A few basic concepts in disk sizing

re: A few basic concepts in disk sizing

Goran Husman — Mon, 11 Oct 2004 19:27:00 GMT

Excellent information, Nicole!
Thank you very much!!

/Göran
Exchange MVP

XP, Exchange, SMS, Tablet PC, New Design - Old Site

Windows Server Clustering — Tue, 12 Oct 2004 06:57:00 GMT

re: A few basic concepts in disk sizing

keith hanna — Tue, 12 Oct 2004 08:46:00 GMT

Excellent article!!!
Much appreciated :)

re: A few basic concepts in disk sizing

Karan — Tue, 12 Oct 2004 22:48:00 GMT

Thanks Nicole ... I'd read the Exchange Disk sizing whitepaper but this post does a nice job of summing it up quite nicely.

Exchange Disk Sizing (it's not just for cluster anymore!) :)

Evan's WebLog — Thu, 14 Oct 2004 07:16:00 GMT

Exchange Disk Sizing (it's not just for cluster anymore!)

Evan's WebLog — Thu, 14 Oct 2004 07:24:00 GMT

re: A few basic concepts in disk sizing

Lok — Thu, 14 Oct 2004 04:25:00 GMT

As mentioned, Exchange typically has a Read-to-Write (R:W) ratio of 3:1 or 2:1. is there any perfmon counter that we can calculate the real life read write ratio in the production environment for particular company?

re: A few basic concepts in disk sizing

Nicole Allen — Fri, 15 Oct 2004 21:45:00 GMT

To measure your R:W ratio, look at the ratio of LogicalDisk\Disk Reads/sec to LogicalDisk\Disk Writes/sec for the database drives. (You can look at the same counters in PhysicalDisk if you don't have the LogicalDisk counters enabled).

Note that for corporate servers with a large number of users (I am defining large here as 500 users or more), the R:W ratios are usually 3:1 or 2:1. However, servers that have fewer than 500 users will have lower R:W ratios (approaching 0:1 as the number of users decreases and as the amount of data in the database decreases). This is because for servers with few users, much of the user’s data will be in the database cache (in memory), so some of the read actions will be satisfied by data in memory. This reduces the number of read operations. Of course, all of the write operations will still have to be written to disk. Thus, the net effect of having a smaller number of users on the server is that the ratio of R:W goes down.

Below are the values for R:W ratio of 0:1.

R:W ratio \ Disk speed: 130 IOs per second 180 IOs per second
0:1 104 IOPS 144 IOPS

Table 2. Estimated maximum disk throughput for Raid 0+1 (or Raid 10):

R:W ratio \ Disk speed: 130 IOs per second 180 IOs per second
0:1 52 IOPS 72 IOPS

Table 3: Estimated maximum disk throughput for Raid 5:

R:W ratio \ Disk speed: 130 IOs per second 180 IOs per second
0:1 26 IOPS 36 IOPS

re: A few basic concepts in disk sizing

Steve McGovern — Wed, 20 Oct 2004 07:03:00 GMT

Do you have any metrics for write latency when SAN synchronous replication is enabled?

XP, Exchange, SMS, Tablet PC, New Design - Old Site

Windows Server Clustering — Wed, 20 Oct 2004 18:48:00 GMT

re: A few basic concepts in disk sizing

Barry — Wed, 20 Oct 2004 17:20:00 GMT

Question on disk latency... While deriving disk latency, if we use the counter "physical disk reads/sec, if we see an average value of 13.882, does that translate to latency of 72ms? (Use 1/13.882=0.072035...) Or I should not use that counter, or my interpretation is totally wrong?

Thanks,

re: A few basic concepts in disk sizing

Barry — Wed, 20 Oct 2004 21:56:00 GMT

ummm, found another reference that points out the counter should be "physical disk\average disk sec /read" and "physical disk\average disk sec/write". So, please ignore the last question. Instead, can you confirm these are the counters that you refer to?
Thanks,

re: A few basic concepts in disk sizing

Steve McGovern — Thu, 21 Oct 2004 07:05:00 GMT

Barry you are correct. Disk I/O latency can be identified by using the physical disk\average disk sec/write & physical disk\average disk sec/read counters. You should also monitor the MSExchangeIS\RPC Average Latency

My company is cheap!

