Re: Exchange Hardware Question
- From: "John Fullbright [MVP]" <fjohn@donotspamnetappdotcom>
- Date: Fri, 12 Jan 2007 17:45:38 -0800
Store = RAID 1 or RAID 10 depending on the usage.
1. The dell, with three drives in a RAID 5 configuration, implies the OS
was built using the dell config utility. This is a worst case for
performance. When an OS is built using such a utility, the file system
begins life as FAT and is converted to NTFS. This results in an allocation
unit size of 512 bytes. This means a write to the page file (assumes page
file on the os drive which is typical in small deployments). This means
every 4K write to the page file will result in a split IO and 8 IOs to the
drive. That goes for anything else on the OS drive. You can see this by
looking at the perfmon physical disk counter split IOs per second. I have
one customer who did a study comparing this default configuration to a
mirror and manually building the OS with a 4K (the default) allocation unit
size. The measured performance increase was 400%.
2. You don't mention the type of drive, but with a perc it will be SCSI.
For a 20ms response time you can use a P value of 85 for 10K drives and 130
for 15K drives. For RAID 5 read perf = P* (n-1) where n is the number of
spindles and P is the IOPS/spindle of the drive type you are using. Write
perf =P *(n-1)/4. For exchange you have to apply the read:write ratio to
determine the mixed value. assuming 10K drives and a 3 spindle array, read
perf = 170 IOPS and write perf - 42.5 IOPS. With a 1:1 read:write ratio,the
mixed perf is 106 IOPS.
For a 2 drive RAID 1 mirror, read = P*N and write = P*N/2. With 10K
spindles write perf = 170, read perf = 85 and mixed perf = 127.5 IOPS.
That's a 20% performance increase over RAID 5 with a third less spindles.
I suppose this is why the MS paper "Optimizing Storage for Exchange Server
2003" states "In general, Raid-5 does not provide the best trade-off between
reliability/availability and performance". Considering the math, I'd call
that an understatement. I would not recommend RAID 5 except in extremely
small Exchange 2003 deployments (under 50 IOPS load).
http://technet.microsoft.com/en-us/library/bb124875.aspx
You don't state the number of users or your IOPS requirement, however the
configuration Bharat talks about provides:
85 IOPS for the OS
85 IOPS for the logs
127.5 IOPS for the databases
If you move to 6 spindes and replace the mirror with a RAID 10, then
85 IOPS for the OS
85 IOPS for the logs
255 IOPS for the databases
This configuration would require 6 drives and cover many small
organizations. Moving to 15K drives in RAID 10 would give you
130 IOPS for the OS
130 IOPS for the logs
540 IOPS for the databases
In RAID 1 you would get
130 IOPS for the OS
130 IOPS for the logs
270 IOPS for the databases
You could even go three mirrors to provide fault tolerence for all the
volumes and, in 6 15K spindles, have
270 IOPS for the OS
270 IOPS for the logs
270 IOPS for the databases.
John
"Bharat Suneja [MVP]" <bharatsuneja@xxxxxxxxxxx> wrote in message
news:efYhVZpNHHA.1252@xxxxxxxxxxxxxxxxxxxxxxx
Before you start - how many users? What's the usage? Disk I/O required?
Generally speaking, no... but if you want to optimize (without knowing
more details.. ), the rule of thumb **given the number of drives you have
=6**:
OS: RAID1 mirror
Transaction Logs: RAID1 mirror
Store (in this situation): RAID1 mirror
--
Bharat Suneja
MVP - Exchange
www.zenprise.com
NEW blog location:
www.exchangepedia.com/blog
----------------------------------------------
"-=gu=-" <gu@xxxxxxxxxxxxxxxxxxxxxxxxx> wrote in message
news:D84C18AC-F2A9-405A-8549-765C9D0765AB@xxxxxxxxxxxxxxxx
Hello,
We are a single site running Exchange 2003 on 2003 server.
I am not too far from running out of disk and would appreciate your
thoughts
on migrating to a different server. Currently, my priv1.edb and stm are
totaling close to 40 GB and I have only 15 GB Free. The existing server
is
ancient.
The company is going to grow (almost double) in 2007.
I have a decent server waiting in the wings doing nothing which I can
use.
It's a Dell PowerEdge 2850 that currently has a PERC raid contoller and
(3)
147 GB drives. There are an additional 3 empty bays, which I could also
fill
with 147 GB drives. Under a RAID5 configuration, I should get 700+ GB of
disk.
I know that the optimal configuration of Exchange would be to use
separate
redundant (not partitioned) drives for: system \ pagefile \ logs and
exchange.
So my question is...
If I buy the 3 additional drives and put them all on a single (or even
two
RAID5s), am I asking for trouble in terms of a disk I\O bottleneck?
Thank you all very much for your responses.
-=gu=-
.
- References:
- Re: Exchange Hardware Question
- From: Bharat Suneja [MVP]
- Re: Exchange Hardware Question
- Prev by Date: Re: Exchange Hardware Question
- Next by Date: Re: Exchange 2007 transition from Exchange 2000
- Previous by thread: Re: Exchange Hardware Question
- Index(es):
Relevant Pages
|
