Why are drive hardware and partitioning recommendations so difficult?
David Schrag
clients into better software and hardware, so there's gonna be a lot of new
server configuration going on. It doesn't seem that hard to figure out what
CPUs or RAM to recommend. Most folks are advocating dual Xeons and 1 GB,
respectively, for all but the tiniest of SBS installs. But I'm finding
little consistency in suggestions for how to configure and partition hard
drives.
I've seen recommendations for everything from a single big hard drive with a
single partition to dual RAID arrays (RAID-1 for the OS plus RAID-5 for the
data) and multiple partitions (for the purpose of separating OS, Exchange
info store, page file, log files, user data, SQL, SharePoint,
backup-to-disk, etc.).
The standard answer from MS is always "it depends on your environment," and
I can understand why they don't want to be in the business of hardware
recommendations. But why is there so little agreement among SBS consultants
out there?
With all the different opinions, it's very hard to distinguish between
choices that will really have an impact on the performance of the server and
the experience of the users vs. choices that may appeal to a techie's
sensibilities but have no real-world impact ... or maybe that only make a
difference once you start getting to a large number of users (and who's to
say what large is ... 25? 45?). It's also easy to get recommendations mixed
up and end up with benefits that cancel each other out (for example, you
might be recommended to put X on a different "drive" to increase
performance, but if that "drive" is just a partition in a RAID array then
the I/O might not actually be any faster).
With hardware costs relatively low, I think my clients will be happy to
spend a few hundred or maybe even a couple thousand dollars more to get the
"right" solution, but I also want to make sure I'm not wasting their money.
(I don't resell servers, so at least I have no financial incentive for them
to overbuy.)
If someone can point me to a resource like a decision-tree that walks you
through the options for hard drive array configurations and partitioning and
makes *practical* recommendations based on inputs about projected user
numbers and data sizing, that would be great. Otherwise, some advice on my
particular scenario would be helpful:
Typical client: 10-20 users, 2-4 GB Exchange data, 3-5 GB user and shared
data, running Premium edition
Proposed disk configuration: 3x36 GB drives in RAID-5:
* 12 GB C: partition for OS, apps, exchange and system log files, page file;
* 16 GB D: partition for Exchange information store;
* 44 GB E: partition for everything else.
Are there any changes I can make to this configuration that are really
"worth it?" And why is it so hard to get an answer to this question?
James Reather
> Lots of us are taking advantage of SBS2003 as a way of getting our
> SBS 4.5 clients into better software and hardware, so there's gonna
> be a lot of new server configuration going on. It doesn't seem that
> hard to figure out what CPUs or RAM to recommend. Most folks are
> advocating dual Xeons and 1 GB, respectively, for all but the tiniest
> of SBS installs. But I'm finding little consistency in suggestions
> for how to configure and partition hard drives.
I know this doesn't answer your question, ... but isn't there the same
problem with the spec of desktop PCs? IMHO many businesses would be happy
if you specified 1GHz/512MB Celeron workstations; but most end up buying
machines that cost *way* more, for very little extra benefit. *So* many
times I've looked at a spec for a machine, and ended up saying "double or
triple the RAM" while recommending downgrading the processor by several
steps - so the overall cost remains the same. Also, the machines seem to
run better as a result :-)
James
Edwin Garst
The disk configuration you propose looks fine. I think the theoretical
maximum for Exchange is 16GB for the mailboxes plus 16GB for public folders,
but 16GB is a lot for 20 users. Exchange defaults to a 200MB per mailbox
limit.
You asked about performance and the impact on the users. The main disk
issue I see is the Volume Shadow Service (VSS) backup that happens at 7AM
and noon by default. The backup space allocation is defined on the last
screen of the backup wizard.
I have reset VSS backup to happen at 4AM only. The process locks up the
disk and suspends all other activity until it finishes. I have daily alerts
that prove this. All of my experience is with IDE drives, so it may be that
SCSI drives don't do this. I suggest setting VSS to run on your data
partition only, not Exchange and not the OS. You should define the Users
Shared Folders on this partition.
