Question: What do you get when Pure Storage gets to build a system that can start small, grow big, handle file requests quickly and is simple to manage?

FlashBlade: Pure’s newest addition to its hardware portfolio. The Pure Storage FlashBlade is not just another NAS filer. It’s an all-flash, scale-out storage for file (NFSv3 for now) and object (soon), delivering some pretty good performance as you can see in the sheet above. And the chassis just looks sexy…

FlashBlade, with the emphasis on Blade

I don’t like rack mounted servers and that’s mostly due to the cabling mess. If you need to connect a whole lot of servers together, connectivity is bound to become a problem. Pure Storage noticed the same problem in their earlier FlashBlade prototypes and decided to take the blade chassis approach.

There’s quite a bit of proprietary hardware involved here: for example the NAND flash chips are accompanied with super-capacitor backed NVRAM. The reason for this is that you want to write to flash with full pages of data, for reasons of flash endurance and performance. If you don’t have enough data coming in, you will want to save up some data and write it out to flash in one go. That’s where the NVRAM comes in. The supercaps make sure that the data in NVRAM isn’t lost in case of a power outage, but safely written out to a persistent media.

The blades come in two sizes: 8TB (which is actually 8.8TB) and 52TB, allowing you to choose between a somewhat more capacity oriented platform or a pure race car. A cluster starts out with 7 blades minimum, after which you can add individual blades and expand your cluster step by step. Par Botes from Pure Storage spoke about 1GB/s sustained speed per blade regardless of the request size, which is quite impressive…

FlashBlade Networking

The blade chassis contains two software-defined network switches in the rear.

These switches collapse the front-end, back-end and control networks into one high speed fabric (as you can read in the picture). Out the back come eight 40Gb/s connections which you can plug into your network. And some power cables…

The cluster of 15 FlashBlades in a chassis will be presented out to the external network as a single endpoint. The switches are programmed to create virtual IP addresses for the individual blades. Load balancing across the available blades is done by the switches: there’s no DNS loadbalancing in your clients or your external network.

Large files are split into multiple chunks to prevent large files from becoming a bottleneck and resulting in traffic that just goes to one node. Files are currently split into 16MB chunks based on Pure Storage R&D results, finding a balance between too many chunks and too few.

My thoughts on the FlashBlade…

I’m happy to see Pure Storage talk about a second product in their portfolio. Dan mentioned people calling Pure Storage a “one trick pony” and I’ll confess to thinking that myself a couple of times. With the FlashBlade system, Pure is branching out into a completely different market segment, which is always good for company stability.

Pure knows how to create a good presentation, and they didn’t disappoint at SFD10. Make sure you check out Brian Gold’s presentation on the FlashBlade hardware architecture! The FlashBlades use a lot of proprietary hardware, but Pure has a good explanation why they do.

Other Storage Field Day 10 delegates also have opinions well worth reading: here’s an extensive post from Max, one from Ray and last but not least a post by Enrico. From a networking perspective, make sure to check out Tom’s post on the network capabilities.

Disclaimer: GestaltIT paid for the flight, hotel and various other expenses to make it possible for me to attend SFD10. I was however not compensated for my time and there is no requirement to blog or tweet about any of the presentations. Everything I post is of my own accord.