I have a 3.2GHz P4 with HT running XP SP2 and BOINC V4.13.

Please check out this wu;
http://climateapps2.oucs.ox.ac.uk/cpdnboinc/trickle.php?resultid=359584

Note that the time per time step is slowly increasing.

Also with this wu;
http://climateapps2.oucs.ox.ac.uk/cpdnboinc/trickle.php?resultid=364734

Note that the time per time step is also increasing.

So why is the t/ts slowly increasing?

Also note that these two wu's are being run on the same machine with one on each HT cpu. So why is the t/ts so different?

Love to know why for all these questions.

---

Paul
(S@H1 8888)
<img border="0" height="80" src="http://31411.sah.sig.boinc.dk?188"></a>BOINC/SAH BETA<img src="http://mysite.freeserve.com/gifcollection/img52.gif">

&gt; I have a 3.2GHz P4 with HT running XP SP2 and BOINC V4.13.
&gt;
&gt; Please check out this wu;
&gt; http://climateapps2.oucs.ox.ac.uk/cpdnboinc/trickle.php?resultid=359584
&gt;
&gt; Note that the time per time step is slowly increasing.
&gt;
&gt; Also with this wu;
&gt; http://climateapps2.oucs.ox.ac.uk/cpdnboinc/trickle.php?resultid=364734
&gt;
&gt; Note that the time per time step is also increasing.
&gt;
&gt; So why is the t/ts slowly increasing?
&gt;
&gt; Also note that these two wu's are being run on the same machine with one on
&gt; each HT cpu. So why is the t/ts so different?
&gt;
&gt; Love to know why for all these questions.
&gt;
&gt;
It's normal for the sec/ts to slowly increase throughout the model, but your first model listed is increasing quite a bit more than would be considered average. It's also normal for one model to run a little faster on a HT P4 than the other, but once again, the difference between your two models is well above average.

Are you doing something else in background on the PC that is sharing CPU time on one of the virtual CPUs? Check the Task Manager at times and see if there is something sucking CPU cycles. You're not running the visualization are you?

I have noticed the CPDN calculations don't behave well under hyperthreading.

Initially, I had configured my system to run 4 units in parallel (it is a dual Xeon with hyperthreading). The estimated time for completion (for each of the units was 1600 hours).
However, at first, all process did seem to run in parallel (in taskmanager, the cpu time for each unit was equal to real time). But as the model went forward, the parallelism went (in taskmanager, the cputime for each unit was much less than real time).

Now, my system is configured to run 2 units in parallel, and this yields an estimated time of 600 hours (for each of the units). This means that if I run two units in parallel, and after that again two units in parallel, I'd have spent 1200 hours to complete. This is a huge difference from the 1600 hours.

Apparently, the type of calculations do change over time, yielding less and less benefit from hyperthreading.
You can check whether you benefit from it in the taskmanager: just look at the CPU time spent on both workunits. If this is over half of real time, you'll benefit from hyperthreading; if it is less than half of real time, the processes are taking turns on the CPU (no more parallelism, even made worse by the time spent swapping).



Jörg

---

CPDN is very dependent on memory I/O, so even with hyperthreading you're still hitting the same memory bus etc, i.e. you have four times the I/O which I imagine is slowing things up.

&gt; CPDN is very dependent on memory I/O, so even with hyperthreading you're still
&gt; hitting the same memory bus etc, i.e. you have four times the I/O which I
&gt; imagine is slowing things up.

Yes, that could also be.
But I gather it doesn't explain the slowdowns The Gas Giant is seeing...?


Jörg

---

A bit more info....mobo is 800MHz FSB, memory is 1GB PC3200 Corsair. The computer is sitting there crunching BOINC 24hrs a day with out much else going on except for dial-up for about an hour a day. Task manager shows the 2 wu's within 5 seconds of cpu time of each other with them typically at 50% each.

I understand that with HT on the wu's will be slower overall but should be faster with the 2 running at once...but the 2nd wu times appear to blow that concept out of the water.

---

Paul
(S@H1 8888)
<img border="0" height="80" src="http://31411.sah.sig.boinc.dk?188"></a>BOINC/SAH BETA<img src="http://mysite.freeserve.com/gifcollection/img52.gif">

It could just be this parameter set is leading to longer or more calculations for this model, which often means it is going unstable, however your 1st phase temp/precip look stable.