I hate to admit it but this discussion reminds me of Linux OS planning!
Edwin Garst
"David Schrag" <david-...@schrag.net> wrote in message
news:ujQREO7p...@TK2MSFTNGP10.phx.gbl...
Ron Korszo
Glad you asked this question, because I was going to ask it soon. As I'm in
the process of designing my first SBS server.
I personally like the operating system on one drive, data on the other
(simple description for now). That's how my personal PC is setup and it's
only logical the disk I/O is distributed between two controllers / hard
drives. I actually have an old 1 GB drive dedicated for the swap file, and
indeed noticed an increase in performance after changing to this
configuration. Even though it's an old slow drive.
With this concept, I believe two RAID arrays make the most sense. For the
simple sake of having two arrays accessing completely different files
simultaneously. We know this happens in the real world, so why not
distribute the load between two systems which can work simultaneously.
Now what I don't understand, is why recommendations are made to make
multiple partitions on the same drive or array. The only advantage I can see
is to minimize fragmentation. Wish someone would post the advantage of
splitting a drive into three or more partitions, this does not make sense to
me. And as in the day of partitition size limitations, downstream ending up
with
one partition that's empty and another that's full. For this reason I've
gotten into the habit of using all available space on the drive or array and
use subdirectories. And quite frankly, have never had a problem with this
concept. Takes longer to defrag yes, but so what? One thing is for sure, all
drive space is available for all subdirectories to use.
I've just received a Dell Poweredge 2600 for my client:
Dual Xeon processors, 2GB of memory
Split Backplane:
RAID 5 - 3x18 15k RPM drives
RAID 5 - 3x73 15k RPM drives
How to partition?
1st Array
7 GB C: Operating System, Apps
29 GB D: Apps, Log Files
2nd Array
100GB E: Data, Everything Else
40 GB F: Client Apps (Distribution), Virtual CD store, Everything Else
6 GB G: Swap File
Partitioning the swap file by itself makes sense, everything else, not
sure??? Maybe cluster size...Smaller drive searches file system faster?
Would be nice for someone to shed some concrete logic / facts that makes
sense on
why to create so many partitions as some people recommend...
Any Takers?
"David Schrag" <david-...@schrag.net> wrote in message
news:ujQREO7p...@TK2MSFTNGP10.phx.gbl...
SuperGumby
> Proposed disk configuration: 3x36 GB drives in RAID-5:
> * 12 GB C: partition for OS, apps, exchange and system log files, page
file;
I'll agree with this with minor reservations.
Why keep the exchange logs on your OS partition? If some virus sends you
37GB of traffic in one day your system is unbootable. You have a single RAID
array and the main advantage of seperating exchange log files from the
database is to put the logfiles on a sequential file system (FAT) on
seperate physical drives so that performance can be improved. However we're
talking about performance of a TINY exchange (<75 users) and less than 16GB
over the whole life of the server.
> * 16 GB D: partition for Exchange information store;
OK, so your Exchange is approaching 16GB. What happens when your users
archive 2GB of mail and Exch needs some more space to 'shuffle' the pages in
order to accept today's 2GB of mail. What I'm eluding to I suppose is that
though the limit on an Exchange Server Standard store is 16GB there may be
times when the filespace required by the store exceeds this amount. You've
also gotta consider that NTFS file access performance drops as the drive
approaches high 'percentage used' amounts. Some people suggest NTFS suffers
when slack space approaches 20% but someone recently suggested to me that
30% was more realistic, 20% was obvious, and 10% was pathetic. So your 16GB
maximum exchange database needs at least 19GB and would probably be better
residing on a drive with a further 21GB so you can defrag it occasionally
(exch defrag requires more space than the database to work). So there's
50GB+ for your 16GB exchange, most of which will be used once every 3 years.
AND, as these 'drives' are actually logical drives on the same volume of a
single RAID array and are all partitioned NTFS you lose MOST of the
advantages.
