arXiv subjects

Mor Harchol-Balter

Publications and source records attributed to Mor Harchol-Balter.

32 records · Page 2Linked to original sources

arXiv2020

heSRPT: Optimal Parallel Scheduling of Jobs With Known Sizes

When parallelizing a set of jobs across many servers, one must balance a trade-off between granting priority to short jobs and maintaining the overall efficiency of the system. When the goal is to minimize the mean flow time of a set of jobs, it is usually the case that one wants to complete short jobs before long jobs. However, since jobs usually cannot be parallelized with perfect efficiency, granting strict priority to the short jobs can result in very low system efficiency which in turn hurts the mean flow time across jobs. In this paper, we derive the optimal policy for allocating servers to jobs at every moment in time in order to minimize mean flow time across jobs. We assume that jobs follow a sublinear, concave speedup function, and hence jobs experience diminishing returns from being allocated additional servers. We show that the optimal policy, heSRPT, will complete jobs according to their size order, but maintains overall system efficiency by allocating some servers to each job at every moment in time. We compare heSRPT with state-of-the-art allocation policies from the literature and show that heSRPT outperforms its competitors by at least 30%, and often by much more.

cs.PF↗

arXiv2020

heSRPT: Parallel Scheduling to Minimize Mean Slowdown

Modern data centers serve workloads which are capable of exploiting parallelism. When a job parallelizes across multiple servers it will complete more quickly, but jobs receive diminishing returns from being allocated additional servers. Because allocating multiple servers to a single job is inefficient, it is unclear how best to allocate a fixed number of servers between many parallelizable jobs. This paper provides the first optimal allocation policy for minimizing the mean slowdown of parallelizable jobs of known size when all jobs are present at time 0. Our policy provides a simple closed form formula for the optimal allocations at every moment in time. Minimizing mean slowdown usually requires favoring short jobs over long ones (as in the SRPT policy). However, because parallelizable jobs have sublinear speedup functions, system efficiency is also an issue. System efficiency is maximized by giving equal allocations to all jobs and thus competes with the goal of prioritizing small jobs. Our optimal policy, high-efficiency SRPT (heSRPT), balances these competing goals. heSRPT completes jobs according to their size order, but maintains overall system efficiency by allocating some servers to each job at every moment in time. Our results generalize to also provide the optimal allocation policy with respect to mean flow time. Finally, we consider the online case where jobs arrive to the system over time. While optimizing mean slowdown in the online setting is even more difficult, we find that heSRPT provides an excellent heuristic policy for the online setting. In fact, our simulations show that heSRPT significantly outperforms state-of-the-art allocation policies for parallelizable jobs.

cs.DC↗

arXiv2020

Optimal Multiserver Scheduling with Unknown Job Sizes in Heavy Traffic

We consider scheduling to minimize mean response time of the M/G/k queue with unknown job sizes. In the single-server case, the optimal policy is the Gittins policy, but it is not known whether Gittins or any other policy is optimal in the multiserver case. Exactly analyzing the M/G/k under any scheduling policy is intractable, and Gittins is a particularly complicated policy that is hard to analyze even in the single-server case. In this work we introduce monotonic Gittins (M-Gittins), a new variation of the Gittins policy, and show that it minimizes mean response time in the heavy-traffic M/G/k for a wide class of finite-variance job size distributions. We also show that the monotonic shortest expected remaining processing time (M-SERPT) policy, which is simpler than M-Gittins, is a 2-approximation for mean response time in the heavy traffic M/G/k under similar conditions. These results constitute the most general optimality results to date for the M/G/k with unknown job sizes. Our techniques build upon work by Grosof et al., who study simple policies, such as SRPT, in the M/G/k; Bansal et al., Kamphorst and Zwart, and Lin et al., who analyze mean response time scaling of simple policies in the heavy-traffic M/G/1; and Aalto et al. and Scully et al., who characterize and analyze the Gittins policy in the M/G/1.

