XGRID: lost half my cpu's per agent!

Question

XGRID: lost half my cpu's per agent!

We had a 10.5 xgrid controller and a bunch of xgrid agents also running 10.5, and everything was dandy.

Today we upgraded the controller and some of the agents to snow leopard 10.6, and now we have lost half the cpu's on those agents that were upgraded.

For example, we have a late model 8-core Mac Pro upgraded to 10.6 reporting only 4 CPUs in the xgrid admin. Calculations are only using 4 CPUs not 8 according to activity monitor.

In the terminal on this agent the command "sysctl hw.ncpu" reports 8 correctly.

I tried setting the "MaximumTaskCount" field in /Library/Preferences/com.apple.xgrid.agent.plist to 8 and it has no effect (originally the field is not defined in the plist). However if I set this field to a smaller value than 4, then I can further limit the number of CPUs to less than 4.

It is not an issue with hyperthreading. We have one 4-core hyperthreaded Mac Pro, it used to report 8 CPUs (running as expected at the speed of 4) now it reports 2 CPUs, which run at the speed of 2.

Losing half my cluster is not a nice upgrade, does anybody know how to restore their contribution under SL?

thanks!!
Eli

mixed-all xeon, Mac OS X (10.6)

Posted on Sep 11, 2009 5:44 PM

Reply

Answer 1

Sep 12, 2009 3:39 AM in response to Eli Rotenberg

I noticed that as well. At first I thought I was going crazy but it looks like a big that will need to be fixed by Apple.

Reply

Answer 2

cwbob

Level 1

0 points

Sep 13, 2009 7:10 AM in response to Anthony_01

I noticed this as well on MacPro system. Xgrid in 10.6 fails to find half the processors. True in 32-bit and 64-bit mode. Also true for Snow Leopard Server.

Reply

Answer 3

Sep 14, 2009 9:53 AM in response to Eli Rotenberg

I am glad we are not the only ones. I have posted the bug to Apple's developer site.

Reply

Answer 4

xgridfan

Level 1

1 points

Sep 25, 2009 6:21 AM in response to Eli Rotenberg

Same problem here. Good idea I decided to try it first in a stand alone server vs my 48 cores production grid.

Reply

Answer 5

Oct 10, 2009 3:21 AM in response to Eli Rotenberg

Hi, I have had the same problem with two servers as below:
I have a dual- quad core 2.26GHz xserve (brand new) set up with 10.6 server, variety of services installed and working fine, however - Xgrid reports that I only have 9.04 GHz of processing power to work with (everything else implies I have, as expected 8x2.26GHz or 18.08GHz)?
I had a similar issue recently on a separate, dual 2.8 GHz quad-core MacPro - under 10.5 server it said I had 22.4GHz total processing power available to Xgrid, but upon upgrading to 10.6 server this was instantly halved to 11.2 GHz (not to mention the fact that the "if you have 10.5 server all you have to do is run the upgrade and type in the new serial number" "...everything else will be the same..." from the install manual is clearly nonsense)- So falling back to my 10.5 server backup - it's back to 22.4GHz?
Apples response from their enterprise team to the first query with the MAcPro was "try a clean install on a spare server..." I wish we all had a spare server lying around!
Anyway, once my new xserve was here - and I consider that a clean install - it is the same under 10.6 server - it's clearly a serious problem with Xgrids cpu usage. Activity monitor reports everything OK (8 processors), it's just anything to do with xgrid under 10.6

Reply

Answer 6

Oct 21, 2009 2:28 PM in response to Eli Rotenberg

The solution to fixing the missing-half-CPUs problem (and also the loss of hyperthreaded CPUs) has been posted by an Apple engineer on the xgrid mailing list.

You can follow the thread starting here:
http://lists.apple.com/archives/Xgrid-users/2009/Oct/msg00001.html

We have tested this out on a bunch of 10.6 agents/server and it works fine now.

I don't know why Apple has changed the default number of CPUs, apparently for some performance reason, but as I type on this 4-core machine all 8 HT-CPUs are blazing away on a calculation, and yet there is no perceivable delay for web browsing and other desktop apps. I hope in the future before introducing such limitations that apple could add more preferences to the agent control panels to allow the user to better control cpu utilization.

Reply