Re: Disck write latency > 50ms

Exmon.


"SW" <SW@xxxxxxxxxxxxxxxxxxxxxxxxx> wrote in message
news:49A5BD65-D062-45D9-B922-7D328F6FF229@xxxxxxxxxxxxxxxx
Thanks, is there any possible way to find out what user causes the
"slowness"? would be nice to see some thing like domain\username?

Also how can I tell how many users are connected to the Exchange 2003
server?

Thanks



"John Fullbright" wrote:

I use the figure 85 IOPS (IO operations per second) for a 10K spindle.
Some
folks use the value 130, which is derived by tahing the average seek time
plus the rotational latency, and dividing 1 second by the sum. This does
not account for the response time. Yes, a 10K spindle can handle more
than
130 IOPS, however at 130 IOPS, the response time is roughly 60 - 80ms.
85
IOPS is at a target 20ms IO.

Physical Disk - avg sec/write, sec/read, and sec/transaction tell you how
long IO operations against a given physical disk are taking. MOM uses
the
criteria from "Optimizing Storage for Exchange Server 2003" to alert;
"average 20ms writes with no peaks lasting more than a few seconds over
50ms".

Physical Disk - Transactions/sec, read/sec, write/sec tell you how many
IOs
and of what type a physical disk is servicing in a second.
Transactions/sec
translates to IOPS. You can use these three counters to calculate your
actual read/write ratio.

Database - Log Record Stalls/sec. This tells you haow many times the log
buffers filled to the high water mark, and a forced commit commenced.
During a forced commit, all cliet IO is quiesced. This value should not
exceed an average of 10 or peak of 100. Exceeding this value generally
means your log disk is inadequate.

Database - Page fault stalls/sec. This tells you how many times per
second
new pages were not able to be read into deatabase buffers in memory.
Generally this occurs when all pages are locked; either assembling
transactions or awaiting transfer to the log buffers. Any value greater
than zero is a sign of a serious IO bottleneck.



"SW" <SW@xxxxxxxxxxxxxxxxxxxxxxxxx> wrote in message
news:C4AD0940-C4CC-4105-96AE-C61C066D3C61@xxxxxxxxxxxxxxxx
At the moment I am only monitoring Physical Disk Avg. Disk sec/write
and
seing if it goes over 0.02 (20ms) and 0.05 (50ms). What esle should I
check
and what should the warning thresholds be?

"Vicnice" wrote:

Input and Output Operations per Second. Basically, anytime that
Exchange
has
to read or write that is an IOP. The standard is 2:1 (two reads one
write -
people are reading more email than they are usually sending).

With that being said an average IOP that a single disk can handle is
between
80-140 per second. That means that it can do 80 to 140 operations per
second.
When users exceed this, it is good bye good performance and HELLO
popup
city
(Outlook is waiting for Communication with the Exchange Server).


http://www.microsoft.com/technet/prodtechnol/exchange/guides/StoragePerformance/f4dc0eaf-3d21-4650-83b7-86526694f05a.mspx

There you go.

Honestly the calculations made no sense to me as I am a bit
"inexperienced"
(read:stupid) compared to the other people here. I would like an
explaination
to the formula as well.


"SW" wrote:

John that is quite amazing. We will go to RAID 10, we have enough
bays
to
put in extra disks, we are to but another RAID card for this too.
You
calculations are no doubt correct, can you explain how you got those
results,
I just want to explain to my bosses. What does IOPS stand for?

"John Fullbright" wrote:

Dump the RAID 5 and go RAID 10. The problem is write performance
on
Drive
F:

Let's assume a 4 disk RAID 5 vs. a 4 disk RAID 10 array. Further,
let's
assume 10K RPM SCSI drives and a 2:1 read/write ratio.

RAID 5:

Write perf = P*N'/4 = 85*3/4 = 63.75 IOPS
read perf = P*N' = 85*3 = 255 IOPS
mixed perf = 255*.66+63.75*.33 = 168.3 + 21 = 189.3 IOPS


RAID 10

write perf = P*N/2 = 85*4/2 = 170 IOPS
read perf = P*N = 85*4 = 340 IOPS
mixed perf = 340*.66 + 170*.33 = 224.4 + 56.1 = 280.5 IOPS


That's a 100% increase in write performance and a 55.5% increase
in
overall
performance for RAID 10 for the same number of spindles. I would
never
recommend RAID 5 for exchange.




"SW" <SW@xxxxxxxxxxxxxxxxxxxxxxxxx> wrote in message
news:E072839C-D358-40B3-B67D-CA89B83840A2@xxxxxxxxxxxxxxxx
It's the "writes" that are the problem, and when this is high
our
users
"nag". I use MOM 2005 with the Exchange Management Pack which
alerts me
when
the Physical Disck Avg. Disk sec/write is greater than 50ms for
10
minutes
(it polls every 60 seconds). Here is one I got earlier:

Severity: Warning
Status: New
Source: PhysicalDisk: Avg. Disk sec/Write: 0 F:
Name: Disk Write Latencies > 50 msec (F: Drive)
Description: High Disk Write Latencies for the past 10 minutes
PhysicalDisk: Avg. Disk sec/Write: 0 F: value =
6.41240062645155E-02.
The
average over last 10 samples is 0.064124.
Domain: DOMAIN
Agent: EXCHANGE2003
Time: 2/22/2006 09:57:00


"Neil Hobson [MVP]" wrote:

RAID5 probably isn't the best answer. Are you using MOM to
monitor this
server? Is it actually causing a problem to the users? In a
nutshell,
you
should look to size your server correctly regarding disk IOPS.
There are
a
number of articles on doing this, such as those by Rui Silva at
msexchange.org, or Microsoft's own paper called something like
Optimizing
Storage for Exchange 2003.

--
Neil Hobson
Exchange MVP


"SW" <SW@xxxxxxxxxxxxxxxxxxxxxxxxx> wrote in message
news:582F4999-F46F-4DAA-9B59-E60B3227B368@xxxxxxxxxxxxxxxx
What can we do about disk latency that goes over 5ms at
various
time of
the
day?

We have an Exchange 2030 Ent server (400 users), 2 DB's 80gb
and
10GB.

The high latency is on our F drive, he is our setup:

c: OS, Exchange program, paging file (RAID 1 channel 0)
e: Transaction logs (RAID 1 channel 1)
f: Databases (RAID 5 on a separate RAID card)

Usually the server runs below 20ms on the Avg. Disk sec/write
what we
hit
50ms at stages throughout the day, especially first thing
when
everyone
logs
in.











.

Relevant Pages

  • Re: Disck write latency > 50ms
    ... With that being said an average IOP that a single disk can handle is between ... (Outlook is waiting for Communication with the Exchange Server). ... we are to but another RAID card for this too. ...
    (microsoft.public.exchange.admin)
  • Re: Disck write latency > 50ms
    ... Also how can I tell how many users are connected to the Exchange 2003 server? ... IOPS is at a target 20ms IO. ... long IO operations against a given physical disk are taking. ... we are to but another RAID card for this too. ...
    (microsoft.public.exchange.admin)
  • RE: Restore Mailbox
    ... mailbox for the users from the old disk. ... it's highly not recommended and supported to restore Exchange ... server to your production environment because ...
    (microsoft.public.windows.server.sbs)
  • Re: Have I destroyed the Exchange 2003 active directory and can it be repaired
    ... After you've replaced the disk you might consider creating a new mailbox ... Took the disk out and put it in another server, ran chkdsk on it there, ... I then put the disk back in the exchange ... Had a look at the usual suspects (network settings etc) and all seems ...
    (microsoft.public.exchange.admin)
  • Re: Big Prob. E: is not accessible.There is not enough space on t
    ... The server is in Bruxelles actually and I am setting in Copenhagen (but ... There is not enough space on the disk. ... Exchange Services either!! ... logfiles to a disk where you have enough space. ...
    (microsoft.public.windows.server.general)