Abstracts of the 2019 SIGMETRICS/Performance Joint International Conference on Measurement and Modeling of Computer Systems最新文献

{"title":"Optimizing the Cost of Executing Mixed Interactive and Batch Workloads on Transient VMs","authors":"Pradeep Ambati, David E. Irwin","doi":"10.1145/3309697.3331489","DOIUrl":"https://doi.org/10.1145/3309697.3331489","url":null,"abstract":"Container Orchestration Platforms (COPs), such as Kubernetes, are increasingly used to manage large-scale clusters by automating resource allocation between applications encapsulated in containers. Increasingly, the resources underlying COPs are virtual machines (VMs) dynamically acquired from cloud platforms. COPs may choose from many different types of VMs offered by cloud platforms, which differ in their cost, performance, and availability. While transient VMs cost significantly less than on-demand VMs, platforms may revoke them at any time, causing them to become unavailable. While transient VMs' price is attractive, their unreliability is a problem for COPs designed to support mixed workloads composed of, not only delay-tolerant batch jobs, but also long-lived interactive services with high availability requirements. To address the problem, we design TR-Kubernetes, a COP that optimizes the cost of executing mixed interactive and batch workloads on cloud platforms using transient VMs. To do so, TR-Kubernetes enforces arbitrary availability requirements specified by interactive services despite transient VM unavailability by acquiring many more transient VMs than necessary most of the time, which it then leverages to opportunistically execute batch jobs when excess resources are available. When cloud platforms revoke transient VMs, TR-Kubernetes relies on existing Kubernetes functions to internally revoke resources from batch jobs to maintain interactive services' availability requirements. We show that TR-Kubernetes requires minimal extensions to Kubernetes, and is capable of lowering the cost (by 53%) and improving the availability (99.999%) of a representative interactive/batch workload on Amazon EC2 when using transient compared to on-demand VMs.","PeriodicalId":268462,"journal":{"name":"Abstracts of the 2019 SIGMETRICS/Performance Joint International Conference on Measurement and Modeling of Computer Systems","volume":"4 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-06-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123736715","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Architecture-Aware Approximate Computing","authors":"M. Karakoy, Orhan Kislal, Xulong Tang, M. Kandemir, Meenakshi Arunachalam","doi":"10.1145/3309697.3331508","DOIUrl":"https://doi.org/10.1145/3309697.3331508","url":null,"abstract":"Observing that many application programs from different domains can live with less-than-perfect accuracy, existing techniques try to trade off program output accuracy with performance-energy savings. While these works provide point solutions, they leave three critical questions regarding approximate computing unanswered: (i) what is the maximum potential of skipping (i.e., not performing) data accesses under a given inaccuracy bound?; (ii) can we identify the data accesses to drop randomly, or is being architecture aware critical?; and (iii) do two executions that skip the same number of data accesses always result in the same output quality (error)? This paper first provides answers to these questions using ten multithreaded workloads, and then presents a program slicing-based approach that identifies the set of data accesses to drop. Results indicate 8.8% performance improvement and 13.7% energy saving are possible when we set the error bound to 2%, and the corresponding improvements jump to 15% and 25%, respectively, when the error bound is raised to 4%.","PeriodicalId":268462,"journal":{"name":"Abstracts of the 2019 SIGMETRICS/Performance Joint International Conference on Measurement and Modeling of Computer Systems","volume":"4 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-06-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116489253","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A Structural Result for Personalized PageRank and its Algorithmic Consequences","authors":"Daniel Vial, V. Subramanian","doi":"10.1145/3309697.3331481","DOIUrl":"https://doi.org/10.1145/3309697.3331481","url":null,"abstract":"Many systems, such as the Internet, social networks, and the power grid, can be represented as graphs. When analyzing graphs, it is often useful to compute scores describing the relative importance or distance between nodes. One example is Personalized PageRank (PPR), which assigns to each node v a vector whose i-th entry describes the importance of the i-th node from the perspective of v. PPR has proven useful in many applications, such as recommending who users should follow on social networks (if this i-th entry is large, v may be interested in following the i-th user). Unfortunately, computing n PPR vectors exactly for a graph of n nodes has complexity O(n3), which is infeasible for many graphs of interest. In this work, we devise a scheme to estimate all n PPR vectors with bounded l_1 error and complexity O(nc), where c<2 depends on the degrees of the graph at hand, the desired error tolerance, and a parameter that defines PPR. This improves upon