Cutting Tail Latency in Cloud Data Stores via Adaptive and Batch-aware Replica Selection and Scheduling

The Faculty of Informatics is pleased to announce a seminar given by Marco Canini

DATE: Wednesday, June 10th 2015
PLACE: USI Lugano Campus, room SI-008, Informatics building (Via G. Buffi 13)
TIME: 10.30

ABSTRACT:
Predictable performance is crucial for many data center applications such as interactive web services and data analytics. It is however difficult to achieve predictable performance due to many factors that introduce performance outliers and skew the tail of the latency distribution even in well-provisioned systems. To make matters worse, modern applications are highly distributed, and likely to get more so as cloud computing continues to separate users from their data and computation.
In this talk, we present our ongoing work to design and analyze algorithms as well as system techniques for reducing tail latencies in order to achieve more predictable performance for replicated and partitioned data stores in multi-tenant environments.
We discuss the fundamental challenges involved in designing a replica selection scheme that is robust in the face of performance fluctuations across servers. We illustrate these challenges through performance evaluations of the Cassandra distributed database on Amazon EC2. We then present the design and implementation of an adaptive replica selection mechanism, C3, that is robust to performance variability in the environment. We demonstrate C3's effectiveness in reducing the latency tail and improving throughput through extensive evaluations on Amazon EC2 and through simulations.
Our results show that C3 significantly improves the latencies along the mean, median, and tail (up to 3 times improvement at the 99.9th percentile) and provides higher system throughput (up to 50%).
We then discuss our ongoing work to extend C3 with scheduling algorithms to further lower the latency in accessing the data store.
Based on the observation that applications access several elements from the data store in batches, our key idea is to leverage batch characteristics as an opportunity to minimize latencies coupled with adaptive, batch-aware replica selection.

The talk is based on the paper at NSDI'15 and on ongoing work in the Absinthe project.

BIO:
Marco Canini is an assistant professor in the ICTEAM institute at the Université catholique de Louvain. Marco obtained his Ph.D. in computer science and engineering from the University of Genoa in 2009 after spending the last year as a visiting student at the University of Cambridge, Computer Laboratory. He holds a laurea degree with honors in computer science and engineering from the University of Genoa. He was a postdoctoral researcher at EPFL from 2009 to 2012 and after that a senior research scientist for one year at Deutsche Telekom Innovation Labs & TU Berlin. He also held positions at Intel Research and Google
(personal page)

HOST: Prof. Robert Soulé