I'll happily point out that I'm not sure whether the Exch Server Standard
Edition is limited to 16GB of content or 16GB of filespace.
> * 44 GB E: partition for everything else.
Yep. DATA=REST.
---------------------
I'll still go with a three partition model.
PART1 = OS and anything not easily moved off it + swapspace + slackspace +
projected upgrades
PART2 = anything considered fairly static. Your program install points, your
application push install point.
PART3 = DATA
and there's a good case for creating the DATA partition between the OS and
static partitions so that that static stuff which gets accessed infrequently
doesn't degrade performance.
HTH :-)
David Schrag
Celerons. A year or two ago I was recommending Celerons for clients until I
realized that P4's with bigger caches cost only about $50 more per computer.
I decided that it was worth the extra money to not have to wonder whether a
different chip would make a difference in performance.
"James Reather" <james...@reather.com> wrote: in message
news:ezh3sp7p...@TK2MSFTNGP09.phx.gbl...
David Schrag
GB partition for Exchange was probably a red herring, since people will
think I chose 16 GB to match the store limit (which as you point out is
really 32 GB). But I was really just thinking "I want a big partition for
Exchange to allow growth." Maybe I should go up to 20 GB for Exchange and
down to 40 GB for the data partition.
Enabling VSS on only the data partition makes sense (and that's a good
argument for not putting everything on a single partition whether or not you
have RAID arrays).
"Edwin Garst" <egarst...@online.epconsystems.com> wrote in message
news:uaDiio9p...@TK2MSFTNGP11.phx.gbl...
David Schrag
to said this was really unnecessary.
I know you're still in the process of building your first SBS 2003 server,
but do you have any sense as to the street-level implications of having that
second array? Will users be able to do anything noticeably faster with that
configuration?
"Ron Korszo" <anon...@msnews.microsoft.com> wrote in message
news:%23xIE0X$pDHA...@TK2MSFTNGP12.phx.gbl...
David Schrag
12 GB C: for OS, apps, pagefile
40 GB D: for data, including Exchange, SharePoint, and regular old users
folders
20 GB E: for CD images, downloaded program installs/drivers, and other
static stuff
Yeah?
"SuperGumby" <n...@your.nellie> wrote in message
news:eZUNsFEq...@tk2msftngp13.phx.gbl...
James Reather
> Yes, it's the same problem, but I don't know that P4's cost "way
> more" than Celerons. A year or two ago I was recommending Celerons
> for clients until I realized that P4's with bigger caches cost only
> about $50 more per computer. I decided that it was worth the extra
> money to not have to wonder whether a different chip would make a
> difference in performance.
I'm talking about the people who demand the latest P4 3.06GHz system
(processor's street price UK £210) because they've heard in the pub it will
run Outlook at least three times faster than their workmate's 1GHz PC.
FWIW I built myself a micro PC with an 800MHz VIA C3 processor (current UK
street price £11) with 512MB RAM, 80GB hdd, it makes a great client for my
test SBS.... go ahead and tell me the P4 chip is really nineteen times (210
/ 11) better for the average small business :-)
James
SuperGumby
I consider the \clientapps folder as relatively static. Also much of the
stuff under \programfiles, we service pack it occassionally and hotfix it as
necessary but by and large the content doesn't change, e:\exchsrvr, e:\isa,
e:\faxsvr etc... makes sense to me. Note however that one of the reasons why
I developed this last opinion was in response to the SFN problem associated
with disaster recovery, which is lessened if not abolished in W2K3. OTOH it
was 7 or so years ago that I commented 'what's this "\program files" dir and
why the hey does every application want to install into my OS partition?',
but this was more due to limitations on partition size.
"David Schrag" <david-...@schrag.net> wrote in message
news:eC0O4jIq...@TK2MSFTNGP12.phx.gbl...
