Evaluating the Effectiveness of Replication for Tail-Tolerance

Qiu, Zhan; Pérez, Juan F.

doi:10.1109/ccgrid.2015.22

Cited by 12 publications

(15 citation statements)

References 17 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In these systems it is not feasible to cancel all replicas of a request upon completion of the first one, which is a mechanism commonly used to limit the additional load introduced by replication. In fact, most existing modeling studies [6]- [11] assume that canceling is performed. This assumption simplifies the analysis, thanks to the implicit synchronization introduced by the canceling mechanism (all replicas of a request finish service at the same time).…”

Section: Contributionsmentioning

confidence: 99%

See 1 more Smart Citation

Cutting Latency Tail: Analyzing and Validating Replication without Canceling

Qiu

Pérez

Birke

et al. 2017

IEEE Trans. Parallel Distrib. Syst.

Self Cite

View full text Add to dashboard Cite

Abstract-Response time variability in software applications can severely degrade the quality of the user experience. To reduce this variability, request replication emerges as an effective solution by spawning multiple copies of each request and using the result of the first one to complete. Most previous studies have mainly focused on the mean latency for systems implementing replica cancellation, i.e., all replicas of a request are canceled once the first one finishes. Instead, we develop models to obtain the response-time distribution for systems where replica cancellation may be too expensive or infeasible to implement, as in "fast" systems, such as web services, or in legacy systems. Furthermore, we introduce a novel service model to explicitly consider correlation in the processing times of the request replicas, and design an efficient algorithm to parameterize the model from real data. Extensive evaluations on a MATLAB benchmark and a three-tier web application (MediaWiki) show remarkable accuracy, e.g., 7% (4%) average error on the 99 th percentile response time for the benchmark (respectively, MediaWiki), the requests of which execute in the order of seconds (respectively, milliseconds). Insights into optimal replication levels are thereby gained from this precise quantitative analysis, under a wide variety of system scenarios.

show abstract

Section: Contributionsmentioning

confidence: 99%

“…Request replication has been considered in [3], [6]- [11], [14]- [18] as an efficient way to combat the variability in latency. Joshi et.…”

Section: Related Workmentioning

confidence: 99%

Cutting Latency Tail: Analyzing and Validating Replication without Canceling

Qiu

Pérez

Birke

et al. 2017

IEEE Trans. Parallel Distrib. Syst.

Self Cite

View full text Add to dashboard Cite

show abstract

“…Recent research have highlighted the potential advantages of replicating requests in systems in order to consistently achieve low latencies [27]. We here consider a variant of the proposed model to assess the potential impact of query replication on Cassandra.…”

Section: A What-if Scenario: the Impact Of Query Replication In Cassamentioning

confidence: 99%

A Queueing Network Model for Performance Prediction of Apache Cassandra

Dipietro¹,

Casale²,

Serazzi³

2017

Proceedings of the 10th EAI International Conference on Performance Evaluation Methodologies and Tools

View full text Add to dashboard Cite

NoSQL databases such as Apache Cassandra have attracted large interest in recent years thanks to their high availability, scalability, flexibility and low latency. Still there is limited research work on performance engineering methods for NoSQL databases, which yet are needed since these systems are highly distributed and thus can incur significant cost/performance trade-offs. To address this need, we propose a novel queueing network model for the Cassandra NoSQL database aimed at supporting resource provisioning. The model defines explicitly key configuration parameters of Cassandra such as consistency levels and replication factor, allowing engineers to compare alternative system setups.Experimental results based on the YCSB benchmark indicate that, with a small amount of training for the estimation of its input parameters, the proposed model achieves good predictive accuracy across different loads and consistency levels. The average performance errors of the model compared to the real results are between 6% and 10%. We also demonstrate the applicability of our model to other NoSQL databases and other possible utilisation of it.

show abstract

“…For instance, [1] observes in trace-driven experiments that processing replicas concurrently is effective in mitigating the effect of latency. [17] explores concurrent replication with canceling as a tail-tolerant approach, and shows that this approach can be powerful in keeping the response time tail short. In particular, [7,15,16] evaluate the effectiveness of this approach in improving the reliability for systems with request failures.…”

Section: Related Workmentioning

confidence: 99%

“…Concurrent replication has been considered recently as a means to reduce latency in interactive and deadline-driven applications [1,3,7,16,15,10,17,20,22]. For instance, [1] observes in trace-driven experiments that processing replicas concurrently is effective in mitigating the effect of latency.…”

Section: Related Workmentioning

confidence: 99%

Tackling Latency via Replication in Distributed Systems

Qiu

Pérez

Harrison

2016

Proceedings of the 7th ACM/SPEC on International Conference on Performance Engineering

Self Cite

View full text Add to dashboard Cite

Consistently high reliability and low latency are twin requirements common to many forms of distributed processing; for example, server farms and mirrored storage access. To address them, we consider replication of requests with canceling -i.e. initiate multiple concurrent replicas of a request and use the first successful result returned, canceling all outstanding replicas. This scheme has been studied recently, but mostly for systems with a single central queue, while server farms exploit distributed resources for scalability and robustness. We develop an approximate stochastic model to determine the response-time distribution in a system with distributed queues, and compare its performance against its centralized counterpart. Validation against simulation indicates that our model is accurate for not only the mean response time but also its percentiles, which are particularly relevant for deadline-driven applications. Further, we show that in the distributed set-up, replication with canceling has the potential to reduce response times, even at relatively high utilization. We also find that it offers response times close to those of the centralized system, especially at medium-to-high request reliability. These findings support the use of replication with canceling as an effective mechanism for both fault-and delay-tolerance.

show abstract

Evaluating the Effectiveness of Replication for Tail-Tolerance

Cited by 12 publications

References 17 publications

Cutting Latency Tail: Analyzing and Validating Replication without Canceling

Cutting Latency Tail: Analyzing and Validating Replication without Canceling

A Queueing Network Model for Performance Prediction of Apache Cassandra

Tackling Latency via Replication in Distributed Systems

Contact Info

Product

Resources

About