[talks] Raghav Sethi will present his masters thesis talk on Friday, May 8, 2015 at 1pm in CS 302

Nicki Gotsis ngotsis at CS.Princeton.EDU
Fri May 1 13:25:24 EDT 2015

Raghav Sethi will present his masters thesis talk on Friday, May 8, 2015 at 1pm in CS 302. 

His readers are Michael Freedman (adviser) and Jennifer Rexford. 

Everyone is invited to attend his talk. His abstract follows below. 

Exoskeleton: Fast Cache-enabled Load Balancing for Key-Value Stores 

Key-value stores are a fundamental building block for many large web-scale applications. These applications typically generate highly-skewed traffic, i.e. the most popular keys are requested many orders of magnitude more frequently than the least popular keys. Schemes that simply balance storage capacity across a cluster are therefore unable to adequately balance traffic load across cluster nodes. Dynamic load-balancing is therefore often necessary, but if it is imperfect or slow, operators must still over-provision individual nodes to be able to deal with rapidly-changing traffic distributions, increasing equipment and operational costs. An effective technique to provide load-balancing is to use popularity based caching. 

Exoskeleton is a cache architecture which enables the routing of requests for popular keys directly to cache nodes at line-rate. This architecture leverages the power and speed of programmable switches to perform content-aware routing. Clients do not need to maintain any state to take advantage of this architecture, and simply encode key information directly into request packet headers. Exoskeleton removes the constraints of previous architectures, which required operators to optimize for either low latency or scalability, but not both. In addition, Exoskeleton deals with rapidly-changing workloads through a cache update mechanism designed specifically for the limitations imposed by switch hardware. Effective load-balancing enables this system to efficiently deal with modern web workloads using a heterogeneous cluster architecture comprising cheap SSD-based backend nodes and a small number of powerful cache nodes. 

More information about the talks mailing list