Flash - saviour of the (I/O) universe?

If you read certain sections of the media you would not be wrong in thinking that flash storage is like ‘the cavalry’ coming over the hill to save the day, but is it going to be the industry saviour?

There is no doubt that I/O has become the biggest bottleneck in the computing model. This is as a result of the difference between the performance growths of ‘silicon’ vs mechanical technologies over time. Today, compared to 10 years ago, a CPU produces about 1000x the number of cycles per second, compared to the I/O rate of a mechanical drive, which is essentially the same (2x if you are generous!). Hence the entry of the enterprise flash storage device!

For a few years now, leading storage vendors have been incorporating flash storage into their arrays, as a means to deliver performance, while mechanical disks provide the low-cost storage. With the right amount of automation the cost of storage systems can be reduced, in some cases around the 30% mark. However, there is even more benefit to using flash as a storage medium if we rethink the whole thing. New start-up companies are beginning to engineer storage products from the ground up to use flash as the storage media and the initial results are fantastic.

As an example of this change, consider the RAID mechanism for protecting data. One of the fundamental constructs of the current RAID algorithms is that you are dealing with a set of physical disk drives. The algorithms have therefore been designed to understand the locality of the data, and, in essence, attempt to overcome the physical limitations of physical drives, such as seek times.

Now consider flash memory, where there are no seek times and, in fact, there is no concept of ‘devices’. Add flash and you get more capacity! With these factors in mind, the data protection algorithms can we rewritten to produce better protection with less overhead - certainly dramatically different to the reads and writes that have to be done in a traditional, mechanical disk, RAID system.

So what is the downside? Well, this makes performance of these devices harder to define. Our training has us looking at seek, access and transfer times on mechanical disks to compare performance metrics, but when it comes to flash devices, the parameters are different. Think about the operation of a flash device in a storage array. When it’s new, and the buffer is empty, and it’s merrily writing to the next cell, everything is working smoothly. Then the buffer fills, and it needs to flush before it accepts new requests. Either that or the flash fills and it has to ‘garbage collect’ to find the cells that can be overwritten. Then, there is the impact of the ‘intelligence’ built into the flash device as well as when and how will it perform these operations.

Each solid state drive (SSD) manufacturer in the past has utilised different measurement methodologies to derive performance specifications for the products they make. Thus, it’s been difficult for buyers to fairly compare the performance specifications of SSDs from different manufacturers. This is the reason SNIA decided to take the complexity out of this situation and to create a series of independent standard conformance tests, which will allow buyers to quickly and accurately compare these devices. The Solid State Storage Performance Test Specification (SSS PTS) defines a suite of tests and test methodologies to measure the performance characteristics of SSDs. The SSS PTS provides performance measurements that may be fairly compared to those of other SSDs measured the same way, in the same environment. For further information visit http://www.snia.org/forums/sssi.

This is just the beginning of a major shift towards flash storage, and as we move into the ‘big data’ world, it is technologies like this which will enable us to make best use of all the data that is available, and make a profound impact on the world we live in!