Ron Korszo
No, I don't think user's will see a difference using the 2nd array with
current usage. But, for how much money is involved, I believe its extends
server life for expandability in the future. Insurance, just in case...If
database app(s) or anything else are required downstream, it will be no
problem. FWIW...
RK
"David Schrag" <david-...@schrag.net> wrote in message
news:uCehThI...@TK2MSFTNGP11.phx.gbl...
Ron Korszo
Also, I believe *more* smaller drives are better than *less* larger drives.
In other words, I'd probably go with 6x18GB, instead of 3x36GB, and utilize
the included 2nd RAID channel...Definately not going to hurt
anything....again FWIW.
RK
"David Schrag" <david-...@schrag.net> wrote in message
news:uCehThI...@TK2MSFTNGP11.phx.gbl...
AT
AT
"SuperGumby" <n...@your.nellie> wrote in message
news:%232YPeuJ...@TK2MSFTNGP12.phx.gbl...
SuperGumby
"AT" <torgny@@direcway.com> wrote in message
news:tJgsb.9998$6c3....@newsread1.news.pas.earthlink.net...
AT
one of the hard disks?
AT
"SuperGumby" <n...@your.nellie> wrote in message
news:uM1FAOMq...@tk2msftngp13.phx.gbl...
Les Connor [SBS MVP]
--
Les Connor [SBS MVP]
-------------------------------------
SBS Rocks !
"AT" <torgny@@direcway.com> wrote in message
news:Wdjsb.10140$6c3....@newsread1.news.pas.earthlink.net...
jimbeh...@mindspring.com
than a few boxes. I just install a new hard drive and the Raid 1
starts to rebuild automatically. The Raid 5 sometimes rebuilds
automatically with no user intervention if you have setup a hot spare
correctly. If you do not set it up correctly the server squeels like a
pig until you replace the dead hard drive. Then it starts to rebuild.
No problems booting in those hardware raid situations.
There is a mention in software raid where you have to modify the
boot.ini so it can see the second disk if you are using software raid
1.
"Les Connor [SBS MVP]" <les.c...@cfiveDEL.ca> wrote:
>Hardware raid.
Jim B. SBS MVP
remove the mvp to send email
Jerry Dubuke
I added the Raid 1 after the initial install. Was pleasantly surprised on a
subsequent reboot of the server in the fact that it now offered a boot menu
which was not there previously. The second option of the boot menu was.....
BOOT FROM SECONDARY disk!
It automatically modified the boot.ini after it realized I had a raid 1
system. *VERY* cool addition!
Jerry
<jimbeh...@mindspring.com> wrote in message
news:bgd4rv89mkjebsun0...@4ax.com...
AT
used RAID5 for the OS of two reasons.
1. Best Practices tell me not to.
2. MicroSoft do not support OS on RAID5.
If I have the OS on RAID5 and one of the disks is malfunctioning I expect my
OS to stop responding and I probably need a reboot for it. Will the
Controller Card (from its BIOS) then be able to take care of mounting my hot
spare and start rebuilding my RAID5 without having an OS up running?
Normally that is taken care of by my management software.
AT
"Jerry Dubuke" <jdu...@not.gpdservices.com> wrote in message
news:ezZpwYSq...@TK2MSFTNGP09.phx.gbl...
SuperGumby
> I understand that it takes a HardWare RAID to make it possible. I have
never
> used RAID5 for the OS of two reasons.
> 1. Best Practices tell me not to.
whose 'best practices' ?
> 2. MicroSoft do not support OS on RAID5.
BALDERDASH
> If I have the OS on RAID5 and one of the disks is malfunctioning I expect
my
> OS to stop responding and I probably need a reboot for it. Will the
> Controller Card (from its BIOS) then be able to take care of mounting my
hot
> spare and start rebuilding my RAID5 without having an OS up running?
No to the first part and yes to the 2nd.
> Normally that is taken care of by my management software.
Actually, until recently GUI's had much less control than could be achieved
from the RAID bios.