Justin — Thu, 21 Oct 2004 08:11:00 GMT

Great Article, Im running servers with only 5 physical disks, and I dont see my company helping me out with anything more. What are the pro's and con's of seperating the logs\os\page\etc... via logical disk but leaving them on the same physical disk. Then putting the databases on a seperate physical disk?
Thanks,

re: A few basic concepts in disk sizing

Nicole Allen — Thu, 21 Oct 2004 15:34:00 GMT

Barry,
As Steve said, yes, you are correct. It's important to look at the seconds/write or seconds/read counters.

re: A few basic concepts in disk sizing

Nino Bilic — Thu, 21 Oct 2004 16:06:00 GMT

Justin,

There are close to no advantages to putting things on separate logical disks if they are on the same physical disk. One could say that doing this could help because of possibly less file level fragmentation, but one could also say that disk will do a lot more seeking on the same physical drive.

There are definitely advantages to putting the database onto a separate physical disk. Keeping the database random IO from the rest of the IO that could be sequential (transaction logs, other databases like SQL etc) is generally a good thing.

re: A few basic concepts in disk sizing

Justin — Thu, 21 Oct 2004 16:21:00 GMT

Thanks Nino, so would you say that it is more important to seperate the database or the logs? smtp\system\mta\and Indexes are all random I\O, like the databases. The Logs are sequential, so would it be more beneficial for me to just give the logs thier own physical disk?

re: A few basic concepts in disk sizing

Nino Bilic — Thu, 21 Oct 2004 16:34:00 GMT

I would definitely say the database. On your typical Exchange server, the database is responsible for about 90% of disk IO as opposed to the logs that are responsible for about 10% of disk IO. So - it is the database that will get disk IO bottlenecked much sooner than the transaction log drive.

re: A few basic concepts in disk sizing

Mike Salim — Thu, 21 Oct 2004 21:21:00 GMT

Hi,

Nice article. You have mentioned the word "database" a few times, could you please expand and define what is a "database" in the Exchange context. particularly the statement "all the databases" - how do I identify "all the databases" ?

Thanks!
Mike

re: A few basic concepts in disk sizing

Nino Bilic — Thu, 21 Oct 2004 21:26:00 GMT

Hi Mike,

"The database" in Exchange 200x world really means the .edb and .stm files. Those two files are really the same database - meaning, every database (mailbox or public folder) will have two database files, one .edb and one .stm. It is those database files that are referred as the "database".

re: A few basic concepts in disk sizing

Louis — Mon, 25 Oct 2004 21:02:00 GMT

Here is the basic info.
Server A
Time slice from 9 am to 6 pm.

Disk Transfers/Sec
Min/Avg/Max
34.400/339.052/1012.903

MsExchangeIS/Active User Count MsExchangeIS
Min/Avg/Max
519/920/1177

Disk Reads/sec
Min/Avg/Max
10.2/206.3/577.8

Disk Writes/sec
Min/Avg/Max
18.467/132.75/766.715

Total Mailboxes
563 on the server

Calculate your IOPS/user
I've seen a couple different options here.
Worst Case scenario
Max Disk Transfers/sec / Total Mailboxes on the server
1012/563
1.797

Max Disk Transfers/sec / Max Active user count from MSExchangeIS
1012/1177
.85

From the Exchange 2003 performance and scalability guide
IOPS/mailbox = (average disk transfer/sec) ÷ (number of mailboxes)
339/563
.602

and then
Avg Disk Transfers/sec / Avg Active user count from MSExchangeIS
339/920
.368

1. So which one is correct or should be used ?

The table Tables to lookup recommended maximum disk throughput per disk:
Table 1. Estimated maximum disk throughput for No Raid or Raid 0

2. Where do these numbers come from ? I understand the 180, but how do you reduce the achived throughput to 115 or 108 (using RAID 0+1)

3. The R:W ratio, is it as simple at looking at the averages on Disk Reads/sec and Disk Writes/Sec ?
So my ratio is roughly about 2:1 from the numbers listed above ?