existing methods, the best of which have complexity O(n2 log n) in our setting. Our complexity guarantee holds with high probability, for certain choices of the PPR parameter, and for a certain class of random graphs (roughly speaking, the sparse directed configuration model with heavy-tailed in-degrees); our accuracy guarantee holds with probability 1 and for arbitrary graphs and PPR parameters. The complexity result arises as a consequence of our main (structural) result, which shows that the dimensionality of the set of PPR vectors scales sublinearly in n with high probability, for the same class of random graphs and for a notion of dimensionality similar to matrix rank. It is this coupling of the PPR vectors for the nodes on a common underlying graph that allows for estimating them faster. Hence, at a high level, our scheme is analogous to (but distinct from) low-rank matrix approximation. We also note that our scheme is similar to one that was proposed in [Jeh and Widom 2003] but lacked accuracy and complexity guarantees, so another contribution of our paper is to address this gap in the literature.","PeriodicalId":268462,"journal":{"name":"Abstracts of the 2019 SIGMETRICS/Performance Joint International Conference on Measurement and Modeling of Computer Systems","volume":"28 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-06-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124592061","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A Distributed Algorithm to Calculate Max-Min Fair Rates Without Per-Flow State","authors":"Lavanya Jose, Stephen Ibanez, Mohammad Alizadeh, N. McKeown","doi":"10.1145/3309697.3331472","DOIUrl":"https://doi.org/10.1145/3309697.3331472","url":null,"abstract":"Most congestion control algorithms, like TCP, rely on a reactive control system that detects congestion, then marches carefully towards a desired operating point (e.g. by modifying the window size or adjusting a rate). In an effort to balance stability and convergence speed, they often take hundreds of RTTs to converge; an increasing problem as networks get faster, with less time to react. This paper is about an alternative class of congestion control algorithms based on proactive-scheduling: switches and NICs \"pro-actively\" exchange control messages to run a distributed algorithm to pick \"explicit rates\" for each flow. We call these Proactive Explicit Rate Control (PERC) algorithms. They take as input the routing matrix and link speeds, but not a congestion signal. By exploiting information such as the number of flows at a link, they can converge an order of magnitude faster than reactive algorithms. Our main contributions are (1) s-PERC (\"stateless\" PERC), a new practical distributed PERC algorithm without per-flow state at the switches, and (2) a proof that s-PERC computes exact max-min fair rates in a known bounded time, the first such algorithm to do so without per-flow state. To analyze s-PERC, we introduce a parallel variant of standard waterfilling, 2-Waterfilling. We prove that s-PERC converges to max-min fair in 6N rounds, where N is the number of iterations 2-Waterfilling takes for the same routing matrix. We describe how to make s-PERC practical and robust to deploy in real networks. We confirm using realistic simulations and an FPGA hardware testbed that s-PERC converges 10-100x faster than reactive algorithms like TCP, DCTCP and RCP in data-center networks and 1.3--6x faster in wide-area networks (WANs). Long flows complete in close to the ideal time, while short-lived flows are prioritized, making it appropriate for data-centers and WANs.","PeriodicalId":268462,"journal":{"name":"Abstracts of the 2019 SIGMETRICS/Performance Joint International Conference on Measurement and Modeling of Computer Systems","volume":"41 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-06-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133411835","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"HyperBench: A Benchmark Suite for Virtualization Capabilities","authors":"Song Wei, Kun Zhang, Bibo Tu","doi":"10.1145/3309697.3331478","DOIUrl":"https://doi.org/10.1145/3309697.3331478","url":null,"abstract":"Virtualization is becoming increasingly common in data centers due to its various advantages. However, how to choose among different platforms, including both software and hardware, is a considerable challenge. In this context, evaluating the virtualization capabilities of different platforms is critically important. Regrettably, the existing benchmarks are not qualified for meeting this requirement. Different hardware mechanisms and hypervisor designs introduce many different hypervisor-level events, such as transitions between VMs and the hypervisor, two-dimensional page walk, and binary translation. These events are key factors affecting virtualization performance. Existing benchmarks either overlook these changes or are tightly coupled to a particular hypervisor. In this paper, we present HyperBench, a benchmark suite that focuses on the capabilities of different virtualization platforms. Currently, we design 15 hypervisor benchmarks covering CPU, memory, and I/O. The virtualization-sensitive operation in each benchmark triggers hypervisor-level events, which examines the platform's ability in the target area. HyperBench is designed as a custom kernel