Tuesday, 12 December 2017

Session 2: Best Paper Candidates

Reducing Task Completion Time in Mobile Offloading Systems Through Online Adaptive Local Restart

Authors:

Qiushi Wang (Freie Universität Berlin)
Katinka Wolter (Freie Universität Berlin)

Abstract:

Offloading is an advanced technique to improve the performance of mobile devices. In a mobile offloading system, heavy computations are migrated from resource constrained mobile devices to powerful cloud servers through a wireless network connection. The unreliable wireless network often disturbs system operation. Task completion can be delayed or interrupted by congestion or packet loss in the network. To deal with this problem the offloaded jobs can be locally restarted and completed in the mobile device itself. In this paper, we propose a dynamic scheme to determine whether and when to locally restart a task. First, we design an experiment to explore the impact of packet loss and delay in unreliable networks on the completion time of an offloading task. Then, we mathematically derive the prerequisites for local restart and selection of the optimal timeout. The analysis result confirms that local restart is beneficial when the distribution of task completion time has high variance. Further, a dynamic local restart scheme is proposed for mobile applications. This scheme keeps track of the variance of the probability density function of the distribution of task completion time. This is done using a dynamic histogram, which collects and updates data at run time. The efficiency of the local restart scheme is confirmed by experimental results. The experiment shows that local restart at the right time achieves better performance than always offloading.

DOI: 10.1145/2668930.2688041

Full text: PDF

[#][]

Automated Detection of Performance Regressions Using Regression Models on Clustered Performance Counters

Authors:

Weiyi Shang (Queen's University)
Ahmed E. Hassan (Queen's University)
Mohamed Nasser (BlackBerry)
Parminder Flora (BlackBerry)

Abstract:

Performance testing is conducted before deploying system updates in order to ensure that the performance of large software systems did not degrade (i.e., no performance regressions). During such testing, thousands of performance counters are collected. However, comparing thousands of performance counters across versions of a software system is very time consuming and error-prone. In an effort to automate such analysis, model-based performance regression detection approaches build a limited number (i.e., one or two) of models for a limited number of target performance counters (e.g., CPU or memory) and leverage the models to detect performance regressions. Such model-based approaches still have their limitations since selecting the target performance counters is often based on experience or gut feeling. In this paper, we propose an automated approach to detect performance regressions by analyzing all collected counters instead of focusing on a limited number of target counters. We first group performance counters into clusters to determine the number of performance counters needed to truly represent the performance of a system. We then perform statistical tests to select the target performance counters, for which we build regression models. We apply the regression models on new version of the system to detect performance regressions. We perform two case studies on two large systems: one open-source system and one enterprise system. The results of our case studies show that our approach can group a large number of performance counters into a small number of clusters. Our approach can successfully detect both injected and real-life performance regressions in the case studies. In addition, our case studies show that our approach outperforms traditional approaches for analyzing performance counters. Our approach has been adopted in industrial settings to detect performance regressions on a daily basis.

DOI: 10.1145/2668930.2688052

Full text: PDF

[#][]

System-Level Characterization of Datacenter Applications

Authors:

Manu Awasthi (Samsung Semiconductor, Inc.)
Tameesh Suri (Samsung Semiconductor, Inc.)
Zvika Guz (Samsung Semiconductor, Inc.)
Anahita Shayesteh (Samsung Semiconductor, Inc.)
Mrinmoy Ghosh (Samsung Semiconductor, Inc.)
Vijay Balakrishnan (Samsung Semiconductor, Inc.)

Abstract:

In recent years, a number of benchmark suites have been created for the “Big Data” domain, and a number of such applications fit the client-server paradigm. A large volume of recent literature in characterizing “Big Data” applications have largely focused on two extremes of the characterization spectrum. On one hand, multiple studies have focused on client-side performance. These involve fine-tuning server-side parameters for an application to get the best client-side performance. On the other extreme, characterization focuses on picking one set of client-side parameters and then reporting the server microarchitectural statistics under those assumptions. While the two ends of the spectrum present interesting results, this paper argues that they are not enough, and in some cases, undesirable, to drive system-wide architectural decisions in datacenter design. This paper shows that for the purposes of designing an efficient datacenter, detailed microarchitectural characterization of “Big Data” applications is an overkill. It identifies four main system-level macro-architectural features and shows that these features are more representative of an application’s system level behavior. To this end, a number of datacenter applications from a variety of benchmark suites are evaluated and classified into these previously identified macro-architectural features. Based on this analysis, the paper further shows that each application class will benefit from a very different server configuration leading to a highly efficient, cost-effective datacenter.

DOI: 10.1145/2668930.2688059

Full text: PDF

[#][]

Capacity Planning and Headroom Analysis for Taming Database Replication Latency: Experiences with LinkedIn Internet Traffic

Authors:

Zhenyun Zhuang (LinkedIn Corporation)
Haricharan Ramachandra (LinkedIn Corporation)
Cuong Tran (LinkedIn Corporation)
Subbu Subramaniam (LinkedIn Corporation)
Chavdar Botev (LinkedIn Corporation)
Chaoyue Xiong (LinkedIn Corporation)
Badri Sridharan (LinkedIn Corporation)

Abstract:

Internet companies like LinkedIn handle a large amount of incoming web traffic. Events generated in response to user input or actions are stored in a source database. These database events feature the typical characteristics of Big Data: high volume, high velocity and high variability. Database events are replicated to isolate source database and form a consistent view across data centers. Ensuring a low replication latency of database events is critical to business values. Given the inherent characteristics of Big Data, minimizing the replication latency is a challenging task. In this work we study the problem of taming the database replication latency by effective capacity planning. Based on our observations into LinkedIn’s production traffic and various playing parts, we develop a practical and effective model to answer a set of business-critical questions related to capacity planning. These questions include: future traffic rate forecasting, replication latency prediction, replication capacity determination, replication headroom determination and SLA determination.

DOI: 10.1145/2668930.2688054

Full text: PDF

[#][]