Who: Stephen Ingram, R&D Engineer at Coho Data, Vancouver, Canada

When: March 9, 2014, 10am

Where: Währinger Straße 29, 1090 Wien, Hörsaal 2

Title: Storage Workload Analysis and Simulation with Counter Stacks, a Streaming Algorithm Approach

Abstract:

Much of our world's public and private information is being stored in today's corporate data centers.  Modern, high-performance data-center hardware utilizes a hybrid of storage media like flash and spinning disks that exhibit vastly differing performance characteristics. Using existing cache evaluation methods to analyze the behavior of storage workloads on this hardware presents numerous computational challenges. In this talk I present the Counter Stack data structure.  Counter Stacks utilize ideas from streaming algorithms to efficiently process and store cache-utility information for each workload on a system.  In addition to computing standard cache statistics in orders-of-magnitude less space than existing approaches, I present how Counter Stacks can be used to perform novel cache simulations, opening the door to powerful new approaches in storage system design.

short CV:

Stephen Ingram is an R&D Engineer at Coho Data, a computer storage startup in Vancouver, Canada, where his team designs and builds the analytics stack of the company's data centre storage products.  His research interests include streaming algorithms, dimensionality reduction, data journalism, information visualization, and computer graphics.  He received his PhD in 2013 from the University of British Columbia.  His homepage is http://www.cs.ubc.ca/~sfingram