EMC is officially pushing it’s all-Flash array, XtremIO, with full force. This is in direct competition with PURE storage who has been the traditional leader in the All-Flash Array (AFA) market. The biggest differences between XtremIO and PURE is in the architecture of the box and the data services each offers. The interesting thing is that the PURE box is a Scale-Up, dual-node, active-passive architecture while the XtremIO is a tightly integrated Scale-Out design. PURE scales by just by adding drives and daisy chaining the enclosures to the Storage Processors. This begs the question, how much Flash can a PURE storage hold before their backend bus is potentially over saturated?
This was an issue for the older EMC Clariions which leveraged a 4Gbps FC back end. The issue came with people who had a CX4 and wanted to add SSDs for Fast Cache. EMC has learned from their past and have made adjustments in how they handle flash for when they released XtremIO. As you add XtremIO X-bricks you add addition compute power and network connectivity to the controllers. This is all built on a 40Gbps Infinaband RDMA backend, while PURE only has 6 x 6Gb/s SAS ports for controller for connectivity to all those MLC drives for data and the SLC caching drives in their enclosures.
This lead me to do some math on how many SSDs can you load in a PURE box before it is possible to over saturate the backend bus to the controllers. After doing some quick math it seems that 50 SSDs would be the breaking point for a PURE storage’s 6Gbps 6 lane backend…
6 Lanes * 6Gbps = 36 Gbps (essentially)
36 Gbps * 125 = 4,500 MBps
4,500 MBps / 90 MBps (The max read bandwidth of an enterprise SSD) = 50 SSDs
This is also assuming we are looking at more powerful enterprise grade MLC drives like in XtremIO and not commodity grade MLC drives that PURE uses. Keep this in mind you would would need to slam all those SSDs at once. So this is dependent on actually having a workload that is pushing the maximum that much bandwidth. While on XtremIO it would be impossible to ever over saturate the backend because of it’s Scale-Out architecture.
Now I am by no means an expert on PURE’s architecture but it seems to me that at scale it is flawed. I am just looking at this from the problems I have seen in my years of experience with the Clariion and other Scale-Up architectures like it. I would love to hear from an engineer at PURE on how they address this traditional Achilles’ heel of Scale-Up storage.