Kevin Weilbacher
on a Raid-5 disk subsystem is simply outdated. Frankly, all of my servers
are configured using only RAID-5 disks. I partition the Raid-5 into three
segments: one for O/S (usually 10GB), one for Exchange, and one for all
other applications.
-kw
"AT" <torgny@@direcway.com> wrote in message
news:uyWBs3Yq...@tk2msftngp13.phx.gbl...
AT
If I understand you correct I will not be able to use my HotSpare with this
setup? How would I find the malfunctioning hard disk without the management
software? I realise that some problems is easy to find. Dead HD, bad heads
e.t.c. But how can I find the one with write or read problems in for exampel
a 5disk RAID5?
I am thankful for any information that can make me feel better about OS on
RAID5.
AT
"SuperGumby" <n...@your.nellie> wrote in message
news:%23WTOQBZ...@TK2MSFTNGP10.phx.gbl...
Daryl Maunder
> If you say MS support OS on RAID5 I accept that.
> If I understand you correct I will not be able to use my HotSpare with
this
> setup? How would I find the malfunctioning hard disk without the
management
> software? I realise that some problems is easy to find. Dead HD, bad heads
> e.t.c. But how can I find the one with write or read problems in for
exampel
> a 5disk RAID5?
I use Compaq RAID controllers in Compaq server, using the supplied
management agents. If a disk fails, or is having errors, the agent will send
me an email to say so.
AT
the RAID5. Then you do not have an available management agent either. Or do
I miss something?
AT
"Daryl Maunder" <dmaunde...@midnightoil.nospam.com.nospam.au> wrote in
message news:OFeD0Kbq...@tk2msftngp13.phx.gbl...
SuperGumby
hotspare is that if the the RAID controller decides a member of the existing
array has had sufficient errors that it should be marked as bad the HS can
automagically take over the failing drives role. You should need to do
nothing in whatever GUI is provided. When the system alerts you (or you
notice) that a member of the array has failed your GUI should be providing
information about how it is incorporating the hotspare into the array,
otherwise it's a coldspare, sitting there waiting for user intervention to
save the system.
How do you find out which drive is kaput? Either in your windows GUI or in
the RAID adapter's BIOS.
A simple example:
Adaptec 2120S SCSI RAID. Let's imagine:
A three drive RAID5 array with hotspare. The array is drives 0, 1 and 2 on
it's single channel. Drive0 dies while the system is operating. What
happens?
As far as the OS is concerned, nothing. It could see one drive, the array,
before the drive failure and it continues to see one drive. Meanwhile, the
adapter is busy incorporating the HS into the array. Things may slow down
for a while. An alarm is raised, so you inspect the server. As the OS is
still happily OS'ing along you may use your Windows GUI to inspect the raid
configuration. BUT DAMN, the array manager isn't available (this is almost
impossible these days), WHAT DO I DO?
Restart the box and when it gets to the SCSI BIOS screen it will halt, for a
time, indicating a drive is offline (0:0) and ask if you wish to continue.
Only one drive is bad so you allow it to restart and, walk to shop, purchase
replacement drive, back to server. Meanwhile the hotspare has been
incorporated into the array. At some convenient time you log out all the
users (which I forgot to mention you do before the previous restart) and
swap dead0:0 for new0:6, and fire her up. Whether you enter the SCSI RAID
bios to mark the new drive as a hotspare or use the fancy windows GUI is a
matter of choice.
feet up, beer, cigar.
"AT" <torgny@@direcway.com> wrote in message
news:%23MDOMWa...@TK2MSFTNGP12.phx.gbl...
Daryl Maunder
> I Understand that. But the question is how you do that if you have the OS
on
> the RAID5. Then you do not have an available management agent either. Or
do
> I miss something?
I think I am missing something. If a disk on the RAID5 array is having
problems, or fails, whether the OS is on it or not, it will still be
running. So the alert will get sent by the management agent. If the
controller dies, or 2 disks die, then the server will crash and you wont get
a notification, but then someone is likely to notice.
AT
This is probabl;y where you get me confused.
