We are big fans of Lustre here at insideHPC. And when I learned that Aeon Computing was building their Data Oasis storage platform on the open source file system, I sought out the company’s co-founder Jeff Johnson to learn more
insideHPC: Why build a Lustre appliance? What was the customer problem you set out to solve?
Jeff Johnson: This is all about performance. We originally set out to plan a Lustre deployment and brought in storage solutions from various well known manufacturers for Lustre benchmarking. We were surprised to discover that many storage devices that looked to be high performance on paper performed quite poorly in a Lustre environment. As a result we decided to develop one ourselves. In this particular case the customer needed Lustre performance without all of the extra, and costly, active failover high-availability features found in the very high end Lustre solutions available. That is what we built.
insideHPC: What sets the Oasis Appliance from other available storage solutions available out there?
Jeff Johnson: There are many storage solutions available in the market but not all of them do Lustre well. We set out to design a Lustre platform that was good at Lustre data and I/O profiles. Part of that design, in addition to performance, is that it follows Aeon Computing’s business philosophy in that there is no unnecessary, extraneous bull___t that gets in the way.
insideHPC: Is Oasis just for the HPC market, or does it have appeal to the Big Data and Cloud-based applications?
Jeff Johnson: The Data Oasis filesystem is specifically an HPC application. The EclipseSL appliance is well suited for any application where reliable, dense high-performance storage is needed. We have mainly focused on Lustre, the EclipseSL would perform well in Gluster and other distributed filesystem applications.
insideHPC: How does the Aeon design change the way you architect a Lustre system?
Jeff Johnson: We designed the EclipseSL with no bottlenecks. The data flow from the network interfaces through bus infrastructures and memory to block storage and back out encounter no bottlenecks or oversubscription. I don’t know that what we did necessarily changes the industry in any way. We simply started with a blank sheet of paper.
insideHPC: What advantages does Lustre offer as a foundation for a storage system?
Jeff Johnson: Bandwidth. Its performance scales out linearly as the file system scales in build out. The more object servers you have, the more network paths you have, the faster your potential. It is the opposite of a large scale monolithic NFS appliance with one spigot.
insideHPC: What was your first reaction when you learned you won the Best of Show Award?
Jeff Johnson: We checked the calendar to make sure it wasn’t April Fools Day. There may have been some expletives of disbelief uttered in the office. Once it sunk in, humbled, flattered and still a bit surprised.
Check out Aeon this week at SC12 booth #2119.
So it looks like they have put Lustre on a Supermicro storage server?
Many OEMs do. Standard chassis with special insides. DDN uses a Newisys chassis. Very few companies actually built their own chassis.
@w, yes, these are Supermicro servers. The difference is that Aeon took the time to design capability around Supermicro’s offerings.