Joint Latency and Cost Optimization for Erasure-Coded Data Center Storage

Xiang, Yu; Lan, Tian; Aggarwal, Vaneet; Chen, Yih-Farn R.

doi:10.1109/tnet.2015.2466453

Cited by 83 publications

(117 citation statements)

References 40 publications

Supporting

Mentioning

115

Contrasting

Order By: Relevance

“…In the above applications, load-balancing policies (see, e.g., [19,20,31,37,40] are usually used for assigning tasks to servers. For scenarios where either low-overhead is desired or information accessibility is constrained (such as in a distributed setting), workload arXiv:1710.00296v3 [cs.PF] 16 Sep 2018 agnostic assignment policies [19,31,37] can be preferred. Our limited fork-join model assumes a random task assignment policy, which is suitable for such application scenarios.…”

Section: Motivationmentioning

confidence: 99%

Delay Asymptotics and Bounds for Multi-Task Parallel Jobs

Wang

Jiang

Scheller‐Wolf

et al. 2019

SIGMETRICS Perform. Eval. Rev.

View full text Add to dashboard Cite

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) ≤ 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-andbins problems where there is a negative correlation among bins, we prove that our variant exhibits positive correlation.

show abstract

Section: Motivationmentioning

confidence: 99%

Delay Asymptotics and Bounds for Multi-Task Parallel Jobs

Wang

Jiang

Scheller‐Wolf

et al. 2019

SIGMETRICS Perform. Eval. Rev.

View full text Add to dashboard Cite

show abstract

“…Since the problem of finding optimal assignment is challenging, we use a probabilistic scheduling approach where the VMs are assigned to jobs with certain probabilities which can be optimized for improved performance. Such scheduling approaches have been used in [16]. This approach results in the arrival distribution for the different types of jobs in the networking phase also being a Poisson process.…”

Section: Introductionmentioning

confidence: 99%

Joint Information Freshness and Completion Time Optimization for Vehicular Networks

Al-Abbasi

Aggarwal

2022

IEEE Trans. Serv. Comput.

Self Cite

View full text Add to dashboard Cite

The demand for real-time cloud applications has seen an unprecedented growth over the past decade. These applications require rapidly data transfer and fast computations. This paper considers a scenario where multiple IoT devices update information on the cloud, and request a computation from the cloud at certain times. The time required to complete the request for computation includes the time to wait for computation to start on busy virtual machines, performing the computation, waiting and service in the networking stage for delivering the output to the end user. In this context, the freshness of the information is an important concern and is different from the completion time. This paper proposes novel scheduling strategies for both computation and networking stages. Based on these strategies, the age-of-information (AoI) metric and the completion time are characterized. A convex combination of the two metrics is optimized over the scheduling parameters. The problem is shown to be convex and thus can be solved optimally. Moreover, based on the offline policy, an online algorithm for job scheduling is developed. Numerical results demonstrate significant improvement as compared to the considered baselines.

show abstract

“…By optimizing these probabilities, we quantify SDTP through a closed-form, tight upper bound for CDNbased video streaming with arbitrary cache content placement and network resource allocation. We note that the analysis in this paper is fundamentally different from those for distributed file storage, e.g., [7], [8], because the stall duration of a video relies on the download times of all its chunks, rather than simply the time to download the last chunk of a file. Further, since video chunks are downloaded and played sequentially, the download times and playback times of different video chunks are highly correlated and thus jointly determine the SDTP metric.…”

Section: Introductionmentioning

confidence: 99%

“…Mean latency and tail latency have been characterized in [7], [8] and [22], [23], respectively, for a system with multiple files using probabilistic scheduling. However, these papers consider only file downloading rather than video streaming.…”

Section: Introductionmentioning

confidence: 99%

FastTrack: Minimizing Stalls for CDN-Based Over-the-Top Video Streaming Systems

Al-Abbasi

Aggarwal

Lan

et al. 2021

IEEE Trans. Cloud Comput.

Self Cite

View full text Add to dashboard Cite

Traffic for internet video streaming has been rapidly increasing and is further expected to increase with the higher definition videos and IoT applications, such as 360 degree videos and augmented virtual reality applications. While efficient management of heterogeneous cloud resources to optimize the quality of experience is important, existing work in this problem space often left out important factors. In this paper, we present a model for describing a today's representative system architecture for video streaming applications, typically composed of a centralized origin server and several CDN sites. Our model comprehensively considers the following factors: limited caching spaces at the CDN sites, allocation of CDN for a video request, choice of different ports from the CDN, and the central storage and bandwidth allocation. With the model, we focus on minimizing a performance metric, stall duration tail probability (SDTP), and present a novel, yet efficient, algorithm to solve the formulated optimization problem. The theoretical bounds with respect to the SDTP metric are also analyzed and presented. Our extensive simulation results demonstrate that the proposed algorithms can significantly improve the SDTP metric, compared to the baseline strategies. Small-scale video streaming system implementation in a real cloud environment further validates our results.

show abstract

Joint Latency and Cost Optimization for Erasure-Coded Data Center Storage

Cited by 83 publications

References 40 publications

Delay Asymptotics and Bounds for Multi-Task Parallel Jobs

Delay Asymptotics and Bounds for Multi-Task Parallel Jobs

Joint Information Freshness and Completion Time Optimization for Vehicular Networks

FastTrack: Minimizing Stalls for CDN-Based Over-the-Top Video Streaming Systems

Contact Info

Product

Resources

About