No to the first part and yes to the 2nd.
I take that as
Will the Controller Card (from its BIOS) then be able to take care of
mounting my hot spare NO
start rebuilding my RAID5 without having an OS up running YES.
You tell me to rely on the GUI? How if I don't get a GUI up? How can the OS
keep running on a system with only 2/3 of it's files available. I expect it
to hang with a reboot as a result. That would be impossible without
rebuilding the faulty disk fiirst. Are you telling me that I can loose 1/3
of the OS and 1/3 of the management program and everything will still work
just fine? Are you telling me that the SCSI BIOS can determine which HD has
a read problem?Perhaps I don't need a HotSpare at all?
I'm sorry.....you make it sound like nothing can ever go wrong with a RAID5
system.
AT
"SuperGumby" <n...@your.nellie> wrote in message
news:Orn2UYc...@TK2MSFTNGP10.phx.gbl...
SuperGumby
No to the first sentence and yes to the 2nd.
Sentence 1 > If I have the OS on RAID5 and one of the disks is
malfunctioning I expect my OS to stop responding and I probably need a
reboot for it.
Sentence 2> Will theController Card (from its BIOS) then be able to take
care of mounting my hot spare and start rebuilding my RAID5 without having
an OS up running?
A 3 disk RAID5 does not store 1/3 of the files on each disk. Typically, half
of each file is stored on disk0, the other half on disk1, and redundancy
information on disk2. If any one disk fails there is enough information in
the remaining pair to maintain the integrity of all files. FOR PETE'S SAKE,
that's the whole purpose.
"AT" <torgny@@direcway.com> wrote in message
news:1Q0tb.205$sb4...@newsread2.news.pas.earthlink.net...
David Schrag
"AT" <torgny@@direcway.com> wrote in message
news:1Q0tb.205$sb4...@newsread2.news.pas.earthlink.net...
AT
have been diskussing so far?
AT
"David Schrag" <david-...@schrag.net> wrote in message
news:%23jhHz0r...@TK2MSFTNGP10.phx.gbl...
AT
1. My OS will keep on working if any one of my HDs is failing and I will
still be able to use the management software.
2. If I need to reboot the OS will still work and be able to boot on any
two disks.
I have to figure this out to understand how this is possible.
AT
"SuperGumby" <n...@your.nellie> wrote in message
news:%23biKkfp...@tk2msftngp13.phx.gbl...
SuperGumby
> You tell me that
> 1. My OS will keep on working if any one of my HDs is failing and I
will
> still be able to use the management software.
Yes
> 2. If I need to reboot the OS will still work and be able to boot on
any
> two disks.
Yes, this is the WHOLE point of RAID5. Lose any single member of a RAID5
array and you lose no information. HOPEFULLY, you have a hotspare which
automatically replaces the failed drive but this is not necessary, just
desirable.
If you were to lose a 2nd drive before the hotspare, or replacement drive,
was incorporated into the array you'd be in all sorts of strife.
> I have to figure this out to understand how this is possible.
Yes
David Schrag
http://redundantarrayofindependentdisks.com/.
"AT" <torgny@@direcway.com> wrote in message
news:Pm%tb.4490$n56....@newsread1.news.pas.earthlink.net...
Jim Behning
setups. Neither one did I set up correctly to incorporate the hot
spare. The server worked fine, rebooted fine but worked slow while it
incorporated the new drive to replace the dead drive.
They will also boot up fine sometimes in Hardware Raid 1 and in
software Raid 1. That is if the second drive is the one that fails. I
have had this happen on a few servers also. No one knew a hard drive
had failed but I noticed it during a servicepack reboot. Installed new
hard drive and repaired mirror.
"SuperGumby" <n...@your.nellie> wrote:
Jim B. SBS MVP
AT
have been away from the RAIDs too long time to realise the evolution that
has happened.
AT
"Jim Behning" <jimbeh...@mindspring.com> wrote in message
news:ec5jrvoi8c1au74ub...@4ax.com...