HARVEY MUDD COLLEGE

I will talk about Cumulus, a system I have built for efficiently storing backups of data from end-user computers to “cloud storage systems”—services that offer to store data for customers over the Internet, generally at a flat rate per gigabyte of data stored and transferred. For portability, Cumulus is restricted to only a simple get/put interface to the remote storage, but I show that this restriction imposes only a low overhead in terms of storage, bandwidth, and cost—even compared with more complex approaches. Cumulus achieves this by aggregating small files together for storage and incorporating ideas from log-structured file systems. I may also talk about research and graduate school more generally.