In days gone by, there were only two ways of us scaling up on storage performance, both of which were reliant on – and directly affected by – the number of disks. Adding more disks to an existing pair of controllers, and adding more controllers [with their own sets of disks]. The second option is/was suitable on the assumption that your dataset which requires the performance, is able to be divided between separate physical devices (unless you use LVM-style tools to stitch volumes together, however these approaches don’t really allow you to drive linear performance).

Continue reading →

 518 total views

The funny thing here is that achieving certain levels of storage array performance (be it IOPS, throughput, latency or a combination) can be relatively trivial if you put your mind to it (expect a future post re array performance benchmark testing); chuck a load of things together, smooth off some of the rough edges, step back and marvel at the wonders of your creation. However, what happens if you then decide you want a bit more…?

Continue reading →

 514 total views