How will the remote site users be using the local DC? Will
Exchange be local? Anything besides domain function on the DC? If no to both of
those items, a single RAID 1 will _probably_ be ok but that is shooting from the
hip knowing nothing about your environment or your directory so YMMV. As
Eric likes to say, for all cases you should be testing the perf and
configurations in the lab and verifying the actual perf matches what you are
shooting for.
As for the Exchange site, seriously, I would go with a 0+1.
I am not a fan of the 3 RAID-1 design for AD. It always overkill for the logs in
every case I have seen (again with the exception of the testing Eric has told me
about where he artificially pushes the write ops to the logs through the roof)
and even the OS usually doesn't do much to the disk. The DIT is what tends to
get pounded with Exchange. If you go to, I think (I am not a hardware guy), the
385 you get the 64 bit and can jam the machines with RAM and run x64 and cache
as much of the DIT as possible and then disk criticallity goes way down once you
have warmed the cache.
From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of Teo De Las Heras
Sent: Saturday, May 20, 2006 1:08 PM
To: [email protected]
Subject: Re: [ActiveDir] [OT] RAID 5 Best Practice
Joe,
What would you recommend for remote sites that are part of an AD domain
with 40,000 users (in terms of spindles)? If the remote site has 1,000 -
3,000 users would a DL360 be enough? Everything would be on a single RAID
1 partition.
We're consolidating the Exchange environment and we're going to create a
seperate site for Exchange 2003 and assign 4 domain controllers to that
site. Those DC's will be DL 380's with three (3) RAID 1 sets (OS, logs,
databases). Does this sound on the mark?
Teo
On 5/18/06, joe
<[EMAIL PROTECTED]>
wrote:
Classic Exchange type design. ;o)
For AD, I pretty generally recommend people do a single 0+1/10[1] first and
then 5 second and go with either because usually they don't have enough
slots for the disk internally to break it all up into a bunch of 1's and I
prefer the disk internal for AD and you want as many spindles in the set as
possible.
The good thing is that 0+1 will stand up to the IO (mostly DIT read) load
that you get out of even really busy DCs. I may change my thoughts after I
start seeing big x64 machines cruising along, haven't seen any yet in
customer sites. The log load on DCs is usually miniscule except in cases I
have heard of ~Eric testing some funky stuff in EEC and actually getting log
write ops into triple digits. Ditto for OS too unless you are doing a bunch
of other stuff on the DC.
For file sharing, I would consider 0+1 but 5 would be more likely since you
probably want/need the space more than the speed. File sharing doesn't
really beat the disks up relative to a busy DC even in large multi-thousand
user file servers I have seen. It is why most normal server admins really
have no clue what to look for in terms of IO load on servers but any
Exchange Admin worth anything is looking at that right away in a problem
situation and able to quote IOPS stats off the top of their head and know
what they can get from the underlying disk subsystem. Exchange disk configs
are critical.
Anyway, I don't have a problem with 5 for file servers. There is definitely
a hit on rebuild but you have to ask yourself how often you expect that and
whether or not it is acceptable that you take a hit when you are in that
mode. I consider the fault tolerance for emergencies, not something I have
to deal with weekly. If there are other benefits I want from 5 (say reduced
cost for the space) and having slower rebuild is acceptable then that is
perfectly fine. If you need something that is entirely transparent then you
look at other solutions and you start spending more money.
As for logically partitioning the underlying disk. Not sure what kind of
security gains you are expecting there. Nothing I can think of off the top
of my head. No perf gain except for the possible perf gains in doing a
volume chkdsk or backup/restore of individual volumes maybe. The
partitioning for logical separate of binaries in data can be a good thing.
Kind of nice to know that you absolutely need the D drive back but the C
could be a complete fresh rebuild.
joe
[1] Assuming they wouldn't consider a straight stripe set, recall DCs are
all duplicates and a big stripe set is going to be the fastest...
