Odd results (meaning workunit runtime) under Linux with Hyperthreading
23 Aug 2006 8:40:19 UTC
Topic 191726
(moderation:
At work I use two "server" (in fact Fujitsu-Siemens Esprimo with Intel Pentium 4 630 at 3000 MHz each) under SuSe Linux 10.1 two crunch for Einstein@home.
Because these clients do not have internet connectivity (in fact connectivity is not allowed) i use my own notebook (Intel Pentium M 1200) at my desktop to report the workunits under VMWarePlayer with a Fedora Core 6 Test 1.
I frankly copy the BOINC directories with a USB-stick from the clients onto my notebook and vice versa.
But the runtime on the clients is much to long for my flavor:
Intel Pentium 4 630 (3000 GHz)
WU: 13,50 C; ~ 17995 s
WU: 13,19 C; ~ 17119 s
Intel Pentium D 805 (2666 MHz) (comparison, runing Windows 2003 Server)
WU: 13,80 C; ~3440 s
I do not think that Hyperthreading which is enabled on the Esprimos has that much influence on the workunit runtime.
Has anyone any ideas what is going on here?
(Mr. Machenschalk has been recomended to me ;) )
Odd results (meaning workunit runtime) under Linux with Hyperthr
)
Have you the ability to compare the wall-clock running times of the tasks?
If I'd use the same client information (i.e. client_state.xml file) on two completely different machines (in this case even with different number of CPUs), I'll simply won't give anything about information such as benchmark or runtime. It may work to copy the whole BOINC directory to transfer tasks to other machines, but the Client isn't made for it and will most likely get confused.
BM
BM
Work is only allowed in the
)
Work is only allowed in the time between 06:00 pm and 06:00 am and the computers only get four and two half workunits in this time and this is consistent with the reported workunit time.
I'm not quite sure if this is
)
I'm not quite sure if this is possible in your case, but afaik SuSe servers have a so-called "power savings" option in their kernel, so as long as the cpu supports it maybe the CPU is not running full speed at night? I think that option is mainly for laptops but you CAN use it on servers if the hardware supports it... I'm actually quite sure about that because one of our servers had problems with that because it was switched on by default and we didn't want it ;-) But, as I said, I have no idea if this is the case with the servers you are using for crunching. It was just an idea that came up in my mind because I'm a laptop user and have made similar experiences on my lappi when I couldn't let the CPU run full speed due to heat reasons in July.
RE: I'm not quite sure if
)
FYI, it's not just SuSe and it's not just on servers. I run Slackware 10.1 on one compter and Slackware 10.2 on another and both of them have this feature. It's a part of any late model Linux kernel. If it's built into the kernel there's nothing much you can do (afaik) except possibly with config files, but if it's built as a module you can disable loading the apm modules and it won't power down or throttle back. I run BOINC on both computers so I simply don't load the modules so both machines are "alive and kicking" at all times.
When asked a question and you are not sure of the right answer, I've found that the best answer is always "I don't know for sure, but I'll find out!"
RE: RE: I'm not quite
)
Please forgive me if I say a supudity...
But, if you allow a higher priority at "einstein..." the kernel wouldn't slow the computer.
By default, boinc's priorities are very low and the kernel can slow the processor if only low priorities jobs are running. If 1 or more high priorities jobs are running, the kernel won't slow the machine.
Sorry if I seem to be stupid, but work week was a hell and I am "burn out" till monday.
"Entia non sunt multiplicandam praeter necessitatem"
(OKHAM)
I think you actually can
)
I think you actually can deactivate this feature or alter the settings... but of course I'm not sure if that is a smart thing to do if it's not really your server ;-).
This is what my SuSe manual says about power saving: