Re: Huge Av Disk Queue Length
- From: "John Fullbright [MVP]" <fjohn@donotspamnetappdotcom>
- Date: Sat, 6 Jan 2007 10:42:40 -0800
By itself, a disk queue length doesn't tell much about whats going on. The
important counters are the physical disk counters read/sec, write/sec,
sec/read and sec/write. Read and write per second tell you how many read
and write IOs are occurring on the drive in question. sec/read and
sec/write tell you how long each IO takes on average. If your average write
latency is above 20ms (.020) or you see spikes lasting more than a few
seconds over 50 ms (.050), then this is a sign of poor disk subsystem
performance according to "Optimizing Storage for Exchange Server 2003".
After you take your measurements, you can use this paper to determine how
many spindles you will need to support your measured IO.
When outlook grinds to a halt, generally this means RPC latency issues that
can be explored with the exmon utility on a per user basis. Unfortunately,
over 99% of the time the source of RPC latency is inadequate disk. With
Unity in the picture, root cause would certainly appear to be leaniung that
way. Integration of voice mail is a great feature, but it comss at a price;
increased disk utilization. If this wasn't accounted for in the initial
sizing of the system, you're bound to come up short. You may save yourself
a bit of time by focusing on disk bottlenecks.
<JustCol@xxxxxxxxxxxxxx> wrote in message
news:1168089581.053431.130840@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
Hi guys
We have a 2 cluster Exchange 2003 environment. Approx 150 mailboxes,
average 500MB, about 20 or so as large as 3GB. 4 storage groups, 4
mailbox stores per group, each store on it's own physical mirrored
disk, logs on seperate physical disks. Compaq servers, dual Xeon, 3GB
RAM, fast SCSI disks.
The problem we have is occasionaly Outlook will grind to a halt on
certain machines, particularly the PA's who have up to 10 shared
calenders open. I've set a performance monitor to check average disk
queue length on each of the disks. Most are fine and rarely go above
10 or so. One particular one has peaks up to several thousand which
can last up to a minute.
This particular storage group has sevarl of the gigabyte mailboxes on
it, also the mailbox for our Cisco VOIP unity system.
How can I monitor down to the mailbox level, or find out what the
problem is? Moving the unity mailbox is going to cost us a days
consulttancy which I'd rather not pay!
Thanks in advance.
Col
.
- Follow-Ups:
- Re: Huge Av Disk Queue Length
- From: Mark Arnold [MVP]
- Re: Huge Av Disk Queue Length
- References:
- Huge Av Disk Queue Length
- From: JustCol@xxxxxxxxxxxxxx
- Huge Av Disk Queue Length
- Prev by Date: MTA Message Bytes
- Next by Date: Re: Huge Av Disk Queue Length
- Previous by thread: Re: Huge Av Disk Queue Length
- Next by thread: Re: Huge Av Disk Queue Length
- Index(es):
Relevant Pages
|
