Sorry, I messed up my previous posting (evreything after "<" got lost) - text should be:
From what you write I suspect that you are running a Linux kernel 2.6.x with x < 10 (which seems to be true). "top" there doesn't report correct values for multi-threaded programs (like einstein), so don't trust it - or upgrade your kernel.
Before i upgrade the kernels on the dual boxes one question remains.
Why does it only seem to affect Einstein, or what is Einstein doing differently.
The rest of the linux fleet are on 2.4 kernels this will be OK yes ?
They aren't SMP.
Upgrading them would not be easy as the kernel isn't standard {LTSP}
Why does it only seem to affect Einstein, or what is Einstein doing differently.
Einstein is the only BOINC App I currently know of that has graphics on Linux, resulting in a multi-threaded program (even if you don't show the graphics). All other projects are, I think, single-threaded, so this top bug doesn't apply there.
BTW: you shouldn't need to upgrade your kernel to _run_ E@H.
However i am not convinced that this is a 'top bug'.
Judging by the amount of work that was actually done by my linux boxes einstein was not processing.
Exact same hardware (AMD 2600+) running WinXP vs. Gentoo Linux:
WinXP
1940.9 floating point
3276.32 integer
22,600 seconds average time to complete w/ app ver. 4.79
Linux
1003.82 floating point
2125.63 integer
28,200 seconds average time to complete w/ app ver. 4.81 (command line only)
28,650 seconds average time to complete w/ app ver. 4.81 (boincmgr also running minimized)
On my machine, Linux only benchmarks at only 60% of the WinXP level (with less claimed credit as a result) and is roughly 20% slower in actual crunch times. I understand that it was much worse before, but there’s still quite a wide gap to bridge.
Can't make any response to linux issues, sorry still ignorant to linux, want (need) to change that in the future!! I can say my 3700+ grinds out wu's in about 2hrs wall clock time, just a click over sometimes, with cpu times of @19k. Fast or slow, it's good to be AMD!! Want to build a 64 X2 later this year, maybe even a dual proc, duel core opteron, now wouldn't that kick ass!!!!
Can't make any response to linux issues, sorry still ignorant to linux, want (need) to change that in the future!! I can say my 3700+ grinds out wu's in about 2hrs wall clock time, just a click over sometimes, with cpu times of @19k. Fast or slow, it's good to be AMD!! Want to build a 64 X2 later this year, maybe even a dual proc, duel core opteron, now wouldn't that kick ass!!!!
My bad on time, sorry, was thinking LHC, takes my machine 5 to 5.5 hrs per wu.
2.6.9-5.0.3.ELsmp
)
2.6.9-5.0.3.ELsmp kernel
No X
command line only
./boinc -allow_remote_gui_rpc -return_results_immediately > /dev/null &
One of my windows boxes runs boincview so i can easily keep tabs on the clients.
Sorry, I messed up my
)
Sorry, I messed up my previous posting (evreything after "<" got lost) - text should be:
From what you write I suspect that you are running a Linux kernel 2.6.x with x < 10 (which seems to be true). "top" there doesn't report correct values for multi-threaded programs (like einstein), so don't trust it - or upgrade your kernel.
BM
BM
OK So I am on x = .9
)
OK
So I am on x = .9 :)
Before i upgrade the kernels on the dual boxes one question remains.
Why does it only seem to affect Einstein, or what is Einstein doing differently.
The rest of the linux fleet are on 2.4 kernels this will be OK yes ?
They aren't SMP.
Upgrading them would not be easy as the kernel isn't standard {LTSP}
BTW
Bernd
Thanks for taking the time to answer my question.
RE: Why does it only seem
)
Einstein is the only BOINC App I currently know of that has graphics on Linux, resulting in a multi-threaded program (even if you don't show the graphics). All other projects are, I think, single-threaded, so this top bug doesn't apply there.
BTW: you shouldn't need to upgrade your kernel to _run_ E@H.
BM
BM
Bernd Thanks for the
)
Bernd
Thanks for the explanation
However i am not convinced that this is a 'top bug'.
Judging by the amount of work that was actually done by my linux boxes einstein was not processing.
It was just sitting in memory doing nothing.
There are other projects so no worrys.
PCZ: You may want to try our
)
PCZ:
You may want to try our new Beta App (see this thread). It contains newer BOINC code that may fix your problem.
BM
BM
It's quite a substantial
)
It's quite a substantial improvement but still approx 12% slower than Windows.
Exact same hardware (AMD
)
Exact same hardware (AMD 2600+) running WinXP vs. Gentoo Linux:
WinXP
1940.9 floating point
3276.32 integer
22,600 seconds average time to complete w/ app ver. 4.79
Linux
1003.82 floating point
2125.63 integer
28,200 seconds average time to complete w/ app ver. 4.81 (command line only)
28,650 seconds average time to complete w/ app ver. 4.81 (boincmgr also running minimized)
On my machine, Linux only benchmarks at only 60% of the WinXP level (with less claimed credit as a result) and is roughly 20% slower in actual crunch times. I understand that it was much worse before, but there’s still quite a wide gap to bridge.
Can't make any response to
)
Can't make any response to linux issues, sorry still ignorant to linux, want (need) to change that in the future!! I can say my 3700+ grinds out wu's in about 2hrs wall clock time, just a click over sometimes, with cpu times of @19k. Fast or slow, it's good to be AMD!! Want to build a 64 X2 later this year, maybe even a dual proc, duel core opteron, now wouldn't that kick ass!!!!
RE: Can't make any response
)
My bad on time, sorry, was thinking LHC, takes my machine 5 to 5.5 hrs per wu.