Exchange Server

Darrel · Guest

Hi

We are currently designing a system for around 1500 users but are unsure of
the hardware specs to use. The system is to be very resilient.

We are looking at the following

2 X DL380's SMTP and OWA in a DMZ, twin cpu's 4Gb ram. one to send mail the
other to recieve.2 mx records with costs pointing at both boxes.

2X DL380's Cluster Nodes, twin CPU's 4Gb of ram local mirrored drives for OS
and pagefile, HBA cards

1x MSA1000 SAN Array
300Gb drives RAID 0+1 config.built in fibre channel switch.

1X DL360 Bridgehead server for remote site connectivity

Does this sound viable and has any one used the MSA 1000 SAN ?

Thanks

Al Mulnick · Guest

Really? Just for SMTP and OWA?

How much traffic do you intend to deal with?

Or is this two separate sets of servers, one for internal and one for
external?

Either way, you won't want/need to cluster the SMTP/OWA functions in a MSCS
way. That's usually done with network load balancers and works quite well
in that configuration.

As for the other set (if that's it) again, what type of metrics can you
provide that give an indication of expected usage? Mailbox size,
throughput, etc?

Al

"Darrel" <Darrel@nynuxus.gotadsl.co.uk> wrote in message
news:O8EPvSSEFHA.1936@TK2MSFTNGP14.phx.gbl...

Brian Desmond [MVP] · Guest

4GB DL380s for SMTP & OWA bridgeheads? Thats a big load of box for not a lot
.... how many messages per second were you planning to transfer here? I
usually tend to think a couple of DL360s for XGW & OWA.

Haven't used the MSA1000 though was eyeballing it for a large deployment.
It's nice to scale up from. For your storage requirement I'd expect it to be
more than sufficient. I wouldn't consider it worth putting dual HBAs in the
380s given one of these SAN-in-a-box units. CHances are you're going to
route the fiber through the same duct, and then the switch is built into the
SAN unit. So, the only real point of failure here would be the HBA and
that's prolly not gonna fail during production.

Not sure whether the 380s would in general hold up without some more info on
the usage patterns of the users.

--
--Brian Desmond
Windows Server MVP
desmondb@payton.cps.k12.il.us

www.briandesmond.com

"Darrel" <Darrel@nynuxus.gotadsl.co.uk> wrote in message
news:O8EPvSSEFHA.1936@TK2MSFTNGP14.phx.gbl...

Darrel · Guest

Hi

The system is for around 1500 users we average per day around 12Gb of mail.
The front end servers will be being used by around 300 users for outlook web
access. One box for sending smtp mail and the other for recieving. The
reason the specs are so good for these 2 boxes is they are spare.

I was more concerned with the scalability of the back end, as I know we can
just use a hardware load balancer to add more OWA front ends should we need
them.

Also is the seperate bridgehead server a overkill as we only have 5 remotes
sites. The biggest having around 250 people that will send all external mail
through us.

Thanks for all your replies.

"Brian Desmond [MVP]" <desmondb@payton.cps.k12.il.us> wrote in message
news:OEXLYMkEFHA.1188@tk2msftngp13.phx.gbl...

Brian Desmond [MVP] · Guest

I think having the two OWAs and gateways load balanced would be an OK
solution here for mailfow and OWA access. The boxes are way more than enough
to do the job.

Hopefully Al will jump in on the backend stuff. You're dealing with more
mail volume than I do.

--
--Brian Desmond
Windows Server MVP
desmondb@payton.cps.k12.il.us

www.briandesmond.com

"Darrel" <Darrel@nynuxus.gotadsl.co.uk> wrote in message
news:uosIkZ0EFHA.3928@TK2MSFTNGP15.phx.gbl...

Darrel · Guest

Thanks for your replies

"Brian Desmond [MVP]" <desmondb@payton.cps.k12.il.us> wrote in message
news:uDTTkd8EFHA.4052@TK2MSFTNGP14.phx.gbl...

Al Mulnick · Guest

12 GB? Hmm..

So some quick math makes it look like 12884901888 bytes/24 hours or
536870912/ hour or .5 GB per hour.

500MB / 1500 users, would be about 33MB per user per hour (I'm sure there's
plenty of other traffic; just generalizing).

Really? What kind of files do they send around typically? What is the
average message size that you deal with a) from the external connections and
b) internally?

It makes a difference as to the layout of disk/application and horsepower
needed to process it.

As an example, the smaller/higher volume messages going by (say 50K msg size
at a million per 24 hours) would require one particular type of disk
configuration to handle without bottleneck, vs. larger/smaller volume
messages (one that has to handle 10MB msg size at 50,000 per 24 hours.)

Then there's the client usage patterns. Do all users use MAPI clients (you
answered that; 300 will use OWA via FE servers, but are there other client
types? POP/IMAP?) Do they all logon and use mail during the same hours?
Or are they spread out and will use them at different types of the day?
Maybe there are intersection points during the traffic when all users are on
but most of the time there's only 50% using at any given time? This all
makes a difference if you want to cut it close.

Likely you have enough horsepower in terms of processor/memory/network.
1500 users with 300 of them being OWA is usually not that bad for just
Exchange. If you plan to add other apps, you may want to be a little more
careful and test it more thoroughly to be sure it can handle the load. The
wildcard here is the disk layout and types of traffic patterns/sizes of
messages. If those calculations are correct, you'll put a lot more strain on
the storage subsystem than just about any other part of the system (network
will have to handle as well though right?). You'll want to pay special
attention to the spindle/application layout. Generally smaller/faster
spindles are going to be better vs. the 300GB spindles they sell with the
MSA1000.

Personally? My instinct tells me it's possible to do pretty comfortably. I
think you just need to be careful how you layout the I/O and you'll be fine
assuming you want some sort of normal performance metrics (normal to me is
no client-noticeable delays in action; delivery can be delyed a few
minutes). You may want to consider putting some of the IO in the server
itself by putting the queue/temp drives internally. The DL380 holds 6
drives, so you may consider putting queue/temp on a RAID 0+1 over 4
spindles, while the other two handle the OS/binaries/swap. Then put the
Exchange db's and logs etc on the SAN along with the cluster parts that you
can run on the SAN.

Does that help?

"Brian Desmond [MVP]" <desmondb@payton.cps.k12.il.us> wrote in message
news:uDTTkd8EFHA.4052@TK2MSFTNGP14.phx.gbl...

	Exchange Server Forum Index -> Design	All times are GMT
Page 1 of 1