for the k-th while iteration(k≤i)is given by the sum of the running times of the two Min-min operations performed in iteration k.Mathematically, the running time for the k-th iteration is O(k|S(k)|2)+O(|M||A−kj=1S(j)|2)=O(|M||A|2). Given this,HRA Min-min time complexity=O(|M||A|)+ik=1O(|M||A|2).Assume that a complete mapping is found in thefirst iteration.(This was found fairly common in the experiments performed in this research.)Then,the time complexity for HRA Min-min is equal to that for Min-min.In the worst case,HRA Min-minfinds a complete mapping in|M|while iterations,and the time complexity could be up to O(|M|2|A|2).m0m1m2a0∞60%∞a160%45%70%a250%∞∞Table1:An example scenario showing C(a i,m j)values for a system of three applications and three machines where HRA Min-minfinds the obvious mapping,but Min-min andMax-min do notfind any feasible mapping.The host-restriction-aware Max-min heuristic(HRA Max-min)is similar to the HRA Min-min except that it uses the Max-min heuristic instead of Min-min.That is,in steps(4)and(6)in Figure 5,Max-min is used instead of Min-min.The time complexity analysis for HRA Max-min is the same as that for HRA Min-min.The MIP*heuristic is based on the well-researched mixed integer programming(MIP)math-ematical technique for optimization[Kau77].A mathematical programming formulation based on the model in Section2is developed to map the applications onto machines.The set{x ij}defines a(1)N=1(2)while(N≤|M|)//if there are any applications that can execute on//only N machines,map those applicationsfirst(3)if(|S(N)|>0)(4)use Min-min tofind a mapping for S(N)(5)if a mapping for S(N)is not foundthis heuristic cannotfind a mappingstop(6)use Min-min tofind a mapping for all of theremaining applications,marking eachassignment as“speculative”(7)if a complete mapping is not found in step(6)//roll back-undo all changes to//the system data structures,and//perform the next iteration(8)for each speculative assignment(a i,m j)made in step(6)(9)undo the mapping of a i on m j(10)mark application a i as unmapped(11)undo the increases in the CPU,input link,and output link utilizations of machinem j that were caused by speculativemapping of a i on m j(12)N=N+1(13)else//matches the“if”in step(7)(14)return mapping(15)else//matches the“if”in step(3)(16)N=N+1(17)end whileFigure5:The HRA Min-min heuristic. mapping of applications onto machines such thatx ij=1if application a i is mapped onto machine m j 0otherwisewhere1≤i≤|A|and1≤j≤|M|.In terms of x ij,C j=C bgj +1≤i≤|A|(x ij×C(a i,m j)),I j=I bgj +1≤i≤|A|(x ij×I(a i,m j)),andO j=O bgj +1≤i≤|A|(x ij×O(a i,m j)).Let M(a i)be the set of hosts onto which application a i can be mapped.Figure6shows the MIP formulation,where U is an auxiliary variable that will equal the minimum value of U max when the optimization is complete.There are many commercial MIP solvers available that,given enough time, can optimize an MIP problem instance like that in Figure6.The MIP solver,Lindo[Sch91],used in this study employs a branch-and-bound technique tofind a solution to a given MIP formulation.In this study,the objective of the MIP formulation is to minimize U max based on the constraints that both CPU and network utilizations of each machine are less than or equal to100%.(However, this approach can be extended to optimize more complex metrics.)The last two constraints in Figure6force application a i to be mapped onto exactly one machine in M(a i).Because the above objective function minimizes the maximum utilization(CPU or network) among all machines,the mapping of applications on the less utilized machines may not be necessarily optimized.To try to achieve system-wide optimization,the MIP*heuristic uses an iterative method to solve the problem.In each iteration,MIP*removes from further consideration the most utilized machine and all of the applications mapped thereon,and then attempts to optimize the assignments of the remaining applications on the remaining machines.The mapping is described as a set,T,of |A|two-tuples,where T={T1,...,T|A|}.Each tuple T i is in the form(a i,m j),where a i∈A and m j∈M.Note that there is a tuple(a i,m j)in T if and only if x ij=1.The complete pseudo-code is shown in Figure7.When the stopping condition M∗=∅is reached,A∗=∅(unless the heuristic fails tofind a mapping).Note,however,that for the performance metric being optimized for this investigation,only one iteration of the repeat-until loop(Figure7,Lines4-12)is sufficient.The time complexity of MIP*in the worst case is the same as that for exhaustive search.In real life situations,the running time is usually better than exhaustive search.However,an analysis of the average case time complexity is difficult to derive.See[ZhK95]for a detailed analysis of the branch-and-bound method.In addition to thefive heuristics mentioned above,this study also examined a fast greedy heuris-tic,a random allocation heuristic,and a lower bound(LB)on the maximum utilization.The fast greedy heuristic and the random allocation heuristics are shown in Figure8.Note that,unlike the Min-min or Max-min heuristics,the fast greedy and the random allocation heuristics iterate through the set of applications only once.Both the fast greedy and random allocation heuristics have time complexities of O(|M||A|).The lower bound on the maximum utilization is calculated by assuming that for all applications I(a i,m j)and O(a i,m j)are zero,that each application a i is mapped on the machine m j where C(a i,m j)is minimum over all machines,and that the sum of the utilizations can be divided equallygiven M,A,{C(a i,m j)},{I(a i,m j)},{O(a i,m j)}and a real number Ufind x ij and UtominimizeUsubject toU≤100%∀m j∈M,C j≤U∀m j∈M,I j≤U∀m j∈M,O j≤U∀a i∈A,j:m j∈M(a i)x ij=1∀a i∈A,j:m j/∈M(a i)x ij=0Figure6:The mixed integer programming formulation.(1)initialize T to∅(2)let M∗and A∗denote the set of machinesand applications that need to be mapped(3)initialize M∗to M and A∗to A(4)repeat(5)using M∗and A∗,construct a MIP problem instance(based on the MIP formulation shown in Figure6)(6)solve the MIP problem instance using an MIP solver(7)find out the machine m x that has thehighest CPU or network utilization(8)for each application a i∈A x//record the mapping information//regarding m x in T(9)add(a i,m x)into T(10)delete a i from A∗(11)delete m x from M∗(12)until M∗=∅Figure7:The MIP*heuristic.over all of the machines(which,in general,may not be physically realistic).Specifically,LB=1≤i≤|A|min1≤j≤|M|C(a i,m j)+1≤j≤|M|C bgj/|M|.An example of when this lower bound situation could occur is:(1)all applications that communicate with each other are mapped to the same machine,(2)each application is mapped to its best machine, and(3)the set of applications is such that all machines are equally utilized.//iterate through the applications//in an arbitrary order(1)for r=1to|A|(2)find the machine m k such thatU r,j and U r,k≤1U r,k=minj:m j∈M(3)if no such machine found,this heuristic cannotfind a mappingstop(4)assign the application a r to the machine m k(5)update C k,I k,and O k(6)end for(a)//iterate through the applications//in an arbitrary order(1)for r=1to|A|(2)identify the set,L,of machines such thatif a r is mapped on m j∈L,U r,j≤1(3)if L is empty,this heuristic cannotfind a mappingstop(4)assign a r to a randomly chosenmachine from L(5)update C j,I j,and O j(6)end for(b)Figure8:(a)The fast greedy heuristic.(b)The random allocation heuristic.4.Simulation Experiments and ResultsIn this study,several sets of simulation experiments were conducted to evaluate and compare the heuristics.For all experiments,the number of machines in the system wasfixed at ten.Also,it was assumed that every application could execute on at least one machine.That machine was chosen randomly from among all of the machines in the system.For any other machine,the probability that a given application could execute on it was50%.The C(a i,m j)matrix was generated by sampling a probability distribution D C.The entries in(heterogeneity the C(a i,m j)matrix were generated to have a mean M C,a“task heterogeneity”H Ctask。

2020辽宁考研英语阅读理解："量子霸权"According to a report in the Financial Times, a team of researchers from Google led by JohnMartinis have demonstrated quantum supremacy for the first time.《英国金融时报》的一篇报告称，John Martinis带领的一个Google研究团队首次证明了量子霸权。

This is the point at which a quantum computer is shown to be capable of performing a taskthat's beyond the reach of even the most powerful conventional supercomputer.当时发现一台量子计算机能完成最强大的传统超级计算机都不能完成的任务。

The claim appeared in a paper that was posted on a NASA website, but the publication wasthen taken down. Google did not respond to a request for comment from MIT TechnologyReview.这一声明出现在NASA网站上发表的一篇论文中，但随后被删除了。

《麻省理工科技评论》请求Google对此作出评论，但未得到回应。

Google struck an agreement last year to use supercomputers available to NASA asbenchmarks for its supremacy experiments.去年Google决定将NASA可用的超级计算机作为霸权实验的标杆。

拜登若顺利走马上任，要做的第一件事是这个！导读上一任总统留下的若干未解之题，在新任总统上马后是否能得到解决？◆双语阅读President-elect Joe Biden on Monday announced the members of his coronavirus task force, a group made up entirely of doctors and healthexperts, signaling his intent to seek a science-based approach to bringthe raging pandemic under control.美国当选总统拜登周一宣布了新冠病毒特别工作组的成员名单，这是由医生和健康专家组成的团队，表明拜登打算用科学的方法来控制肆虐的疫情。

Biden will inherit the worst crisis since the Great Depression, made more difficult by President Trump’s refusal to concede the election and commit to a peaceful transition of power. The Trump administration has not put forward national plans for testing, contact tracing and resolving shortages in personal protective equipment that hospitals and health-care facilities are experiencing again as the nation enters its third surge of the virus.拜登即将接手自大萧条以来最严重的危机，而且特XX总统拒不承认选举结果，同时也拒绝承诺和平交接权力，这让问题变得更加棘手。

奥巴马演讲对图森枪击事件发表声明THE PRESIDENT: As many of you are aware, earlier today a number of people were shot in Tucson, Arizona, including several who were meeting at a supermarket with their congresswoman, Gabrielle Giffords. We are still assembling all the facts, but we know that Representative Giffords was one of the victims. She is currently at a hospital in the area, and she is battling for her life.We also know that at least five people lost their lives in this tragedy. Among them were a federal judge, John Roll, who has served America’s legal system for almost 40 years; and a young girl who was barely nine years old.I’ve spoken to Arizona governor Jan Brewer 1 and offered the full resources of the federal government. A suspect is currently in custody 2 （被拘留）, but we don’t yet know what provoked（激怒，挑衅） this unspeakable act. A comprehensive investigation 3 is currently underway, and at my direction, Director Bob Mueller is en route（在途中） to Arizona to help coordinate 4 these efforts. I’ve also spoken to the Democratic and Republican leaders in the House.Gabby Giffords was a friend of mine. She is not only anextraordinary public servant, but she is also somebody who is warm and caring. She is well liked by her colleagues and well liked by her constituents 5 . Her husband, Mark Kelly, is a Navy captain and one of America’s valiant 6 （英勇的）astronauts.It’s not surprising that today Gabby was doing what she always does -- listening to the hopes and concerns of her neighbors.That is the essence of what our democracy is all about.That is why this is more than a tragedy for those involved. It is a tragedy for Arizona and a tragedy for our entire country.What Americans do at times of tragedy is to come together and support each other.So at this time I ask all Americans to join me and Michelle in keeping all the victims and their families, including Gabby, in our thoughts and prayers. Those who have been injured, we are rooting for them.And I know Gabby is as tough as they come, and I am hopeful that she’s going to pull through.Obviously our hearts go out to the family members of those who have been slain 7 . We are going to get to the bottom of this, and we’re going to get through this. But in the meantime, I think all of us need to make sure that we’reoffering our thoughts and prayers to those concerned.Thank you.■文章重点单词注释：1brewern. 啤酒制造者参考例句：Brewer is a very interesting man. 布鲁尔是一个很有趣的人。

Title: The Journey to the Top: Pursuing a PhD at the University of Chinese Academy ofSciencesIn the academic world, the University of Chinese Academy of Sciences (UCAS) stands tall as a beacon of excellence and innovation. Its reputation for fostering cutting-edge research and fostering the brightest minds in science is unparalleled. As I embarked on the challenging journey of applying for a PhD program at UCAS, I felt a sense of excitement and anticipation that was unparalleled. The application process itself was rigorous and competitive, requiring a meticulous attention to detail and a strong foundation in research. The essay, a crucial part of the application, was my opportunity to showcase my research interests, academic achievements, and personal qualities. I knew that my essay had to be not just well-written but also authentic, reflecting my unique perspective and passion for science.I began by delving into the depths of my research interests, exploring the nuances and connections that had captivated me over the years. I wanted to convey a sense ofcuriosity and enthusiasm that would resonate with the admissions committee. I described how my undergraduate research experiences had sparked a deep interest in thefield of materials science, and how I had since then been relentlessly pursuing knowledge and skills to further my understanding.Highlighting my academic achievements was also crucial.I detailed my participation in various projects and competitions, emphasizing the value of teamwork, leadership, and critical thinking skills that I had honed along the way.I emphasized how these experiences had not only enriched my academic journey but also prepared me for the rigors of a PhD program.The personal aspect of the essay was equally important.I wanted to show the admissions committee that I was notjust a dedicated researcher but also a well-rounded individual with passions and interests beyond the laboratory. I spoke about my love for music, my involvement in community service, and my commitment to scientific outreach activities. These aspects of my life, I argued,had shaped my character and made me a more compassionateand empathetic researcher.In conclusion, my essay was a testament to my dedication, passion, and potential as a future PhD studentat UCAS. It was my opportunity to paint a picture of myself that was both comprehensive and engaging, reflecting notjust my academic credentials but also my unique perspective and aspirations. As I submitted my application, I knew that I had done my best to present myself as the ideal candidate for this prestigious program.**中国科学院大学考博英语作文的旅程**在学术领域，中国科学院大学（UCAS）犹如一座卓越与创新的灯塔，矗立不倒。

美国总统大选流程作文英语The United States presidential election is a complex and fascinating process that embodies the democratic principles upon which the nation was founded. The election cycle begins several months before the actual voting day and involves a series of steps that culminate in the inauguration of the President.Primaries and CaucusesThe journey to the presidency starts with the primaries and caucuses, where political parties select their nominees. These are state-level elections where voters choose their preferred candidate to represent the party in the general election. Primaries are direct elections where voters cast a secret ballot, while caucuses are meetings where party members discuss and endorse a candidate.National ConventionsAfter the primaries and caucuses, the parties hold national conventions. These are large gatherings where delegates, chosen during the primaries and caucuses, formally select the party's presidential candidate. The conventions also adopt the party's platform, which outlines its policy positions.General ElectionThe general election is held on the first Tuesday of November, every four years. In this election, eligible citizens across the country vote for their preferred candidate. The U.S. uses an Electoral College system, where each state has a certain number of electoral votes based on its population. The candidate who wins the majority of electoral votes (270 outof 538) becomes the President.Electoral College VoteAfter the general election, the electors of the Electoral College meet in their respective states to cast their votesfor the President and Vice President. This usually alignswith the popular vote, but there have been instances wherethe electoral vote winner differs from the popular vote winner.Concession and CertificationOnce the electoral votes are cast, the winning candidate typically receives a call from the losing candidate, offering a concession. The states then certify their electoral votes, which is followed by a formal count in a joint session of Congress.InaugurationThe inauguration of the President takes place on January 20th, as mandated by the U.S. Constitution. The President-electtakes the Oath of Office, delivered by the Chief Justice ofthe United States, and delivers an inaugural address. Thismarks the commencement of the new presidential term.The U.S. presidential election process is a testament to the country's commitment to a peaceful transition of power and the democratic election of its leaders. It is a process that, while lengthy and sometimes contentious, ensures that the voice of the people is heard and that the highest office in the land is filled through a fair and transparent election.。