cs.PF↗

arXiv2020

Stability for Two-class Multiserver-job Systems

Multiserver-job systems, where jobs require concurrent service at many servers, occur widely in practice. Much is known in the dropping setting, where jobs are immediately discarded if they require more servers than are currently available. However, very little is known in the more practical setting where jobs queue instead. In this paper, we derive a closed-form analytical expression for the stability region of a two-class (non-dropping) multiserver-job system where each class of jobs requires a distinct number of servers and requires a distinct exponential distribution of service time, and jobs are served in first-come-first-served (FCFS) order. This is the first result of any kind for an FCFS multiserver-job system where the classes have distinct service distributions. Our work is based on a technique that leverages the idea of a "saturated" system, in which an unlimited number of jobs are always available. Our analytical formula provides insight into the behavior of FCFS multiserver-job systems, highlighting the huge wastage (idle servers while jobs are in the queue) that can occur, as well as the nonmonotonic effects of the service rates on wastage.

cs.PF↗

arXiv2020

Optimal Resource Allocation for Elastic and Inelastic Jobs

Modern data centers are tasked with processing heterogeneous workloads consisting of various classes of jobs. These classes differ in their arrival rates, size distributions, and job parallelizability. With respect to paralellizability, some jobs are elastic, meaning they can parallelize linearly across many servers. Other jobs are inelastic, meaning they can only run on a single server. Although job classes can differ drastically, they are typically forced to share a single cluster. When sharing a cluster among heterogeneous jobs, one must decide how to allocate servers to each job at every moment in time. In this paper, we design and analyze allocation policies which aim to minimize the mean response time across jobs, where a job's response time is the time from when it arrives until it completes. We model this problem in a stochastic setting where each job may be elastic or inelastic. Job sizes are drawn from exponential distributions, but are unknown to the system. We show that, in the common case where elastic jobs are larger on average than inelastic jobs, the optimal allocation policy is Inelastic-First, giving inelastic jobs preemptive priority over elastic jobs. We obtain this result by introducing a novel sample path argument. We also show that there exist cases where Elastic-First (giving priority to elastic jobs) performs better than Inelastic-First. We then provide the first analysis of mean response time under both Elastic-First and Inelastic-First by leveraging recent techniques for solving high-dimensional Markov chains.

cs.PF↗

arXiv2020

Simple Near-Optimal Scheduling for the M/G/1

We consider the problem of preemptively scheduling jobs to minimize mean response time of an M/G/1 queue. When we know each job's size, the shortest remaining processing time (SRPT) policy is optimal. Unfortunately, in many settings we do not have access to each job's size. Instead, we know only the job size distribution. In this setting the Gittins policy is known to minimize mean response time, but its complex priority structure can be computationally intractable. A much simpler alternative to Gittins is the shortest expected remaining processing time (SERPT) policy. While SERPT is a natural extension of SRPT to unknown job sizes, it is unknown whether or not SERPT is close to optimal for mean response time. We present a new variant of SERPT called monotonic SERPT (M-SERPT) which is as simple as SERPT but has provably near-optimal mean response time at all loads for any job size distribution. Specifically, we prove the mean response time ratio between M-SERPT and Gittins is at most 3 for load $ρ\leq 8/9$ and at most 5 for any load. This makes M-SERPT the only non-Gittins scheduling policy known to have a constant-factor approximation ratio for mean response time.

cs.PF↗

arXiv2019

Load Balancing Guardrails: Keeping Your Heavy Traffic on the Road to Low Response Times

Load balancing systems, comprising a central dispatcher and a scheduling policy at each server, are widely used in practice, and their response time has been extensively studied in the theoretical literature. While much is known about the scenario where the scheduling at the servers is First-Come-First-Served (FCFS), to minimize mean response time we must use Shortest-Remaining-Processing-Time (SRPT) scheduling at the servers. Much less is known about dispatching polices when SRPT scheduling is used. Unfortunately, traditional dispatching policies that are used in practice in systems with FCFS servers often have poor performance in systems with SRPT servers. In this paper, we devise a simple fix that can be applied to any dispatching policy. This fix, called guardrails, ensures that the dispatching policy yields optimal mean response time under heavy traffic when used in a system with SRPT servers. Any dispatching policy, when augmented with guardrails, becomes heavy-traffic optimal. Our results yield the first analytical bounds on mean response time for load balancing systems with SRPT scheduling at the servers.

cs.PF↗

arXiv2018

Optimal Scheduling and Exact Response Time Analysis for Multistage Jobs

Scheduling to minimize mean response time in an M/G/1 queue is a classic problem. The problem is usually addressed in one of two scenarios. In the perfect-information scenario, the scheduler knows each job's exact size, or service requirement. In the zero-information scenario, the scheduler knows only each job's size distribution. The well-known shortest remaining processing time (SRPT) policy is optimal in the perfect-information scenario, and the more complex Gittins policy is optimal in the zero-information scenario. In real systems the scheduler often has partial but incomplete information about each job's size. We introduce a new job model, that of multistage jobs, to capture this partial-information scenario. A multistage job consists of a sequence of stages, where both the sequence of stages and stage sizes are unknown, but the scheduler always knows which stage of a job is in progress. We give an optimal algorithm for scheduling multistage jobs in an M/G/1 queue and an exact response time analysis of our algorithm.

cs.PF↗

arXiv2018

Delay Asymptotics and Bounds for Multi-Task Parallel Jobs

We study delay of jobs that consist of multiple parallel tasks, which is a critical performance metric in a wide range of applications such as data file retrieval in coded storage systems and parallel computing. In this problem, each job is completed only when all of its tasks are completed, so the delay of a job is the maximum of the delays of its tasks. Despite the wide attention this problem has received, tight analysis is still largely unknown since analyzing job delay requires characterizing the complicated correlation among task delays, which is hard to do. We first consider an asymptotic regime where the number of servers, $n$, goes to infinity, and the number of tasks in a job, $k^{(n)}$, is allowed to increase with $n$. We establish the asymptotic independence of any $k^{(n)}$ queues under the condition $k^{(n)} = o(n^{1/4})$. This greatly generalizes the asymptotic-independence type of results in the literature where asymptotic independence is shown only for a fixed constant number of queues. As a consequence of our independence result, the job delay converges to the maximum of independent task delays. We next consider the non-asymptotic regime. Here we prove that independence yields a stochastic upper bound on job delay for any $n$ and any $k^{(n)}$ with $k^{(n)}\le n$. The key component of our proof is a new technique we develop, called "Poisson oversampling". Our approach converts the job delay problem into a corresponding balls-and-bins problem. However, in contrast with typical balls-and-bins problems where there is a negative correlation among bins, we prove that our variant exhibits positive correlation.

cs.PF↗

arXiv2018

Practical Bounds on Optimal Caching with Variable Object Sizes

Many recent caching systems aim to improve miss ratios, but there is no good sense among practitioners of how much further miss ratios can be improved. In other words, should the systems community continue working on this problem? Currently, there is no principled answer to this question. In practice, object sizes often vary by several orders of magnitude, where computing the optimal miss ratio (OPT) is known to be NP-hard. The few known results on caching with variable object sizes provide very weak bounds and are impractical to compute on traces of realistic length. We propose a new method to compute upper and lower bounds on OPT. Our key insight is to represent caching as a min-cost flow problem, hence we call our method the flow-based offline optimal (FOO). We prove that, under simple independence assumptions, FOO's bounds become tight as the number of objects goes to infinity. Indeed, FOO's error over 10M requests of production CDN and storage traces is negligible: at most 0.3%. FOO thus reveals, for the first time, the limits of caching with variable object sizes. While FOO is very accurate, it is computationally impractical on traces with hundreds of millions of requests. We therefore extend FOO to obtain more efficient bounds on OPT, which we call practical flow-based offline optimal (PFOO). We evaluate PFOO on several full production traces and use it to compare OPT to prior online policies. This analysis shows that current caching systems are in fact still far from optimal, suffering 11-43% more cache misses than OPT, whereas the best prior offline bounds suggest that there is essentially no room for improvement.

