PernixData Unbuttons Trench Coat at SFD3 and Reveals..

Flash Virtualization Platform.


It seems in Storage circles there is much discussion around using new technologies to cache data in faster more accessible media. Flash is everywhere.. but there are many choices for where and how you can deploy flash technology in order to alleviate strain on storage systems, whose current SAS/SATA based disk drives are struggling to keep up with the day to day IOPS requirements of many organisations.

Pernix Data believe they have the solution for VMware environments in the form of their Flash Virtualisation Platform (FVP). They certainly have the credentials to be making such claims with team members coming from many leading companies. This includes, Satyam Vaghani, their CTO who came from VMware and was responsible for the creation of VMware’s VMFS filesystem.

FVP is a hardware agnostic server-side flash virtualization solution.  It virtualizes PCIe flash cards and local SSD at the server side. What I found particularly impressive about this solution is that it is a software only solution, that looks very easy to implement. It sits seamlessly between hypervisor and SAN without requiring any configuration changes to the either the datastores at the hypervisor or LUNs at the SAN. It’s just a extension that is easily implemented on vSphere.

The clustered nature of the product also overcomes some of the current server-side flash device challenges. When flash caching is being used in a server, a footprint of hot, commonly accessed data is built up for the workload running. If a VM (and its associated workload) migrate to another host due to vMotion or some other reason, the footprint needs to be recreated from scratch. FVP resolves this by replicating copies of the footprint data to other host in the cluster, making it easy for a VM to pickup it’s cached footprint if it moves. There are also, obviously data protection benefits to keeping multiple copies of the data in the event a server dies along with it’s cache.

In addition, to easy implementation the product also provides solid easy to read stats on what results it’s achieved. A sure fire way to build a solid business case around IOPS saved and the reduced requirement to scale up your SAN to deal with load.

What these new caching capabilities amount to is an entirely new storage tier between RAM & SAN. This new tier (or layer) will definitely come with challenges. One such challenge would be ensuring consistent copies of data at the SAN for things like Backup processes. If FVP caches the data at the server, some reads/writes never actually reach the SAN. So if your backing up from SAN you need a way to flush the data through in a consistent state. FVP does include a “write-through” mode which should flush changes to disk and stop caching (i.e. write-back mode). In order to achieve consistency there will need to be careful orchestration from VSS (or prefreeze/postthaw scripts on Linux) to FVP to VMware Snapshot and beyond. The product will have a PowerShell interface which could be used to switch between write modes for such an operation, but users should be aware that this is a requirement.

All in all, FVP looks like a great product that hist a lot of my hot buttons.. easy-to-install, easy-to-use, transparent but powerful and solid reporting on results. Although not publicly available as yet, it will be interesting to see the licensing model, TCO and ROI information Pernix provide. If they get that right, they could be very successful.

Here is the Intro from Storage Filed Day 3: