Ok two hosts -
#1 P4 2.4 G running win2k
#2 AuthenticAMD AMD Athlon(tm) XP 2000+ (redhat)

Both hosts cruch Seti Classic WU's at about the same rate (#1 ~5wu/day) (#2 ~6wu/day) (based on months of setique data)

But, the benchmarks are:
#1
--- - 2004-06-23 19:49:30 - Number of CPUs: 1
--- - 2004-06-23 19:49:30 - 1536 double precision MIPS (Whetstone) per CPU
--- - 2004-06-23 19:49:30 - 3283 integer MIPS (Dhrystone) per CPU

#2
2004-06-23 19:44:18 [---] Number of CPUs: 1
2004-06-23 19:44:18 [---] 873 double precision MIPS (Whetstone) per CPU
2004-06-23 19:44:18 [---] 2079 integer MIPS (Dhrystone) per CPU
2004-06-23 19:44:18 [---] Finished CPU benchmarks

It seems that 3.19 (on #1) or 3.18 (on #2) result in screwy benchmarks. The faster machine (running seti) give lower benchmark numbers?

You could check that there aren't any background processes eating up CPU time, and also in Linux's case, it could be worth to see if there are any runaway processes (do "ps aux" or "top" in shell to see running processes (if i remeber correctly)).

ALSA (or something like that) in my case sometimes crashes and eats up resources violently.. If there is a runaway, then get rid of it with "kill [pid]" or if it refuses to die, then "kill -9 [pid]". But don't do it if it's an important process :P

Ideally SETI/BOINC should show around 99% CPU usage, if this is not the case, then something is propably eating up the cycles.

Worth a try :)

regards

--J

Edit: Also should be noted, that the results may be weird if the computer was doing something while doing the tests, as in that case, not all the resources were free for the benchmark only.

The #1 machine sounds about right. However, #2 is definitely wrong -- it is reporting only half the FLOPS required to process 6 WUs/day.

> You could check that there aren't any background processes eating up CPU time,
> and also in Linux's case, it could be worth to see if there are any runaway
> processes (do "ps aux" or "top" in shell to see running processes (if i
> remeber correctly)).
>
> Edit: Also should be noted, that the results may be weird if the computer was
> doing something while doing the tests, as in that case, not all the resources
> were free for the benchmark only.

J is correct, even minor changes in the system will make the benchmark numbers incorrect. This is still early days for SETI@Home Powered by BOINC and some things are still a little "odd". Compared to our earlier experiences with the benchmark this is "good enough".

There have been suggestions to, and now they will, run the benchmark once a week (which it does now) and this will catch changes to the system even if the participant forgets to re-benchmark.

Another suggestion was that successive runs of the benchmarks be averaged so that minor changes in the benchmark results slowly converge making those little fluctuations go the way of the dodo bird.

> > You could check that there aren't any background processes eating up CPU
> time,
> > and also in Linux's case, it could be worth to see if there are any
> runaway

Not the case.. The box was rebooted just prior to installing BOINC. (to insure my rc changes to remove Seti1 were correct) This is a "bare bones" system that I use for testing, and has very little extra stuff installed, much less running.

If anything, there was much more happining in the background on the win2k machine.

I am having a similar problem, but the benchmarks on SAH WU's just went to about 7-8 times what they were. I think I will put this in its own thread.

> I am having the same problem. Suddenly I can't get work anymore from SAH or
> PAH projects because the SAH WU's suddenly jumped from approx. 5 hours
> estimated completion time to 35 hours completion time.

Do not worry too much. Something strange happened since 2 days ago.

You can have a look at this thread.

Greetings from Belgium.

I had a similar problem: my dual-2.4Ghz system was showing the approximately the same Integer speed results (per CPU) as my single-866MHz system.

I made sure that NOTHING else was running; empty out the task bar AND the system tray, and check TaskManager (if you have one) to see if the System Idle Process is taking 99% of the CPU.

Then I opened BOINC, selected the Messages tab, and went to the "File" menu to choose "Run Benchmarks". Then do ABSOLUTELY NOTHING AT ALL, not even moving or clicking the mouse, until the Messages window reports that the benchmarks are done. It will take one FULL minute.

My results still came back odd, so I ran the benchmarks again. That looked correct (Integer speed increased A LOT), but I ran them one more time anyway and still got the good results.