Flash is changing the storage landscape and as we move from SLC to eMLC and then onto 3D TLC the cost per TB is plummeting, but it is important to point out that even the cheapest forms of flash are no match for a NL-SAS HDD when it comes to cost per TB. The all-flash Data Centre is therefore still not a reality for the foreseeable future, but for any application that will benefit from increased performance the cost per IO of Flash is unbeatable.
So introducing Flash is inevitable for the majority of organisations, but the challenge for many is that they have existing storage arrays that still have many years of serviceable life, so do they:
- Add SSDs to their existing storage arrays
- Replace/augment what they have with a Hybrid Flash Array
- Replace/augment what they have with an All-Flash Array
The problem with option 1 is that unless the array is fairly new adding SSDs will typically not deliver great performance benefits due to bottlenecks in the controllers, as they were designed more for the HDD era – it will therefore be a false economy. The problem with options 2 and 3 is that budgets may not be available as the organisation wants you to “sweat your assets” for another few years.
So are there any other options?
If there is nothing that can be done with regard to your storage arrays, then the best option is to look to put flash into the hosts even closer to the applications. Now VMware has a product called vSphere Flash Read Cache which does exactly this, the problem is that it only caches reads and many workloads (such as VDI and Databases) are very write heavy. Also it is only included with the Enterprise Plus version of vSphere, which most organisations do not have.
This is where PernixData FVP comes in, a solution designed by the author of VMFS, VAAI and Virtual Volumes. Its goal is very simple, get more performance out of your existing storage, no matter the application type. By caching the working set of the application on the host you deliver the equivalent of all-flash performance with the back-end storage just being a capacity tier that does not really have to be sized for performance. Key features include:
- 100% Software-Defined – any server, any Flash/RAM, any storage
- Dramatically accelerate applications – improve IOPS, throughput and latency of both reads and writes
- Simple and Non-Disruptive – deploy a high-performance Kernel Module in less than 10 minutes with no changes (or reboots) to hosts, storage, network or VMs
- Seamless operations – fully supports VM migrations (vMotion, DRS, HA) and VMs can remotely access data on server-side Flash/RAM on demand for continued acceleration
- RAM compression technology – ensures the best performance at price points similar to flash
- Scale-out storage performance in a predictable manner – add storage performance where and when you need it most by adding more server media
In conclusion FVP aggregates server-side Flash/RAM into a scale-out data acceleration tier that can leverage any storage platform for capacity, creating a decoupled storage architecture that scales performance with compute resources.
It is important to point out that FVP is not just a tactical solution to fix an existing under-performing storage array – it can also become a strategic play as it changes the way you purchase new storage; moving forward you assume the performance is taken care of by the host and therefore the storage array is just sized for capacity. The key point is that if and when you need to increase storage performance you just add commodity SSDs to your servers, no need to add any extra SSDs to your storage arrays or upgrade their controllers (if they have become a bottleneck), or add any further host licenses.
Related Posts
by 