Footprint Methods Independently Validated

In his MS thesis published in Vancouver Canada four months earlier, Zachary Drudi reported the result of using the footprint technique developed by us to predict the hit ratio curve for a number of Microsoft traces, for cache sizes ranging from 0 to gigabytes (up to 1TB).  The footprint prediction is accurate in most cases.  Below is the plot copied from the thesis (page number 32, page 40), where avgfp is the footprint prediction, and mattson is the ground truth calculated from the reuse distance (LRU stack distance).

The author implemented our technique entirely on his own without consulting or informing any of us.

The thesis is titled A Streaming Algorithms Approach to Approximating Hit Rate Curvesavailable online from the University Of British Columbia.

Screen Shot 2015-03-05 at 9.20.34 AM

(copyright Zachary Drudi, 2014)

In a series of papers in PPOPP 2008 (poster), PPOPP 2011, PACT 2011, and ASPLOS 2013, Rochester researchers have developed a set of techniques to measure the footprint and use it to predict other locality metrics including the miss/hit ratio curve and reuse distance, for exclusive and shared cache.  We have shown that the footprint techniques are efficient and largely accurate.  Mr. Drudi’s results provide an independent validation.

This is the first time we know that the technique is used in characterizing storage workloads.  The same implementation was used in their OSDI 2014 paper Characterizing storage workloads with counter stacks, J. Wires, S. Ingram, N. J. A. Harvey, A. Warfield, and Z. Drudi.

One thought on “Footprint Methods Independently Validated

  1. I took a look at their OSDI paper and found that the paper is about how to efficiently approximate MRC. To be honest, I am a little surprised to see this topic is published in OSDI.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s