Excellence in Performance Testing
I’ve spent quite a bit of time working with the group of folks (yes, I’m from the American South) over at Storage Review this year. The primary goal was to take what we consider to be a solid configuration of HPE Apollo servers and present them as block and object storage from SUSE Enterprise Storage and get some solid benchmark data. As a side note, I ate some really good chili (Skyline Chili) and barbecue (Montgomery Inn-Boat House) while in Cincinnati, thanks for the recommendations Edwin!
The SUSE Enterprise Storage cluster has 6 HPE Apollo nodes with 24 6TB drives and 4 SSDs each to be used for journals. We built the cluster using redundant and aggregated 40Gb ethernet connections. The result is a cluster of over 800TB of raw space that is capable of turning in very respectable performance. An implementation guide for this hardware can be found here.
One thing that I was very pleased with from the Storage Review team, was the devotion to doing the job right. They took the time to discuss the config, the protocols, the networking, everything and then made sure their test would exercise it. We even increased the working set sizes to ensure that we ran past caching that would be present where possible. The results can be found here.
As a group, we discovered the industry knowledge and standardization around object testing is still maturing. This is why there are no published benchmarks on object at this time. It is my opinion that Storage Review wants to ensure all the proper things are measured and interpreted correctly before kicking something out the door. I find this refreshing and am glad to have made the acquaintance of each and every person there and look forward to doing further work with them in the future.