Re: Re: SATA SAN Performance
- From: "John Fullbright [MVP]" <fullbrij@xxxxxxxxxxx>
- Date: Mon, 8 May 2006 10:36:24 -0700
It was from a previous post. Instead of retyping a similar respone, I tend
to reuse my posts.
I do write a semimonthly column in "Tech ONTAP". In the April issue I
talked about disater recovery.
"Al Mulnick" <amulnick_No_SPAM@xxxxxxxxxxx> wrote in message
news:%23lU2p88bGHA.3712@xxxxxxxxxxxxxxxxxxxxxxx
An earlier thread on the same day and possibly several others for the last
few years from John. He gets around quite a bit. :)
He really should consider writing technical articles for magazines as a
way to get the word out, but he may be too busy and just needs some more
help or something. But I'm starting to ramble, so......
Al
"Dusty C" <DustyC@xxxxxxxxxxxxxxxxxxxxxxxxx> wrote in message
news:C772687E-A40C-459F-AD28-5D62A1C666CE@xxxxxxxxxxxxxxxx
John,
Thanks for the info. Where is this a reprint from?
--Dusty
"John Fullbright [MVP]" wrote:
<REPRINT>
As an application, the IO profile for Exchange is high IO/Low to
intermediate space with a small block size (4K for the most part). In
today's enviroments, with client side caching, you'll see read/write
ratios
on the order of 2:1 or 3:1. The most important design consideration is
the
load you'll be placing on the storage subsystem. It's always best to
measure. Use the methodology for determining IOPS per user from
environmental data defined in "Optimizing storage performance for
Exchange
server 2003"
Once you've done this, there are a few tricky spots to be aware of:
1. Miscalulating the concurrency ratio. A small error in concurrency
ratio
can add up to a large performace deficit. Today's usage patterns can
change
tomorrow. It's best to assume worst case and go with 100%.
2. Miscalculating the read/write ratio. The read write ratio matters
because most storage systems today have a write penalty. The one
exception
I know of is Data ONTAP (the virtualization layer in every NetApp
storage
system) that has no write penalty. If your selected storage system does
have a write penalty, you have to know the read write ratio and add IO
capacity proportional to the penalty of your system and the read/write
ratio. A small change in the read write ratio can equate to a large
change
in IO demand and a performance deficit.
3. Failure to understand the relationship between IOPS/spindle and
response
time. The more IOPS you load a spindle with, the longer it takes for
each
IO request to be fulfilled (the IO latency increases. Sure, a 15K SCSI
spinde can have a maximum of 200 or so IOPS per spindle, but the
response
time will be on the order of 50ms. Exchange requires an average
response
time of 20ms with no peaks greater than 50ms lasting more than a few
seconds. Realistically, a 10K SCSI spindle can handle about 85 random
IOPS
with a 20ms target response time. A 15K SCSI spindle can do about 130.
A
10K SATA spindle can do about 45. A 7200 RPM SATA spindle is closer to
35.
4. Overstating the effectiveness of cache for random workloads.
5. Failing to factor in backend processes; Snapshot technology,
replication, backups, Exchange online maintenance, etc. All of these
place
load on the storage subsytem. Many times, this load is not visible from
the
host.
That rounds out the top 5 Storage Design Mistakes. Exchange on EMC
AX100
SATA... Given
the above, it should be clear why it's a bad idea.
</REPRINT>
Even at 3 IOPS/user, you're talking 150 DB IOPS, 15 LOG IOPS, and maybe
50
SMTP IOPS peak. Yes, the total is about the performace level that the
AX100
can do, but you'd be far better off with with local RAID 1/10. The cost
is
an order of magnitude less.
The AX is suitable for applications that require a large amout of space,
mostly sequential reads, and few writes. Archiving or CIFS might be a
good
fit, not Exchange.
"Dusty C" <DustyC@xxxxxxxxxxxxxxxxxxxxxxxxx> wrote in message
news:1A1B881D-9399-4580-85C8-8B4B432CA732@xxxxxxxxxxxxxxxx
Hello,
I am planning an Exchange 2003 deployment that will support 50 users.
We
are "medium users"; 50 incoming emails and 20 outgoing per user. Each
user
will probably have about 1GB mailbox.
Because of our existing infastructure the most appealing option
involves
iSCSI SAN. Particuarlly, I was interested the DELL/EMC AX100. But
this
enclosure uses SATA II drives (7200 RPM).
Is it a big no-no to deploy Exchange on SATA? With our number of
users
are
we going to pay a big performance price? Has anyone done this before?
Thanks for your time.
.
- References:
- Re: SATA SAN Performance
- From: John Fullbright [MVP]
- Re: SATA SAN Performance
- From: Dusty C
- OT: Re: SATA SAN Performance
- From: Al Mulnick
- Re: SATA SAN Performance
- Prev by Date: Re: Crashed Active Directory, how can I get my Exchange 2003 server back online?
- Next by Date: Re: New Server for Exchange
- Previous by thread: OT: Re: SATA SAN Performance
- Next by thread: Re: Uninstall 2000 server getting 8240 error
- Index(es):
Relevant Pages
|
