afp server issue - very hign cpu load

Hi, yeah I have done that a few times, but once I did it and the AppleFileServer process didn't start again and I had a lot of angry people looking a spinning beachballs.

We are using standard for the authentication, but we were using 'any' until the problem crept in and i changed it to standard

I changed the afp wanquantum and afp wanthreshold ONLY on the server - no effect
I am now putting the change on all the clients and will post results

Eric Hilferding wrote:
We are using standard for the authentication, but we were using 'any' until the problem crept in and i changed it to standard

I changed the afp wanquantum and afp wanthreshold ONLY on the server - no effect
I am now putting the change on all the clients and will post results

hey Eric

change it back, none of my recommendations were for the server! It was for the clients only!

Please try to change the server back to the 0 value I assume was there before and please try to change the afp wanquantum and afp wanthreshold on the CLIENTS please! There has been about 5 people now that change the afp wanquantum and afp wanthreshold and AFP stopped its pinging

have a go and let me know

Cheers

Back again,
AppleFileServer-process at 200% after one production day., did kill -9 PID Clienst kept connected.... CPU load from 2.something dropped to normal levels. hopefully tomorrow another production day.
keep u posted, greetz, Moi.

I assume none of you are using Network Homes for you users, cos from what I've read afpwanthreshold and afpwanquantum changes have no affect on Network Home users

Kevin Neal wrote:
I assume none of you are using Network Homes for you users, cos from what I've read afp wanthreshold and afp wanquantum changes have no affect on Network Home users

sorry but I have over 200 users with network homes and yes according to the docs from apple afp wanthreshold and afp wanquantum don't work, but according to the FACT that I don't have any freaken problems with AFP any more since I did afp wanthreshold and afp wanquantum then I think maybe try it with your clients then let me know what is going on.

It's all black magic. There is a fundamental problem with the AFP binary and no amount of threshold tweaking is going to make a blind bit of difference.

Stuart Ramdeen1 wrote:
It's all black magic. There is a fundamental problem with the AFP binary and no amount of threshold tweaking is going to make a blind bit of difference.

yeah black magic, because no matter how much I talk about what I did, but how it worked for my problems...and a few other peoples, its like black magic,

going to get myself some xserve voodoo doll and then poke it with pins.. LOL

again though I think that we have two, six, or ten different problems with AFP, my case afp wanthreshold and afp wanquantum tweaking worked, but then afp wanthreshold and afp wanquantum doesn't work for other people

sorry all I know is, major problems with AFP, then afp wanthreshold and afp wanquantum tweaking then no problems, please let me know why it worked for my situation and a few others.....


Black Magic

I did change those settings when the info was first put on apple's support site so this is not relevant in my case

Yep, that was a typo:

xserve:~ root# defaults read -g com.apple.AppleShareClientCore
{
"afp wanquantum" = 131702;
"afp wanthreshold" = 1000;
}

update: last 2 prdoction day's load on server was about 1.2, AppleFileServer 102 %, only one client connected, after kill PID AppleFileServer (after prduction day) load dropped again (as expected), client kept it's connection.

now set network-preferences, disabled ipv6, en manually set network-speed instead of auto (topic Network speeds extremely slow.... reminded me...) lets see what tomorow brings....

keep u posted, kind regards from the Netherlands.

UPDATE *slight optimism* ???... load very low, nothingk to kill,
4 current connected user, max 14 clients logged on today.
load = 19:05 up 3 days, 19:29, 2 users, load averages: 0.02 0.10 0.08
looks like last settings helped ?
- disabled ipv6
- "manually set" ethernet speed settings (not auto),
- loaded quicktime 7.6 not restarted)
but we'll be careful, we'll be back on monday.
keep U posted, for now WEEKEND.
Maybe You Foogs are able to test as well with these latest changes ?
Kind regards, again/still from The Netherlands.

I have solved our problems with this issue. I now have low CPU usage, reduced memory usage, and significantly reduced network traffic. It's been running this way for several days.

Our environment:
K-12 School with a dedicated server for Network Home Directories. Server is a PowerMac G5 dual core 2GHz G5 with 2GB memory. Ever since upgrading the server and clients to 10.5 this past summer, this server (which was running fine with Tiger server and clients) was being hammered with saturated CPUs, high network bandwidth and high I/O rates.

After researching this issue, I implemented a couple of items to reduce server load:
1) disabled spotlight for all the sharepoints on the server
2) implemented local Cache file redirection per John DeTroy's Tips and Tricks document

That still wasn't enough relief. So I looked at the open files on the server using the 'lsof' command in Terminal.

I noticed a large number of open Spotlight files in the homedirs of my users. I decided to find out how to turn off Spotlight on my client computers since everyone only logs in with their network homedir identities. I found instructions on how to do that at: http://www.macosxhints.com/article.php?story=20071102215912892&query=disable%2Bs potlight

After using Remote Desktop to execute the appropriate commands on all my clients, the open Spotlight files disappeared, the CPU load attributed to the AFP server dropped very low, idle CPU time once again appeared on the server, the memory consumed by AFP server dropped significantly and the network and disk traffic really backed down.

After running this way for several days, I'm very satisfied with the results and my impression is that performance has returned to the levels which existed prior to the clients and servers being upgraded to Leopard.

Hope this helps!


Randy

good to hear, bur.... our client need to be able to search ther files on the server volumes as they did without problems on their 10.4 server.
so wat do you use, when client need to find files on the sever ? Easyfind ?
houdah ?
like to hear, kind regards.

Well, maybe not black magic, more gray. 🙂 I applied this last Wednesday to my 3 labs (60 iMacs) and over the weekend to the remaining faculty/staff machines. On Wednesday CPU had reached 90% (or 360% since both processors/cores were involved) before I made the change and installed and restarted - I did kill -9 PID and it dropped to around 10% and stayed that way until today, even after a few spikes Thursday and Friday. Today the first class - 30% stuck, the second class - 50% stuck and I did kill again. So it hasn't fixed it - but it has seemed to keep the spikes from being so high and slowed down the plateauing. More than some of the things I've tried....
Did anyone apply the OS X Server 10.5.6 v1.1 that was released this weekend? It's showing up on my SUP server and I've applied it there, and made it available, but it's not showing up for my Mac Manager Xserve yet. And it's not on the Apple website (as least not as of early today). Maybe it'll help?