Building a Distributed Cloud Storage System for Researchers is Not as Hard as You Think....Or is it?

Stephen Walsh
TrackTrack 1 -- Lecture Hall X on the First Floor
SessionCloud
DescriptionIn 2012, AARNet embarked on a project to build a cloud based storage system using solely open source components. Distributed file-systems had been around for a while, and the edge cases where they didn't suit were few and far between. Master-Master Databases were known to be painful over high latency internet links, but with AARNet owning the backbone links in use, there were technical solutions to ease this pain. 18 months on, the deployment team is now much, much, wiser. Little did the engineering team know, but these perceptions, plus a lot more, would be challenged with every choice of technology.
From grumpy file systems to latency-triggering replication wars, every bug report seemed to find more and more assumptions that had been made on the eventual use of each particular building block by its development team. With each discovery of a limitation, the deploying engineers found themselves thinking further and further out of the box to find a solution. More and more products came into the testing system, including Gluster, Ceph, XtreemFS, and even a multi?master BitTorrent based replication system, in an attempt to find 'The One That Would Work'. Come along on a tour through what should have been a simple-to-deploy research facilitation tool that turned into something that consumed engineers, open source projects and waking hours in ever increasing amounts.

All talks