[cs-talks] NRG Talk: Ahmed Ali-Eldin, Mon. 11/6 @11am, Hariri Seminar Room

Harrington, Jacob Walter jwharrin at bu.edu
Wed Nov 1 10:56:52 EDT 2017


Did We Really Need All These Algorithms?: A Brief History of Autoscaling
Ahmed Ali-Eldin, Post-Doc at UMass, Amherst
Monday, November 6th from 11am-12pm in the Hariri Seminar Room (MCS-180)

Abstract: Datacenter applications are diverse, ranging from request-response workloads with milli/microsecond processing times, to task based workloads that would run for hours. To manage the performance of these applications, many dynamic provisioning algorithms (aka autoscaling algorithms) have been proposed in the literature. Autoscaling algorithms dynamically control the amount of resources allocated to an application based on its workload aiming to improve the application’s QoS and reduce its overall running costs. In this talk, I will present a brief history of autoscaling, how autoscalers are introduced as the holy grail of elastic cloud computing, and how these algorithms have evolved. In most of the published work, when a new autoscaling algorithm is proposed, it is seldom compared to the state-of-the-art, and is often compared only to static provisioning using a predefined QoS target. In order to better understand the performance and the state-of-the-art of these algorithms, we re-implemented and acquired implementations of popular, highly cited, algorithms in the literature and evaluated their performance. I will discuss how, in many cases, the newly published — sophisticated — algorithms perform worse when compared to the mostnaive autoscaling algorithms one can think of. I will discuss approaches that one can use to establish bounds and guarantees on the performance of autoscaling algorithms using stochastic control theory, and simple statistical modeling.

This work was done in collaboration with Lund University (Sweden), TU-Delft (The Netherlands), and Wurzberg (Germany).

Bio: Ahmed Ali-Eldin is a postdoc at UMass, Amherst and a research scientist at Umea University, Sweden. He obtained his PhD in 2015. He is a Heidelberg Laureate Forum Alumni. His work lies somewhere in the intersection between building systems and building performance models trying to understand why (his) systems do (not) work. When he is not doing useful work, he rants about computer science research, benchmarks, and workloads.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://cs-mailman.bu.edu/pipermail/cs-talks/attachments/20171101/a5238c3a/attachment.html>


More information about the cs-talks mailing list