cs.PF↗

arXiv2018

SRPT for Multiserver Systems

The Shortest Remaining Processing Time (SRPT) scheduling policy and its variants have been extensively studied in both theoretical and practical settings. While beautiful results are known for single-server SRPT, much less is known for multiserver SRPT. In particular, stochastic analysis of the M/G/k under multiserver SRPT is entirely open. Intuition suggests that multiserver SRPT should be optimal or near-optimal for minimizing mean response time. However, the only known analysis of multiserver SRPT is in the worst-case adversarial setting, where SRPT can be far from optimal. In this paper, we give the first stochastic analysis bounding mean response time of the M/G/k under multiserver SRPT. Using our response time bound, we show that multiserver SRPT has asymptotically optimal mean response time in the heavy-traffic limit. The key to our bounds is a strategic combination of stochastic and worst-case techniques. Beyond SRPT, we prove similar response time bounds and optimality results for several other multiserver scheduling policies.

cs.PF↗

arXiv2018

SOAP: One Clean Analysis of All Age-Based Scheduling Policies

We consider an extremely broad class of M/G/1 scheduling policies called SOAP: Schedule Ordered by Age-based Priority. The SOAP policies include almost all scheduling policies in the literature as well as an infinite number of variants which have never been analyzed, or maybe not even conceived. SOAP policies range from classic policies, like first-come, first-serve (FCFS), foreground-background (FB), class-based priority, and shortest remaining processing time (SRPT); to much more complicated scheduling rules, such as the famously complex Gittins index policy and other policies in which a job's priority changes arbitrarily with its age. While the response time of policies in the former category is well understood, policies in the latter category have resisted response time analysis. We present a universal analysis of all SOAP policies, deriving the mean and Laplace-Stieltjes transform of response time.

cs.PF↗

arXiv2017

Towards Optimality in Parallel Scheduling

To keep pace with Moore's law, chip designers have focused on increasing the number of cores per chip rather than single core performance. In turn, modern jobs are often designed to run on any number of cores. However, to effectively leverage these multi-core chips, one must address the question of how many cores to assign to each job. Given that jobs receive sublinear speedups from additional cores, there is an obvious tradeoff: allocating more cores to an individual job reduces the job's runtime, but in turn decreases the efficiency of the overall system. We ask how the system should schedule jobs across cores so as to minimize the mean response time over a stream of incoming jobs. To answer this question, we develop an analytical model of jobs running on a multi-core machine. We prove that EQUI, a policy which continuously divides cores evenly across jobs, is optimal when all jobs follow a single speedup curve and have exponentially distributed sizes. EQUI requires jobs to change their level of parallelization while they run. Since this is not possible for all workloads, we consider a class of "fixed-width" policies, which choose a single level of parallelization, k, to use for all jobs. We prove that, surprisingly, it is possible to achieve EQUI's performance without requiring jobs to change their levels of parallelization by using the optimal fixed level of parallelization, k*. We also show how to analytically derive the optimal k* as a function of the system load, the speedup curve, and the job size distribution. In the case where jobs may follow different speedup curves, finding a good scheduling policy is even more challenging. We find that policies like EQUI which performed well in the case of a single speedup function now perform poorly. We propose a very simple policy, GREEDY*, which performs near-optimally when compared to the numerically-derived optimal policy.

cs.PF↗

arXiv2015

Clearing Analysis on Phases: Exact Limiting Probabilities for Skip-free, Unidirectional, Quasi-birth-death Processes

Many problems in computing, service, and manufacturing systems can be modeled via infinite repeating Markov chains with an infinite number of levels and a finite number of phases. Many such chains are quasi-birth-death processes (QBDs) with transitions that are skip-free in level, in that one can only transition between consecutive levels, and unidirectional in phase, in that one can only transition from lower-numbered phases to higher-numbered phases. We present a procedure, which we call Clearing Analysis on Phases (CAP), for determining the limiting probabilities of such Markov chains exactly. The CAP method yields the limiting probability of each state in the repeating portion of the chain as a linear combination of scalar bases raised to a power corresponding to the level of the state. The weights in these linear combinations can be determined by solving a finite system of linear equations.

math.PR↗