I really like many of Kostadis' blog entries, he's pretty informative and appears to have mostly taken a step back from the what was turning into a bit of a MAD world of vendor blogs. His latest blog entry on I/O densities reflects one of my major concerns at the moment, we've been doing a fair amount of modelling on the performance of SATA drives and what workloads are suited for them.
Lots of CIOs and CTOs are looking at SATA drives and thinking that we can put everything on them as the costs are very attractive. There is a real lacking of understanding that there are two measures of capacity in storage; there is the capacity in the amount of data that the disk can store and then there is the capacity in the number of I/Os that a single disk can handle. The former has gone up massively whilst the latter has stayed the same and arguably with low-cost fibre and SATA drives actually got worse.
Some of our modelling suggests that even for fairly low-load workloads, we can only utilise 50% of a tray of SATA drives' useable capacity before we see performance dropping off quite significantly at the host level. Suddenly the attractive costs of SATA drives start to ramp up and head rapidly into FC drive territory. Dedupe on SATA drives may not make a huge amount of sense in a fair few cases; space isn't the problem, it's I/O. Perhaps EMC's I/O dedupe will come into play; who knows?
I'm sure all kind of funky games can be played to drive up the performance of SATA drives or at least the apparent performance but for the time being 'caveat emptor'; you may find those large SATA drives are not as cheap as you thought. And do you want to be the person who has to present reports showing incredibly low utilisation rates; perhaps we need two types of utilisation report, one showing the amount of space and the other showing the utilisation of I/O resource.
I'm interested 2.5" spindles as a potential way to drive up I/O densities whilst still utilising spinning rust; if I can fit 4×2.5" spindles into the space taken by a single 3.5" spindle, it might be an interesting intermediate between flash and huge 3.5" drives.
p.s I notice even Dave is getting in on the act.
In the media world though, the economics of slow disk can end up being a bit different. The odd discussion I used to find myself having was to try to persuade people that they would actually save money if they told me to spec. storage for more rather than less video.
The great thing about broadcast video is its size. Because a couple of thousand hours of archive quality video consumes such prodigious amounts of disk, you can find yourself in the happy situation where storing more is cheaper because storing less just means buying more spindles than you need for performance.
Of course, it helps if the video application uses the disks sensibly (i.e. sequentially – as should be the case for content) but the bottom line is that there’s not usually any reason to use FC outside of the NLE.
I guess I’m just making the point that it’s horses for courses. I’ve spent my fair share of life disabusing people of the idea that you can use SATA for anything and everything. In the video world though, at least until the cloud can deal with archives for us, SATA’s in its sweet spot.