which can adapt to different hypervisors and architectures. What's more, adding a new benchmark is pretty easy. Finally, we perform a series of experiments on the host machine and several popular hypervisors, such as QEMU, KVM, and Xen, demonstrating that HyperBench is capable of revealing the performance implications of the hardware mechanism and hypervisor design.","PeriodicalId":268462,"journal":{"name":"Abstracts of the 2019 SIGMETRICS/Performance Joint International Conference on Measurement and Modeling of Computer Systems","volume":"28 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-06-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116619303","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Information, Memory and Capacity in Dynamic Resource Allocation","authors":"Kuang Xu, Y. Zhong","doi":"10.1145/3309697.3331488","DOIUrl":"https://doi.org/10.1145/3309697.3331488","url":null,"abstract":"We propose a general framework, dubbed Stochastic Processing under Imperfect Information (SPII), to study the impact of information constraints and memories on dynamic re- source allocation. The framework involves a Stochastic Processing Network (SPN) scheduling problem in which the decision maker may access the system state only through a noisy channel, and resource allocation decisions must be carried out through the interaction between an encoding policy (who observes the state) and allocation policy (who chooses the allocation). Applications in the management of large-scale data centers and human-in-the-loop service systems are among our chief motivations. We quantify the degree to which information constraints reduce the size of the capacity region in general SPNs, and how such reduction depends on the amount of memories available to the encoding and allocation policies. Using a novel metric, capacity factor, our main theorem characterizes the reduction in capacity region (under \"optimal\" policies) for all non-degenerate channels, and across almost all combinations of memory sizes. Notably, the theorem demonstrates, in substantial generality, that (1) the presence of a noisy channel always reduces capacity, (2) more memories for the allocation policy always improve capacity, and (3) more memories for the encoding policy have little to no effect on capacity. Finally, all of our positive (achievability) results are established through constructive, implementable policies. Our proof program involves the development of a host of new techniques by combining ideas from information theory, learning and queueing theory. We create a simple yet powerful generalization of the Max-Weight policy, in which individual Markov chains are selected dynamically, in a manner analogous to how schedules are used in a conventional Max-Weight policy.","PeriodicalId":268462,"journal":{"name":"Abstracts of the 2019 SIGMETRICS/Performance Joint International Conference on Measurement and Modeling of Computer Systems","volume":"19 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-06-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125877557","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A TTL-based Approach for Data Aggregation in Geo-distributed Streaming Analytics","authors":"Dhruv Kumar, Jian Li, A. Chandra, R. Sitaraman","doi":"10.1145/3309697.3331491","DOIUrl":"https://doi.org/10.1145/3309697.3331491","url":null,"abstract":"Streaming data analytics has been an important topic of research in recent years. Large quantities of data are generated continuously over time across a variety of application domains such as web and social analytics, scientific computing and energy analytics. One of the key requirements in modern data analytics services is the real-time analysis of these data streams to extract useful and timely information for the analyst. Several distributed data analytics platforms have been developed in recent times to meet this growing requirement of real-time streaming analytics. Nowadays, a large amount of data is generated continuously by geographically distributed sources (e.g., agents, sensors, mobile devices, edge nodes, etc.) in many streaming applications. For instance, services like Facebook, Twitter and Netflix continuously gather data from the end users for a variety of analytical purposes such as finding the popular web content amongst their users or monitoring the QoS metrics. Large content delivery networks (CDNs) like Akamai that serve a significant fraction of content on the Internet continuously collect data from their edge servers and clients from around the globe to understand what, where and how content is accessed for the purpose of providing content analytics insights to businesses.","PeriodicalId":268462,"journal":{"name":"Abstracts of the 2019 SIGMETRICS/Performance Joint International Conference on Measurement and Modeling of Computer Systems","volume":"89 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-06-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126998393","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Retroactive Packet Sampling for Traffic Receipts","authors":"Pavlos Nikolopoulos, C. Pappas, K. Argyraki, A. Perrig","doi":"10.1145/3309697.3331485","DOIUrl":"https://doi.org/10.1145/3309697.3331485","url":null,"abstract":"Is it possible to design a packet-sampling algorithm that prevents the network