ADSM-L

Not All SAN Disks are Created Equal

2002-09-03 08:51:43
Subject: Not All SAN Disks are Created Equal
From: "Seay, Paul" <seay_pd AT NAPTHEON DOT COM>
To: ADSM-L AT VM.MARIST DOT EDU
Date: Mon, 2 Sep 2002 14:53:20 -0400
Folks, Lets be careful.

SANs are not bad for anything.  We must not lump the disk into the SAN
discussion, talk specifically about a disk subsystem being a part of a SAN
and its specific implementation.  What may be bad is the disk subsystem
implementation or how it has been connected to the SAN.  "Not all SAN
connected disks are created equal".  Unfortunately, like all open
technologies, you get to put the "better stuff" label on it eventhough it
may not be as good as previous technologies.

I would challenge anyone that can backup an 71GB database (actual backup
pages, database is 128GB) faster than we are (1.6 hours).  This is done in a
SAN, on an IBM 2105-F20, 36GB 10K disk.  The actual machine that holds the
database has 7.28TB of other applications on it.  I have some problems with
the way we set it up that we are working on.

It is correct that these new disk drives being so large and used native
without some kind of cache are bad for high activity random acess
applications, but sequential the data is packed even tighter and they
actually transfer data faster than a smaller drive at the same RPM.  So, a
15K 36GB drive will actually transfer data twice as fast as a 15K 18GB
drive.  That said, the most significant issue is device busy.  So, if the
device is not doing much it can have losts of data on it.

Take the ESS for example.  The average write response time on an ESS is
about 3ms or less.  I do not know of a hard drive that can write that fast.
It does striped IO on read and write.  It actually converts full stripe
sequential writes to RAID-3 making it faster than other RAID-1 solutions.
My general experience is that this technology is 5 to 20 times faster than
your standard 10K RPM technologies.  But, it can be more expensive.  There
is only one case where the ESS does not perform well.  That is applications
that are very very high intensity write, 10K+ IOPS (I/Os per second) random
write.   I have not seen any applications that do that, so I cannot speak to
that.  But, IBM just announced RAID-10 in their next product that has just
been released, 2105-800 that addresses this issue as well as being
generally, 2.5 times faster than the 2015-F20.

This list has some of the smartest people I have every discussed issues
with.  And, for that reason, just your name being attached to a post can
become a factor in the decision for a customer.

Paul D. Seay, Jr.
Technical Specialist
Naptheon Inc.
757-688-8180

<Prev in Thread] Current Thread [Next in Thread>
  • Not All SAN Disks are Created Equal, Seay, Paul <=