Re: Planning for a new storage array

  • From: "Mulnick, Al" <Al.Mulnick@xxxxxxxxxx>
  • To: "'[ExchangeList]'" <exchangelist@xxxxxxxxxxxxx>
  • Date: Tue, 23 Sep 2003 09:10:55 -0400

What you want to consider is the response times of the spindles when
requested to serve random data.  The biggest bang for you buck is to
separate out the log file I/O from the DB I/O.  The reason is the I/O type;
sequential Logfile I/O vs. random DB I/O.   The next step in the likelihood
of making it optimal category is to separate the queue files from the other
I/O.  Then continue with the swap file etc.

To really understand if you'll benefit from raid 1+0 for database logs, you
have to figure out if you're doing heavy writes or if you're going to use
more than 7 spindles.  If either is true, you'll very likely benefit from
1+0.  If not, Raid 5 is likely going to work fine.  Raid 5 has diminishing
returns when putting in more than 7 spindles to a container/volume. 

Raid 5 is great for many databases that don't require more spindles.  It's
fast for random reads which is what you want, but for scalability, you may
want to consider 1+0.  There are some performance trade-offs if you don't
scale 1+0 since there is overhead at the hardware level for the bifurcated
write op and if you're hardware isn't optimized to read from a 1+0 then you
may not see the same read performance that you want.  It depends on the
algorithm so check with SAN the vendor to find out what they do and what
advantages they can offer.

As always, be sure the cluster hardware is on the cluster HCL and set to the
proper settings to be a recommended configuration.


-----Original Message-----
From: kim [mailto:kim@xxxxxxxxx] 
Sent: Monday, September 22, 2003 5:25 PM
To: [ExchangeList]
Subject: [exchangelist] Re: Planning for a new storage array


I too will be moving my exchange data, to a fibre channel san device. While
i will be using RAID 5, it is mostly due to storage restrictions- i will
need another external drive box and another 7-14 scsi drives to use my
preferred RAID 5+1 (or ADG)...

i agree that with RAID 5 you should separate the containers (volumes- please
excuse the terminology- i have a unix background) by function- one for
exchange one for the db, one for the web pages, etc...

but why striping? what is it you are hoping to gain here? feel free to
contact my offline at kim@xxxxxxxxx should you feel it inappropriate to
continue on the list.


> Hi all,
> In few weeks I'll be moving entire Exchange server into new 
> enterprise-class storage array. Currently its working on single RAID 5 
> volume what for sure is not optimal. I'll have opportunity to make 
> final word what and where put and also what RAID should be used. I'm 
> thinking about RAID 0+1 volumes, however I'd like to use experts 
> experience how to divide exchange logs and databases to achive optimal 
> performance. Its clustered Exchange 2000 with about 1000 users, using 
> mostly Outlook XP, currently about 80 GB of mail data and about 20 GB 
> in public folders. Any input will be appreciated.
> Regards,
> Chris Ciapala
> System Admin
> RZE S.A.

List Archives:
Exchange Newsletters:
Exchange FAQ:
Other Internet Software Marketing Sites:
Leading Network Software Directory: No.1 ISA
Server Resource Site: Windows Security Resource
Site: Network Security Library: Windows 2000/NT Fax Solutions:
You are currently subscribed to this Discussion List as:
al.mulnick@xxxxxxxxxx To unsubscribe send a blank email to

Other related posts: