Looks like no one’s replied in a while. To start the conversation again, simply ask a new question.

afp server issue - very hign cpu load

hallo

i googled an searched this forum al long time but i found no solution.

my problem is that my os x 10.5.4 server with about 30 networked homeddrive users have an issue with the afp server. the afp server process uses all 8 cores of this newest intel xserve with 14 gigs of ram installed. when this happens all users get an spinning wheel. the incoming network traffik is reduced to some kb´s.

ok all users shut down there clients - restart server and about 30 minutes later i have the same problem.

i have dumped the network traffic with wireshark and there i see some tcp retransmissions.
now i need someone who can help me analyse the wireshark protocol, because i cant´s handle that.

so if there is someone out there who can help me plz send me an email to support@premedia.at so that i can send you the wireshark log.

thank you in advice

Macbook Pro, Mac OS X (10.5.4)

Posted on Aug 29, 2008 2:47 AM

Reply
279 replies

Sep 8, 2008 7:34 AM in response to Manfred Rumpl

Same problem here. We did an upgrade from 10.4.11 server to 10.5.4 and this is the result. It used to use around 70-120% usage on a 4x CPU server. Now it jumps to 370% as soon as around 20 Users are logged in.

Our clients are mostly 10.4 at this point, so we can't use the NetworkHomeRedirector.

Does anyone have a resolution or other suggestions for this?

Thanks.

Sep 8, 2008 7:43 AM in response to Brett_X

Prior to 10.5 offering this, there was a client-side option mentioned near the beginning of the afp548 article:
http://www.afp548.com/article.php?story=20060915140425369

I also always advise quite strenuously against upgrade installs, for a whole variety of reasons.
I am not saying that "is" the cause of your problems, but it seems that most upgrade installs are not problem-free. This was also true of 10.2 -> 10.3 and 10.3 -> 10.4

Sep 10, 2008 7:03 AM in response to Brett_X

Follow Up- Short version: Deselected Spotlight search (was enabled by default), and rebooted. (Also added RAM, but we don't think that was the problem.)

Long version- I found that Spotlight searching was enabled on the share that hosts the home folders. Once I de-selected that, it didn't make any difference right off (without rebooting).
We shut down and added more RAM- from 2 GB to 4 GB, but there was already 1GB that was "inactive", so, that didn't seem to be a problem.
Once we brought it back up, things seemed better. The RAM usage was still the same.. only using about 1GB and the rest was either "available" or "inactive", so I really think it was the Spotlight searching that was the real CPU hog. It still does use way more CPU than 10.4.11 did.. which is okay as long as nobody gets slowed down by it.

Here are some of our details an observations in case anyone has issues and wants to compare (and in case I want to look back at my post in the future)...
We're running 1 Master OD and 3 replicas. The Master is a Quad 2.5 Ghz G5, and hosts home folders for about 130 users.. of which 80 or so are live network home directories.
The CPU utilization used to run around 70% for AppleFileServer on a normal day, and jump up to around 120% when people were logging in in the morning. When we used to reach that 120% or so, people would notice a slowdown. I never saw the utilization reach 200% or higher (which was always peculiar to me). This was true even after changing the max threads to 200.
Now, (after disabling Spotlight) the CPU utilization for AppleFileServer seems to spike up to 350% sometimes, but it rarely stays there for any length of time. It most often seems to be bouncing between 50% and 250%. And people aren't reporting any slowdowns. We're very happy.
The behavior we're seeing now is what I would have expected in the first place. I always wondered why AFS didn't use as much CPU as was available before causing slowdowns. My theory is that Tiger's AFS wasn't written well for multi-CPU, and Leopard's is.
I hope this helps someone.

Sep 10, 2008 1:34 PM in response to Brett_X

I am having this same problem.
I at one time attributed it to spotlight, and disabled that on all shares
seemed better for a while, but came back
I have a dual G5 Xserve, 4GB Ram, Xraid, hosting 15 home directories, and five network volumes, to no more then ten people at a time... I have seen afp using 180% and all computers become unusable...

This seems to happen once a week...
oh and I cave done three clean install, set the clients to not make .DS_Store files, redirected their caches to the local drives, increased kernel limits, and afp threads...

Still afp uses %40 percent average cpu
drops to a couple % sometimes
jumps to 180% sometimes, and everything stops...
Server and clients are running 10.5.4
Thanks, Rich

Sep 12, 2008 9:34 AM in response to Rich Simpson

I'm seeing the same problem as Rich. Spotlight not enabled, 2 x 2.8Ghz Quad core Xeon, 8Gb of RAM. OS X Server 10.5.4

Have tried everything, DNS, all the afp tuning tricks that I've found on the net. Users grinding to a halt and general slowness, High cpu load on server (sometimes it hits 100% and won't come down until I reboot). Traffic to and from the server is seemingly minimal (because all the users beachball).

Practically unusable from the end user point of view.

My server has active directory integration though so it's perhaps more complicated than other setups. Home directories on the Mac server.

Up until the summer the server was a Single G5 Xserve with only 1Gb of RAM, 10.4.11 Server with Tiger clients. It struggled but coped a lot better than the new server running Leopard.

Any new ideas would be greatly appreciated.

Sep 14, 2008 11:27 PM in response to Fridgemagnet

hi there

i never had spotlight searching on any share on. but the problem persists.
ram is definitly not the problem, i have 14 gigs installed.

i had an similar problem on 10.4 server. and i figured out that the problem always
comes on a specific day and time. i this case it was the auto check for update function
of os x. turned this of and it never came back. i had this problem on 4 xserves.
on 10.5 it seems that that is not the problem.

Sep 15, 2008 5:53 AM in response to Manfred Rumpl

I'm still experiencing major difficulties. The only thing I have found so far that helps is to redirect user's entire Library folder to /tmp/username/Library using MCX but this has in turn created a new problem (group MCX settings not being applied).

I tried a more granular redirection of most of the folders within Library, leaving Preferences untouched but users still experienced crippling slow down and CPU climbs suddenly.

Out of interest I also have the user's Microsoft User Data redirected as the slow downs seem in part connected to Office.

Sep 22, 2008 9:08 AM in response to Fridgemagnet

I had to come up with a custom solution (for a different reason- not the slowdown issue) for making our Users ~/Library folder local. It works in 10.4 and 10.5 by making a symbolic link from their Library folder to the local drive. I wrote a script that our users have to run which A) moves the ~/Library/Caches folder out of the way, then B) copies the remainder of the ~/Library folder to the local drive and C) installs a launch daemon to back up the [now local] ~/Library folder to ~/Library.backup folder once an hour using rsync.
In doing this, all of their ~/Library folder is local, but it gets backed up. Also, if we change our setup and want to make the ~/Library folder networked again (or when re-imaging machines, etc), I can just blow away the sym link and change the name of the ~/Library.backup to ~/Library.

Does anyone care to see the script?

Sep 22, 2008 1:41 PM in response to Fridgemagnet

It's initially copied from the current location to the /Users/Shared/Local Folder/$username/Library folder. From there, the rsync utility keeps that data backed up in the user's network home folder once per hour.
I can't do the script justice from in this format, so email me at my username at mac and I'll send you all of the files. There's only 3 files - totaling less than 100K.

Oct 6, 2008 6:02 AM in response to Brett_X

Making the entire Library folder local sounds like overkill and may likely cause as much traffic syncing and backing up each time as is saved by making it local in the first place.

The ~/Library/Caches folder on the other hand is where 90% of transient files go and can safely be dropped after a client session with no need to send anything back to the server.

Importantly, cache folder redirection is supported in WGM on Leopard server as standard and doesn't require a work around. It will automatically delete existing Cache folders and create a symlink for each user. The process can even be reversed this way if required.

To access it you have to first add the preference manifest from /System/Library/CoreServices/Managed Client.app in WGM, this opens up a whole bunch of new options!

Its usage is best described by John DeTroye in his awesome Tips and Tricks v14 for Leopard here :-

http://web.me.com/johnd/JohnDsSite/Tips_%26_Tricks/Tips_%26Tricks.html

Download the pdf and check pages 53 & 55.

Good luck.

afp server issue - very hign cpu load

Welcome to Apple Support Community
A forum where Apple customers help each other with their products. Get started with your Apple ID.