node that performs the sampling from treating the sampled packets preferentially? We study this problem in the context of designing a \"network-transparency'' system. In this system, networks emit receipts for a small sample of the packets they observe, and a monitor collects these receipts to estimate each network's loss and delay performance. Sampling is a good building block for this system, because it enables a solution that is flexible and combines low resource cost with quantifiable accuracy. The challenge is cheating resistance: when a network's performance is assessed based on the conditions experienced by a small traffic sample, the network has a strong incentive to treat the sampled packets better than the rest. We contribute a sampling algorithm that is provably robust to such prioritization attacks, enables network performance estimation with quantifiable accuracy, and requires minimal resources. We confirm our analysis using real traffic traces.","PeriodicalId":268462,"journal":{"name":"Abstracts of the 2019 SIGMETRICS/Performance Joint International Conference on Measurement and Modeling of Computer Systems","volume":"35 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-06-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122057195","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"The Segmentation-Thickness Tradeoff in Online Marketplaces","authors":"Reza Alijan, Siddhartha Banerjee, Sreenivas Gollapud, Kostas Kollias, Kamesh Munagala","doi":"10.1145/3309697.3331519","DOIUrl":"https://doi.org/10.1145/3309697.3331519","url":null,"abstract":"A core tension in the operations of online marketplaces is between segmentation (wherein platforms can increase revenue by segmenting the market into ever smaller sub-markets) and thickness (wherein the size of the sub-market affects the utility experienced by an agent). An important example of this is in dynamic online marketplaces, where buyers and sellers, in addition to preferences for different matches, also have finite patience (or deadlines) for being matched. We formalize this trade-off via a novel optimization problem that we term as 'Two-sided Facility Location': we consider a market wherein agents arrive at nodes embedded in an underlying metric space, where the distance between a buyer and seller captures the quality of the corresponding match. The platform posts prices and wages at the nodes, and opens a set of virtual clearinghouses where agents are routed for matching. To ensure high match-quality, the platform imposes a distance constraint between an agent and its clearinghouse; to ensure thickness, the platform requires the flow to any clearinghouse be at least a pre-specified lower bound. Subject to these constraints, the goal of the platform is to maximize the social surplus subject to weak budget balance, i.e., profit being non-negative. Our work characterizes the complexity of this problem by providing both hardness results as well as algorithms for this setting; in particular, we present an algorithm that for any constant ε > 0 yields a (1 + ε) approximation for the gains from trade, while relaxing the match quality (i.e., maximum distance of any match) by a constant factor.","PeriodicalId":268462,"journal":{"name":"Abstracts of the 2019 SIGMETRICS/Performance Joint International Conference on Measurement and Modeling of Computer Systems","volume":"118 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-06-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116368627","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Proactive Caching for Low Access-Delay Services under Uncertain Predictions","authors":"Ran Liu, E. Yeh, A. Eryilmaz","doi":"10.1145/3309697.3331471","DOIUrl":"https://doi.org/10.1145/3309697.3331471","url":null,"abstract":"Network traffic for delay-sensitive services has become a dominant part in the network. Proactive caching with the aid of predictive information has been proposed as a promising method to enhance delay performance. In this paper, we analytically investigate the problem of how to efficiently utilize uncertain predictive information to design proactive caching strategies with provably good access-delay characteristics. We first derive an upper bound for the average amount of proactive service per request that the system can support. We then analyze the behavior of a family of threshold-based proactive strategies using an innovative Markov chain analysis, and show that the average amount of proactive service per request can be maximized by properly selecting the threshold. Finally, we propose the UNIFORM strategy, which is the threshold-based strategy with the optimal threshold. Surprisingly, we show that it outperforms the commonly used Earliest-Deadline-First (EDF) type proactive strategies in terms of delay. We perform extensive numerical experiments to demonstrate the influence of thresholds on delay performance, and explicitly compare performance of the EDF strategy and the UNIFORM strategy.","PeriodicalId":268462,"journal":{"name":"Abstracts of the 2019 SIGMETRICS/Performance Joint International Conference on Measurement and Modeling of Computer Systems","volume":"82 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-06-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126388025","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}