Performance Analysis of Workload Dependent Load Balancing Policies

Hellemans, Tim; Bodas, Tejas; Houdt, Benny Van

doi:10.1145/3309697.3331504

Cited by 20 publications

(34 citation statements)

References 17 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Furthermore, in [10] the authors develop policies that result in fair delays for multiple classes of jobs, under replication constraints. An even more refined model for the server slowdowns, similar to the model that we introduce in Section 4, is introduced [14]. There, the authors use a fluid model to analyse a family of policies based on the celebrated power-of-𝑑-choices [24,34].…”

Section: Previous Workmentioning

confidence: 99%

Delay-optimal Policies in Partial Fork-Join Systems with Redundancy and Random Slowdowns

ZubeldiaMartin

2020

Proc. ACM Meas. Anal. Comput. Syst.

View full text Add to dashboard Cite

We consider a large distributed service system consisting of n homogeneous servers with infinite capacity FIFO queues. Jobs arrive as a Poisson process of rate λn/k_n (for some positive constant λ and integer k_n). Each incoming job consists of k_n identical tasks that can be executed in parallel, and that can be encoded into at least k_n "replicas" of the same size (by introducing redundancy) so that the job is considered to be completed when any k_n replicas associated with it finish their service. Moreover, we assume that servers can experience random slowdowns in their processing rate so that the service time of a replica is the product of its size and a random slowdown. First, we assume that the server slowdowns are shifted exponential and independent of the replica sizes. In this setting we show that the delay of a typical job is asymptotically minimized (as $n\to\infty$) when the number of replicas per task is a constant that only depends on the arrival rate λ, and on the expected slowdown of servers. Second, we introduce a new model for the server slowdowns in which larger tasks experience less variable slowdowns than smaller tasks. In this setting we show that, under the class of policies where all replicas start their service at the same time, the delay of a typical job is asymptotically minimized (as n\to\infty) when the number of replicas per task is made to depend on the actual size of the tasks being replicated, with smaller tasks being replicated more than larger tasks.

show abstract

Section: Previous Workmentioning

confidence: 99%

Delay-optimal Policies in Partial Fork-Join Systems with Redundancy and Random Slowdowns

ZubeldiaMartin

2020

Proc. ACM Meas. Anal. Comput. Syst.

View full text Add to dashboard Cite

show abstract

“…The PTC policy is more advanced but recently has been applied to the cluster management practice [27]; it can achieve a higher utilization of servers [24]. More discussion on dispatching policies can be found in [14,15,18,25]. Once the job j is dispatched to a server, the server will be occupied by the on-demand job owner during the period [a j , d j ], i.e., from the beginning of slot a j until the end of slot d j .…”

Section: Dispatching High Priority Of On-demand Jobsmentioning

confidence: 99%

A Framework for Allocating Server Time to Spot and On-Demand Services in Cloud Computing

Pellegrini

Gao

et al. 2019

ACM Trans. Model. Perform. Eval. Comput. Syst.

View full text Add to dashboard Cite

Cloud computing delivers value to users by facilitating their access to servers at any time period needed. An approach is to provide both on-demand and spot services on shared servers. The former allows users to access servers on demand at a fixed price and users occupy different time periods on servers. The latter allows users to bid for the remaining unoccupied time periods via dynamic pricing; however, without appropriate design, such time periods may be arbitrarily short since on-demand users arrive randomly. This is also the current service model adopted by Amazon Elastic Cloud Compute. In this article, we provide the first integral framework for sharing time on servers between on-demand and spot services while optimally pricing spot service. It guarantees that on-demand users can get served quickly while spot users can stably use servers for a properly long period once accepted, which is a key feature in making both on-demand and spot services accessible. Simulation results show that, by complementing the on-demand market with a spot market, a cloud provider can improve revenue by up to 461.5%. The framework is designed under assumptions that are met in real environments. It is a new tool that other cloud operators can use to quantify the advantage of a hybrid spot and on-demand service, making the case for eventually integrating this service model into their own infrastructures.

show abstract

“…[11,15]). The second category consists of workload dependent load balancing policies, for these policies the dispatcher balances the load on the servers by employing information on the amount of work that is left on some of the servers (see also [4]). This can be done explicitly if we assume the amount of work on servers is known or implicitly by employing some form of redundancy such as e.g.…”

Section: Introductionmentioning

confidence: 99%

“…In order to compute performance metrics such as the mean waiting time, the waiting time distribution, etc. most work relies on mean-field models [1,4,7,8,13]. Mean field models capture the limiting stationary behavior of the system as the number of servers tends to infinity provided that any finite set of servers becomes independent and identically distributed.…”

Section: Introductionmentioning

confidence: 99%

Mean Waiting Time in Large-Scale and Critically Loaded Power of d Load Balancing Systems

Hellemans

Houdt

2021

Abstract Proceedings of the 2021 ACM SIGMETRICS / International Conference on Measurement and Modeling of Computer Systems

Self Cite

View full text Add to dashboard Cite

Mean field models are a popular tool used to analyse load balancing policies. In some exceptional cases the waiting time distribution of the mean field limit has an explicit form. In other cases it can be computed as the solution of a differential equation or a recursive relation. In our paper [6] we study the limit of the mean waiting time E[W λ ] as the arrival rate λ approaches 1 for a number of load balancing policies when job sizes are exponential with mean 1 (i.e. when the system gets close to instability). As E[W λ ] diverges to infinity, we scale with − log(1 − λ) and present a method to compute the limit lim λ→1 − −E[W λ ]/log(1 − λ). We show that this limit has a surprisingly simple form for the load balancing algorithms considered.In addition, we propose an alternate scaling − log(p λ ) instead of − log(1 − λ), where p λ is adapted to the policy at hand. This scaling allows us to obtain good approximations of the mean waiting time when λ is close to one or zero. CCS CONCEPTS• Mathematics of computing → Probability and statistics; • Networks → Network performance modeling.

show abstract

Performance Analysis of Workload Dependent Load Balancing Policies

Cited by 20 publications

References 17 publications

Delay-optimal Policies in Partial Fork-Join Systems with Redundancy and Random Slowdowns

Delay-optimal Policies in Partial Fork-Join Systems with Redundancy and Random Slowdowns

A Framework for Allocating Server Time to Spot and On-Demand Services in Cloud Computing

Mean Waiting Time in Large-Scale and Critically Loaded Power of d Load Balancing Systems

Contact Info

Product

Resources

About