Storage Area Network as a tlug meeting topic

[Lennart Sorensen]

On Thu, Apr 30, 2009 at 12:01:47PM -0400, Christopher Browne wrote:
> Sounds interesting...
> 
> The one challenge I'd foresee is that there are two perspectives on
> this that are quite different in their slants:
> 
> a) Those that are doing "hard-core" SAN work, where maximizing numbers
> of spindles is worth paying extra money, and where multichannel
> FibreChannel is worthwhile.
> 
> I saw the following recently that explains that (though doesn't
> particulary explain *why*) expensive disks can give 10x better
> performance than the cheaper disks.

Well remember that to access data on a harddisk, the head has to move to
the right track (typical 3.5" SATA 7200rpm drive tends to do an average
seek in about 10ms or so, which is of course an eternity for a computer),
and then has to wait for the sector to rotate past the head (which takes
1/7200 of a second of course maximum, and on average 1/3600 of a second.)

The expensive disks running 15000rpm of course half the rotational
delay, and tend to be 2.5" as well, which means a shorter seek distance,
and often have better head controllers that move and settle faster,
and often get the average seek down to the 4 to 5ms range.  So for random
I/O those should be at least twice as fast as your typical desktop drive.

Now some people have found that the desktop drive can have better transfer
rates because of higher areal densities and longer tracks (because they
are 3.5" not 2.5"), and to get a very fast drive you take a modern 1TB
drive for example, and only use the first 10% or so of the disk, which
is at the outside (normally) and hence has the highest density of bits
per revolution, and by only using 10% of the disk, the head movement
now only covers about 5% of the normal distance, and the average seek
goes way down as a result.  Sure you waste 90% of the disk, but get get
a very fast disk as a result.

After all if you can reduce the average seek from 10ms to 3ms, and get
a higher transfer rate when you do reach the right part of the disk,
and you can buy such a 1TB drive (giving you 100GB effective disk)
for $130, that's not bad.  Compared to a 73GB 15000rpm disk which has
a lower transfer rate and similar average seek (but a lower rotational
delay, which doesn't matter nearly as much as the seek time), costing
probably $200 or $300.  Seems wasteful, but its an interesting method.
You effectively turn a 1TB 7200rpm 3.5" disk into a 100GB 1.5" drive
(except with the transfer rate of the outer part of the 3.5" drive).

I wonder if there will ever be common drives with multiple heads to
reduce the seek time overhead.

-- 
Len Sorensen
--
The Toronto Linux Users Group.      Meetings: http://gtalug.org/
TLUG requests: Linux topics, No HTML, wrap text below 80 columns
How to UNSUBSCRIBE: http://gtalug.org/wiki/Mailing_lists