On Mon, Aug 30, 1999 at 10:19:02PM -0700, Tim Moore wrote:
> I have done some performance studies at work. As everyone is doing we
> are looking at the trend of 9GB -> 18GB -> 36GB -> 72GB -> xxxGB driven
> by increased demand for data storage. The measured trend is an overall
> performance decrease. My guess is that major drive manufacturers will
> not be phasing out smaller capacity/higher performing drives any time
> soon. Rather offer a large capacity line for the data hungry, and a
> fast performance line for the critical response oriented.
Well, it depends on what you mean by performance.
I claim the trend is towards increased performance for
sustained sequential I/O. This is because sometimes that's what
matters, it almost comes "for free" with increased density,
and it gives good marketing numbers.
However, for many purposes, sustained sequential I/O rates are not
all that interesting. When you look at random I/O performance, the
trend is for slight increases (largely driven by the increases
in rotational speed, from 5400 -> 10000rpm), but only when you
look at accessing a single drive. If you have many drives,
then the trend is towards decreased random I/O performance,
simply because you have more data on each drive, giving a reduced
net seeking ability per megabyte of data. In the limit, you replace
your big old disk farm with a single high-capacity drive, but it can only
do one seek at a time.
If you are trying to get maximum TPC benchmark results, you probably
want as many drives as possible. So you might want to buy lower
capacity versions of the latest big capacity drives, or just get a lot
of big drives and plan on underutilizing them. Doing the latter has the
added advantage that you can discard the inner (slower) cylinders,
thus ending up with effectively faster small drives.
And, what the heck, the drives are getting so incredibly cheap ... :-)
Jan Edler
NEC Research Institute