--
O'Reilly Active Directory Third Edition -
http://www.joeware.net/win/ad3e.htm
-----Original Message-----
From: [EMAIL PROTECTED]
[mailto:[EMAIL PROTECTED]] On Behalf Of Carlos Magalhaes
Sent: Thursday, May 18, 2006 2:02 PM
To: [email protected]
Subject: Re: [ActiveDir] [OT] RAID 5 Best Practice
I know this is not exactly the RAID 5 Best practices but this is how I
usually setup and recommend the customers to setup their disks (if they can
afford the hardware)
RAID1 for the OS
RAID1 for the logs
RAID0+1 for the database
Carlos
Brian Desmond wrote:
>
> I always do 12GB for C and the rest for D for 'Data'. I can format C
> and not worry about the Data.
>
> *Thanks,**
> *Brian Desmond**
>
> [EMAIL PROTECTED] <mailto:[EMAIL PROTECTED]>
>
> *c - 312.731.3132*
>
> ----------------------------------------------------------------------
> --
>
> *From:* [EMAIL PROTECTED]
> [mailto:[EMAIL PROTECTED]] *On Behalf Of *Timothy
> Foster
> *Sent:* Thursday, May 18, 2006 1:29 PM
> *To:* [email protected]
> *Subject:* RE: [ActiveDir] [OT] RAID 5 Best Practice
>
> Thanks, Brian. That makes sense.
>
> So if I have a 4 disk array on a single backplane, and given that I
> want the benefits of RAID 5, is there any argument for configuring
> more than one partition on the array? I realize that this is
> potentially too much of an open-ended question, but I'm curious :-).
> The basic premise is that this server would be a workhorse domain
> member/file server. Would one partition - C: - combined with carefully
> configured share and NTFS permissions provide adequate security? Or is
> it better to put the OS on C: and the shares on D: ? Or does the
> benefit of partitions lie somewhere else - for example, if I wanted to
> wipe C: and reinstall the OS without touching D: ? (I'm not sure if I
> like this idea, but as I mentioned, I'm curious...).
>
> Thanks,
>
> Tim
>
> ----------------------------------------------------------------------
> --
>
> *From:* [EMAIL PROTECTED]
> [mailto:[EMAIL PROTECTED]] *On Behalf Of *Brian
> Desmond
> *Sent:* Thursday, May 18, 2006 12:53 PM
> *To:* [email protected]
> *Subject:* RE: [ActiveDir] [OT] RAID 5 Best Practice
>
> Tim-
>
> It doesn't really matter. The RAID controller has no idea about the
> partition table. It just presents a LUN to the OS and the OS writes to it.
>
> *Thanks,**
> *Brian Desmond**
>
> [EMAIL PROTECTED] <mailto:[EMAIL PROTECTED]>
>
> *c - 312.731.3132*
>
> ----------------------------------------------------------------------
> --
>
> *From:* [EMAIL PROTECTED]
> [mailto:[EMAIL PROTECTED] ] *On Behalf Of *Timothy
> Foster
> *Sent:* Thursday, May 18, 2006 12:19 PM
> *To:* [email protected]
> *Subject:* [ActiveDir] [OT] RAID 5 Best Practice
>
> Using a RAID controller's configuration utility I can build and
> initialize a RAID 5 container. When installing the OS, I can, if I
> choose, create a partition. Is this a good or bad idea? In other
> words, if I partition RAID 5 container during the OS install will it
> make any difference if I ever need to replace a drive and rebuild the
> array? Will the partition table be recognized during the rebuild?
>
> Thanks for your input.
>
> Tim
>
List info : http://www.activedir.org/List.aspx
List FAQ : http://www.activedir.org/ListFAQ.aspx
List archive: http://www.mail-archive.com/activedir%40mail.activedir.org/
List info : http://www.activedir.org/List.aspx
List FAQ : http://www.activedir.org/ListFAQ.aspx
List archive: http://www.mail-archive.com/activedir%40mail.activedir.org/
