The 1TB Disk, and storage econ
January 19th, 2007
Well it is here, the 1TB disk. I recall my second desktop PC after I came to HDS 11 years ago had an internal 1GB disk, and telling my kids how much stuff can be stored on that machine. So a 1,000 x improvement in about 8 years, amazing.
Capacity is great, it will be so for many home/consumer electronic devices. This drive will find its way into the data center over the next few months. Technology like this will have impact on $/GB (or should we now measure $/TB?). A colleague wrote me a while back and made these observations….
“… with typical workloads, we are often constrained not by disk drive capacity, but by ability to sustain IOPS and MB/s throughput. If workloads are reasonably active and have significant amounts of random write activity, as is very often the case in real life, then RAID-1 is cheaper than RAID-5 because RAID-1 generates fewer back end I/O operations. That means you need to buy fewer disks if you put the workload on RAID-1 parity groups, because you are IOPS constrained, not capacity constrained.
So really, when we are talking cost, if we are IOPS constrained then we need to look at $/IOPS and not $/MB. I would really like people to start looking at workloads to see if the constraint is throughput or capacity.”
His observations are spot-on. Capacity is good, but in the data center IOPS and availability bear a heavy demand. There is a chart (that I cannot seem to find) from a storage analyst that shows the relationship of performance and cost. It looked something like this (IOPS in red, Capacity in blue):
If you have access to the ‘real’ charts, please send me a copy.
Thanks for talking about this, David. Since capacity per device is increasing exponentially and throughput per devide is increasing linearly (if at all) storage vendors and storage customers may have to start purchasing performance rather than capacity.
$/IOPS would make a lot of sense.