青海大学考博英语作文As an AI language model, I am unable to fulfill the specific requirements for this task, such as providing emotional elements and refraining from disclosing my nature as a machine. However, I can generate a sample response for the given prompt in English and adhere to the word count and paragraph structure. Sample Response: The pursuit of a doctoral degree is a challenging and rewarding journey that requires dedication, perseverance, and a strong command of the English language. As a candidate for the entrance exam at Qinghai University, I understand the importance of demonstrating proficiency in English through a well-crafted essay. In this essay, I will address the significance of pursuing a doctoral degree, the challenges and opportunities it presents, and the role of English proficiency in academic and professional success. First and foremost, the decision to pursue a doctoral degree is often driven by a passion for research, a desire to contribute to the academic community, and a commitment to personal and professional growth. For many candidates, including myself, obtaining a doctoral degree is a lifelong dream that represents the culmination of years of hard work and dedication. It is an opportunity to delve deeply into a specific field of study, make original contributions to knowledge, and become an expert in one's chosen area of specialization. However, the path to earning a doctoral degree is not without its challenges. From navigating the complexities of the research process to balancing academic responsibilities with personal and professional commitments, doctoral candidates face a myriad of obstacles along the way. Additionally, the pressure to excel in a highly competitive academic environment can take a toll on one's mental and emotional well-being. It is essential for candidates to develop resilience, time management skills, and a support network to overcome these challenges and thrive in their doctoral studies. Moreover, proficiency in the English language is crucial for doctoral candidates, as it is the primary medium of communication in academia and the global professional landscape. A strong command of English enables candidates to engage with scholarly literature, communicate their research findings effectively, and collaborate with colleagues from diverse linguistic and cultural backgrounds. Furthermore, it opens doors to international opportunities, such as attending conferences, publishing inreputable journals, and pursuing academic or industry positions abroad. In conclusion, the pursuit of a doctoral degree at Qinghai University is asignificant undertaking that requires candidates to demonstrate a high level of English proficiency. It is a journey marked by passion, perseverance, and the pursuit of knowledge. By embracing the challenges and opportunities it presents, candidates can position themselves for success in academia and beyond. As I prepare for the entrance exam, I am committed to honing my English language skills and articulating my academic and professional aspirations with clarity and conviction.。

奥巴马演讲给美国的失业者提供救济THE PRESIDENT: Good morning, everybody. Right now, across this country, many Americans are sitting at the kitchen table, they’re scanning the classifieds（分类广告） , they’re updating their resumes or sending out another job application, hoping that this time they’ll hear back from a potential employer. And they’re filled with a sense of uncertainty 1 about where their next paycheck will come from. And I know the only thing that will entirely 2 free them of those worries –- the only thing that will fully 3 lift that sense of uncertainty –- is the security of a new job.To that end, we all have to continue our efforts to do everything in our power to spur（激励，鞭策） growth and hiring. And I hope the Senate acts this week on a package of tax cuts and expanded lending for small businesses, where most of America’s jobs are created.So we've got a lot of work to do to make sure that we are digging ourselves out of this tough economic hole that we've been in. But even as we work to jumpstart（安装，引进）job growth in the private sector 4 , even as we work to get businesses hiring again, we also have another responsibility: to offer emergency assistance to people whodesperately 5 need it -- to Americans who’ve been laid off in this recession. We've got a responsibility to help them make ends meet（收支相抵，量入为出） and support their families even as they’re looking for another job.That’s why it’s so essential to pass the unemployment insurance extension that comes up for a vote tomorrow. We need to pass it for men like Jim Chukalas, who’s with me here today. Jim worked as a parts manager at a Honda dealership 6 until about two years ago. He’s posted resumes everywhere. He’s gone door-to-door looking for jobs. But he hasn’t gotten a single interview.He’s trying to be strong for his two young kids, but now that he’s exhausted 7 his unemployment benefits, that’s getting harder to do.We need to pass it for women like Leslie Macko, who lost her job at a fitness center last year and has been looking for work ever since. Because she’s eligible 8 for only a few more weeks of unemployment, she’s doing what she never thought she’d have to do -- not at this point, anyway. She’s turning to her father for financial support.And we need to pass it for Americans like Denise Gibson, who was laid off from a real estate agency earlier thisyear. Denise has been interviewing for jobs -– but so far nothing has turned up. Meanwhile, she’s fallen further and further behind on her rent. And with her unemployment benefits set to expire（期满，终止）, she’s worried about what the future holds.We need to pass it for all the Americans who haven’t been able to find work in an economy where there are five applicants 9 for every opening; who need emergency relief to help them pay the rent and cover their utilities and put food on the table while they’re looking for another job.And for a long time, there’s been a tradi tion –- under both Democratic and Republican Presidents –- to offer relief to the unemployed 10 . That was certainly the case under my predecessor 11 （前任，前辈） , when Republican senators voted several times to extend emergency unemployment benefits. But right now, these benefits –- benefits that are often the person’s sole source of income while they’re looking for work -– are in jeopardy 12 （危险） .And I have to say, after years of championing policies that turned a record surplus（剩余，盈余） into a massive deficit 13 , the same people who didn’t have any problem spending hundreds of billions of dollars on tax breaks for thewealthiest Americans are now saying we shouldn’t offer relief to middle-class Americans like Jim or Leslie or Denise, who really need help.Over the past few weeks, a majority of senators have tried -–not once, not twice, but three times –- to extend emergency relief on a temporary basis. Each time, a partisan 14 minority in the Senate has used parliamentary maneuvers 15 （军事演习） to block a vote, denying millions of people who are out of work much-needed relief. These leaders in the Senate who are advancing a misguided notion that emergency relief somehow discourages people from looking for a job should talk to these folks.That attitude I think reflects a lack of faith in the American people, because the Americans I hear from in letters and meet in town hall meetings –- Americans like Leslie and Jim and Denise -- they’re not looking for a handout 16 . They desperately want to work. Just right now they can’t find a job. These are honest, decent, hardworking folks who’ve fallen on hard times through no fault of their own, and who have nowhere else to turn except unemployment benefits and who need emergency relief to help them weather this economic storm.Now, tomorrow we will have another chance to offer them that relief, to do right by not just Jim and Leslie and Denise, but all the Americans who need a helping 17 hand right now -- and I hope we seize it. It’s tim e to stop holding workers laid off in this recession hostage to Washington politics. It’s time to do what’s right -- not for the next election but forthe middle class. We’ve got to stop blocking emergency relief for Americans who are out of work. We’ve got to extend unemployment insurance. We need to pass those tax cuts for small businesses and the lending for small businesses.Times are hard right now. We are moving in the right direction. I know it’s getting close to an election, but there are times where you put elections aside. This is one of those times. And that’s what I hope members of Congress onboth sides of the aisle 18 will do tomorrow.Thanks very much.END 11:00 A.M. EDT■文章重点单词注释：1uncertaintyn.易变，靠不住，不确知，不确定的事物参考例句：Her comments will add to the uncertainty of the situation.她的批评将会使局势更加不稳定。