Thanks this info is very good!

re: A few basic concepts in disk sizing

Nicole Allen — Tue, 26 Oct 2004 21:50:00 GMT

Steve,
Use the same metrics as a non-sync replications scenario (20ms), as measured from the LogicalDisk\Avg. Disk sec/transaction. What really matters to the users is the amount of latency the server is seeing - so you still want to keep Avg disk sec/transaction low.

-Nicole

re: A few basic concepts in disk sizing

David Wilhoit (kidego) Exchange MVP — Wed, 10 Nov 2004 02:04:00 GMT

Nicole,
I'm experiencing high latency with on my 5.5 servers, due to some poor configuration when the servers were originally built. Although write-back cache is enabled, with battery backup (SmartArray 5i with internal disks), and my avg disk sec/transactions are low, I think the cache is full, and it's not flushing out to disk fast enough. Disk usage% time can spike at over 1100, and over 2 hours in the A.M. it can average over 500%, with disk queue lengths hovering around 7. Only 4 spindles for the database in a RAID5, and I know that the server is pounded flat. Would increasing the read cache size help me out on this, or am I wasting time until I convince them to buy the SAN?

MSDTC and Exchange clusters

Evan's WebLog — Thu, 25 Nov 2004 17:47:00 GMT

MSDTC and Exchange clusters

Praveen Kumar's Blog — Thu, 25 Nov 2004 19:40:00 GMT

"Requesting data from the Exchange server..." or "Outlook is trying to retrieve data…" client messages

You Had Me At EHLO... — Wed, 25 May 2005 20:59:44 GMT

Based on the questions that we got on another post, it seemed appropriate to address the "Requesting...

'Requesting data from the Exchange server...' or 'Outlook is trying to retrieve data…' client messages

You Had Me At EHLO... — Wed, 25 May 2005 21:00:11 GMT

Based on the questions that we got on another post, it seemed appropriate to address the "Requesting...

'Requesting data from the Exchange server...' or 'Outlook is trying to retrieve data…' client messages

You Had Me At EHLO... — Wed, 25 May 2005 21:01:38 GMT

Based on the questions that we got on another post, it seemed appropriate to address the "Requesting...

Common Exchange 2003 Cluster Questions (Top 11 List) - Original Posted May 1, 2005

Cluster Help — Fri, 05 Aug 2005 23:05:31 GMT

A few basic concepts in disk sizing

re: A few basic concepts in disk sizing

XP, Exchange, SMS, Tablet PC, New Design - Old Site

re: A few basic concepts in disk sizing

re: A few basic concepts in disk sizing

Exchange Disk Sizing (it's not just for cluster anymore!) :)

Exchange Disk Sizing (it's not just for cluster anymore!)

re: A few basic concepts in disk sizing

re: A few basic concepts in disk sizing

re: A few basic concepts in disk sizing

XP, Exchange, SMS, Tablet PC, New Design - Old Site

re: A few basic concepts in disk sizing

re: A few basic concepts in disk sizing

re: A few basic concepts in disk sizing

My company is cheap!

re: A few basic concepts in disk sizing

re: A few basic concepts in disk sizing

re: A few basic concepts in disk sizing

re: A few basic concepts in disk sizing

re: A few basic concepts in disk sizing

re: A few basic concepts in disk sizing

re: A few basic concepts in disk sizing

re: A few basic concepts in disk sizing

re: A few basic concepts in disk sizing

MSDTC and Exchange clusters

MSDTC and Exchange clusters

&quot;Requesting data from the Exchange server...&quot; or &quot;Outlook is trying to retrieve data…&quot; client messages

'Requesting data from the Exchange server...' or 'Outlook is trying to retrieve data…' client messages

'Requesting data from the Exchange server...' or 'Outlook is trying to retrieve data…' client messages

Common Exchange 2003 Cluster Questions (Top 11 List) - Original Posted May 1, 2005

"Requesting data from the Exchange server..." or "Outlook is trying to retrieve data…" client messages