[THIN] Re: SMB Current Commands and SMB tuning
- From: Dan Dill <DanD@xxxxxxxxxx>
- To: "thin@xxxxxxxxxxxxx" <thin@xxxxxxxxxxxxx>
- Date: Tue, 16 Nov 2010 15:52:32 -0800
So those stats were pulled during the afternoon at peak load (which is light
for our servers, around 20-25 sessions). I kept an eye on it over the course
of an hour or two and saw that same pattern, flat with a couple of slight
changes.
I do appreciate those links, I did read through those before sending the email
to the list and agree that those have are good resources.
I did some testing last night on a Citrix server without MaxCmds set in the
hops of understanding exactly how it works under the hood. The citrix blog
article on SMB
tuning<http://community.citrix.com/display/ocb/2010/10/21/SMB+Tuning+for+XenApp+and+File+Servers+on+Windows+Server+2008>
seemed to suggest 'current commands' = smb mpx limit which didn't seem to be
the case in my environment.
On to the late night very unofficial test: Citrix is hitting CIFS on a Netapp
array with MPX set to 126 so the session was established with the default 50
limit (from the client). Off the start it seemed to take up 2 'commands' just
from a user being logged in. With 50+ directories opened in windows explorer
it seemed to take up a "current command" for each folder opened, until around
50 after which is seemed to be re-using slots or something. It didn't scale up
linearly from that point. Opening 50+ individual files used just one more
'command' on the server. So 50 files = 3 commands, 50 directories = 52
commands. I tested 50 files both in the same directories and each in its own
directory (yay for powershell). I never did hit a point where a file or folder
open request failed, though I only tested up to about 75 of each.
After tuning the server (MaxCmds), opening more than 50 folders resulted in a
'command' for each folder, scaling linearly.
So my guess is that connections to folders and files are counted in that
perfmon counter but that when it hits the negotiated limit the server begins to
re-use or scavenge connections which can cause performance problems. I wonder
if this would only be a hard limit if your MaxCmds was less than the number of
sessions (or users?) on the given machine perhaps? Or perhaps it's only a
problem when 50 files are opened at a given time from that machine and not 50
folders?
So unfortunately I didn't satisfy my question of what exactly that counter is a
reflection of and how that relates to MaxCmds. I'm just going to roll that
setting across the Citrix farm even though I'm not so sure it will provide any
benefit or in what circumstances it would provide benefit. So maybe this was a
snipe hunt, maybe not.
Andrew I fully realize the possibility of there being a problem with this and
it not being reported. It's also possible that this isn't a problem and
changing these settings will make no difference in my environment and I'm just
wasting my time. It's better to err on the side of performance and since this
is quite mainstream the changes will be made.
Thanks for the help everyone.
Dan Dill |Systems Engineer | Harsch Investment Properties
From: thin-bounce@xxxxxxxxxxxxx [mailto:thin-bounce@xxxxxxxxxxxxx] On Behalf Of
Andrew Wood
Sent: Tuesday, November 16, 2010 12:03 AM
To: thin@xxxxxxxxxxxxx
Subject: [THIN] Re: SMB Current Commands and SMB tuning
Joe's figures are the ones recommended - bear in mind you need to change the
file servers as well
To answer your question Dan, perhaps your servers are fairly flat because the
SMB commands were relatively constant when you did the test and by the looks of
it, operating relatively well. When did you do the test? During the day? First
thing in the morning? At the end of the day? Mid afternoon? If you were to set
off some big file copies on a server - does the number go up? If you tune the
values on the TS servers and the file servers, does the file copy take less
time?
What are those perfmon values on your file servers/domain controllers? What is
the maxmpxct value on the file servers those citrix servers communicate with?
"I've got an un-tuned server that hit 57 today without any reported issues" -
its possible you did have a problem, it slowed for a second or two, or wasn't
as responsive as it could have been. "No Issues" often equals "no one said
anything to me" - Maybe no one would have reported it because, unless it was
prolonged, or regular, or stopped your CEO from checking Facebook few would
raise a helpdesk call along the lines of "5 minutes ago the citrix server
paused for a second or two, but its ok, it got better"
This is a great article to read - which details SMB tuning across Windows 2003
x32/x64->windows2008R2
http://community.citrix.com/display/ocb/2010/10/21/SMB+Tuning+for+XenApp+and+File+Servers+on+Windows+Server+2008
A.
From: thin-bounce@xxxxxxxxxxxxx [mailto:thin-bounce@xxxxxxxxxxxxx] On Behalf Of
Joe Shonk
Sent: 16 November 2010 05:01
To: thin@xxxxxxxxxxxxx
Subject: [THIN] Re: SMB Current Commands and SMB tuning
Here are the recommended settings from Microsoft:
(http://support.microsoft.com/kb/324446)
Microsoft Recommended Optimization Settings for Terminal Service
MaxWorkItems = 8196
MaxMpxCt = 2048
MaxRawWorkItems = 512
MaxFreeConnections = 100
MinFreeConnections = 32
MaxCmds = 2048
RegistryLazyFlushInterval = 60
PoolUsageMaximum=60
PagedPoolSize=4294967295
From: thin-bounce@xxxxxxxxxxxxx [mailto:thin-bounce@xxxxxxxxxxxxx] On Behalf Of
Dan Dill
Sent: Monday, November 15, 2010 6:11 PM
To: thin@xxxxxxxxxxxxx
Subject: [THIN] SMB Current Commands and SMB tuning
Hello,
I've been digging into SMB tuning for our PS 4.5 Server 2003 x64 citrix farm as
our farm currently doesn't have MaxMpxCt or MaxCmds set on our servers. I am
confused by the results of the Redirector\Current Commands output in perfmon.
That counter is defined as: "counts the number of requests to the Redirector
that are currently queued for service" however the graph over time is quite
flat when I could expect a queue graph to look spikey. Is that counter a
combination of the outstanding SMB requests plus something else?
I've got an un-tuned server that hit 57 today without any reported issues when
I would have expected it to cap at 50 or start exhibiting problems at that
level.
Anyone know why those counters are fairly flat? Is there a better way to see
if I'm hitting MaxCmds or do people just set it higher regardless?
[cid:image001.png@01CB859E.84F27400]
Thanks,
Dan Dill |Systems Engineer | Harsch Investment Properties
Other related posts:
