Periodic slowness with network accounts

[cainadai]

[b]The server: [/b]10.3.9 Server, Open Directory Master serving Authentication, AFP and DNS.
[b]The clients: [/b]10.3.9 & 10.4.9 (mix of G4 and Intel, ~30 total)

[b]The issue: [/b]Users with network home directories experience pauses throughout the day where their cursor disappears (if they are typing, things typed stop appearing on the screen) and the mouse pointer turns into a spinning pinwheel for anywhere from 5 to 15(ish) seconds. It seems to happen most frequently in FileMaker, MS Word, FireFox, Mail and when switching between applications.

[b]The environment: [/b]Users are doing a lot of work with FileMaker Pro 8.5 and MS Word 2004, including mail merges between the two. There is also a bit of large-quantity printing going on to various network printers. Between automounts and normal file sharing connections, our max connected users is peaking around 45, which doesn’t seem unreasonable.

[b]What I’ve done: [/b]First, I checked to see if the server is being overloaded. As far as I can tell, it’s not. Overall load is low (uptime reports below .5 at peak hours), cpu usage is low (10-15% on average with occasional spikes to ~33%), memory usage is acceptable (1GB total RAM, only ~400MB used) and there is plenty of free space on the hard drive.

I also checked system resources on the client machines; the only thing that seemed odd to me is that “free” memory was extremely low (under 10MB in some cases), but when added to the “inactive” memory there seems to be plenty available to the system. There don’t seem to be any spikes in network activity or cpu/memory usage that coincide with occurrences of the issue.

I noticed that there was some extremely verbose access logging going on, but the issue persists even after turning access logging off entirely.

I noticed that AFP downloads from the server were capping out much lower than I’d expect (it’s only a 10/100 network but we weren’t getting even close to the low-end throughput) but I tracked that down to a problem with one of the switches. So, uploads and downloads are nice and speedy now but the original issue remains unchanged.

I spent a whole day with system and console logs for both my workstation and the server tailed, but no messages logged coincided with occurrences of the issue.

I installed a login hook to redirect ~/Library/Caches to the local hard drive for network users, but that hasn’t seem to make much difference (though admittedly I didn’t do it on every machine, I thought I would try it on a couple and if those machines saw any improvement at all I could deploy it office-wide).

I adjusted Max Threads in com.apple.AppleFileServer.plist from 40 to 200, and while AFS has defnitely been using more system resources than it used to… no luck. (Unclear if that particular tweak is any good for 10.3 or just 10.4.)

Because we are using a computer list to restrict login access to our computers, I tried creating a brand new computer list. No preferences are being managed but the login window. There is no group or user-level preference management in place.

I’m a bit stumped. I’ve heard that AFP performance with network homes is much improved as of 10.4.9 Server, and we are planning to upgrade anyway within the next few months, but I’d hate to push the schedule up and have the problem linger. Reportedly, this only started happening within the last couple of months (I inherited this network, and the problem has been going on as long as I’ve been here, so I can’t say what change may have taken place to provoke it).

Any thoughts?

[deemery]

Just to confirm, we’re talking Networked but not Mobile accounts? On my Mobile account, I see this happen when OS X decides to synch/check the laptop with the ‘mothership’.

dave

[cainadai]

There are no mobile accounts in play.

(But thank you for asking!)

[dfederici]

I tried to anwer your post but it keeps getting tagged as spam.

[Author]

Posts