HDD as the main storage medium
The Hard Disk Drive (HDD) has been around a long time and is likely to be here for a good while yet. It still remains the cheapest way to supply
mass storage to computer systems, but in the Enterprise the need for ever increasing performance is causing the good old HDD some issues.
Firstly the maximum speed of a spinning disk is pretty much bound by the laws of physics, hence the reason it has stuck at 15,000 rpm for years. Anything
much above this and the amount of power required and potential for unsustainable error rates just does not make a good enough trade off for the increase
in performance.
Most importantly though the rise of virtualisation is seeing an exponential increase in demand for performance in the form of IOPS (Input Outputs Per Second) and
conventional disks just cannot deliver what is required of them.
Virtualisation by its nature causes a huge amount of random I/O as multiple servers or desktops
act on the same area of storage - while one is busy writing data another may want to read, one could be a database with many little read and write transactions and another may be
sequentially reading from a huge file. Put all this together and you have very random data in both read and write form and the drive heads just cannot keep up.
There are various mitigation techniques such as RAID that help spread the load across mulitple disks, but as we eluded to in a
blog article this is just a sticking plaster
over a festering wound. More drastic action is needed to be able to move forward.
The age of the SSD
Flash drives, also known as Solid State Drives (SSD) have also been around for quite a while now. Many will be familiar with them as the ubiquitous USB storage key, but in order to be useful as
mass storage medium they had to be able to be used in the same way as conventional drives. This lead to the emergence of the SSD as a like for like replacement for the HDD.
The advantages of the SSD are that there are no moving parts, so the physical limitations of the spinning disk are removed. Also because of this they are just as efficient at
handling random I/O as they are sequential I/O, and therefore are much more suited to virtualisation workloads. They are quicker too, whereas a 15k HDD can typically only deliver
around 160 IOPS it is not uncommon for an SSD to deliver 30,000 IOPS.
Until now the disadvantage of Flash storage was its expense, with some SSDs costing 10 to 20 times per TeraByte compared to a conventional disk. To bring this cost down
cheaper versions of these drives were produced, but these had a significantly lower lifetime and were still relatively expensive so the humble HDD wasn't really threatened.
However developments in technologies to deliver data to and from banks of Flash drives, in the same way RAID came along for the HDD, have produced a product that loses none
of the advantages in performance - but comes at a cost and lifetime now more equivalent to their HDD competitors.
We have now come to a point where HDD cannot sustain the demand for performance from virtualised systems, and more and more we will see the Flash drive in the form of
SSD populated storage arrays become the favoured choice. The spinning disks will maintain the lead in terms or raw capacity for a long time, and as such will be used
for online backup and archiving.
One exciting player in this new market is Nimbus.
