Jabba Storage Array

A lot of the World Community Grid crunchers know that the Harvard Clean Energy Project (CEP) uses a lot of disk space to store the results of our calculations, as each workunit has on average about 20 MB. Our expensive storage arrays “Miss Piggy 1 & 2” started to fill rapidly. Therefore, we set out to search for a low-cost solution where we could potentially also store our data locally. We eventually came across Backblaze, a cloud storage backup company, which shares the design of their 45-hard-drive storage boxes with the online world. The people at Backblaze even publish their component and vendor list!

After laying eyes on those red shiny boxes, we decided to have a go at it and build our own storage machines following the Backblaze template. A closer look at the part and vendor list revealed the first bump in the road: some of the parts from the 2009 original were already outdated and needed to be substituted with modern ones. And some vendors really only wanted to sell in bulk, so that some parts (e.g., the wire harness) had to be custom-made. We put a detailed component and vendor list of our 90TB version of the Backblaze pod - lovingly named 'Jabba’- below.

Building Jabba was quite a blast. People from the Aspuru-Guzik group (Sangwoo, Xavier, Dmitrij, Sergio, Jacob) and other labs at Harvard (Roger) were extremely helpful to creating Jabba. Johannes has documented the whole process with pictures over on the CEP Facebook page, but here’s a how it looks as we neared the end of the setup.

The built went surprisingly smooth and without bigger incidents. We struggled a bit with organizing all the cables and messed up daisy-chaining the power supplies at first. The on/off button was also surprisingly hard to figure out - the instructions must have come from Mars... But that was already it and switching the box on for the first time was truly exhilarating. It runs much more quiet than expected.

After trying different linux distros, we decided to install Ubuntu 10.04 Lucid Lynx. We created three RAID6 arrays, each having 15 disks and 30 TB total raw capacity. After some benchmarking we can now let it go online!

We plan to build at least two more of these. Next time around we may buy only the drives for one RAID section at a time to cash in on falling hd prices. We will definitely keep you updated how our first jabba machine turns out.

Last but not least, we'd really like to thank the people at Harvard's Faculty of Arts and Sciences Research Computing who have immensely helped us with the Ms. Piggy arrays. Now that we take care of little Jabba, they have provided us tons of help in managing it and getting it ready to receive data. Lastly, their motivation via blog posts has really been invaluable.

Jabba's Vendor List

Since we were not buying in bulk, certain items were hard to find and/or came with slightly higher price-tags, especially the port multiplier backplanes, the rubber bands and certain cables.  Performance-pcs were very helpful to provide us custom-made cables. We should have gotten a motherboard with an integrated graphics card, since we will not need an output very often. Finally, it is likely that two 525 W power supplies are sufficient to power the low power Barracudas.