I am using the client 4.13 on windows XP machine with a 2.66ghz pentium 4. For some reason its starts happily but as it gets into the work unit, the secs/TS slow down. When it started its current unit it was crunching at 3.28sec/TS. (Which would seem about the right speed.) Now into its 10th trickle it is down to 4.22 sec/ts. Each successive trickle is a little bit slower than the other.
while crunching a previous work unit, this problem got as slow as 8.10 Sec/ts, while still only halfway through phase one. So I actaully reinstalled Boinc afresh and downloaded a new workunit, as I thought something was wrong. The same machine has delivered a completed (3 phases) unit at a normal speed.
I am running 4 other machines that don't do this. Any one else having this sort of problem?
Is there some work units that take longer than others?

---

Does the visualization look okay? Nothing too strange?

What is the temperature of your CPU? Is it possible P4 thermal throttling is slowing your system down?

> Does the visualization look okay? Nothing too strange?
>
> What is the temperature of your CPU? Is it possible P4 thermal throttling is
> slowing your system down?
>

---

> Does the visualization look okay? Nothing too strange?
>
> What is the temperature of your CPU? Is it possible P4 thermal throttling is
> slowing your system down?
>

The visualisation looks fine. It is late December and cold in the northern hemisphere and warm in Australia.
Thermal throttling-- It is a laptop, but I don't understand why it would be thermal throttleing when it takes weeks to slow down. each trickle is slightly slower. If it was the laptop I would expect it to be a sudden cutting in half of the speed. I have atteched the recent results...
4.2229
21 Jan 2005 19:52:58 hidden 459211 2u6t_000153774_1 1 108020 438241 4.0570
16 Jan 2005 01:37:19 hidden 459211 2u6t_000153774_1 1 97218 374689 3.8541
14 Jan 2005 08:36:59 hidden 459211 2u6t_000153774_1 1 86416 311294 3.6023
10 Jan 2005 06:39:28 hidden 459211 2u6t_000153774_1 1 75614 267605 3.5391
08 Jan 2005 06:31:39 hidden 459211 2u6t_000153774_1 1 64812 213205 3.2896
07 Jan 2005 09:50:00 hidden 459211 2u6t_000153774_1 1 54010 177938 3.2945
06 Jan 2005 03:13:53 hidden 459211 2u6t_000153774_1 1 43208 142245 3.2921
05 Jan 2005 05:39:25 hidden 459211 2u6t_000153774_1 1 32406 106663 3.2915
04 Jan 2005 08:53:45 hidden 459211 2u6t_000153774_1 1 21604 70975 3.2853

---

This is a stab in the dark, but here goes. As you've tried this with two different models I don't think it is model related. That leaves the hardware. What changes slowly as the model runs? Massive amount of data generated, hence is your laptop filling up either in physical RAM, virtual memory or hard disk.

Just a thought.

Ian

---

> This is a stab in the dark, but here goes. As you've tried this with two
> different models I don't think it is model related. That leaves the hardware.
> What changes slowly as the model runs? Massive amount of data generated,
> hence is your laptop filling up either in physical RAM, virtual memory or hard
> disk.
>
> Just a thought.
>
> Ian
>
Most likely answer here. Add fragmented disks to the possibilities and you probably have it.

---

BOINC WIKI

BOINCing since 2002/12/8

23 Jan 2005 01:58:06 82859 459211 2u6t_000153774_1 1 118822 501772 4.2229
21 Jan 2005 19:52:58 82859 459211 2u6t_000153774_1 1 108020 438241 4.0570
16 Jan 2005 01:37:19 82859 459211 2u6t_000153774_1 1 97218 374689 3.8541
14 Jan 2005 08:36:59 82859 459211 2u6t_000153774_1 1 86416 311294 3.6023
10 Jan 2005 06:39:28 82859 459211 2u6t_000153774_1 1 75614 267605 3.5391
08 Jan 2005 06:31:39 82859 459211 2u6t_000153774_1 1 64812 213205 3.2896
07 Jan 2005 09:50:00 82859 459211 2u6t_000153774_1 1 54010 177938 3.2945
06 Jan 2005 03:13:53 82859 459211 2u6t_000153774_1 1 43208 142245 3.2921
05 Jan 2005 05:39:25 82859 459211 2u6t_000153774_1 1 32406 106663 3.2915
04 Jan 2005 08:53:45 82859 459211 2u6t_000153774_1 1 21604 70975 3.2853
03 Jan 2005 22:35:16 82859 459211 2u6t_000153774_1 1 10802 35627 3.2982

Looks like the first five trickles from that result had a duration about 36000 sec (~10 hrs), the next one 54000 (~15 hrs), the next 44000 (~12 hrs), the last three about 63000 (~18 hrs). It's gone in jumps like the speed of the processor has changed, with the intermediate durations possibly due to a speed change during that period. It will be interesting to see if it maxes out at around 63 - 64K secs. You can check the speed of the CPU using CPU-Z from www.cpuid.com

That is, if the throttling is done by decreasing the front side bus instead of periodically inserting halt statements. My laptop would throttle down from 3.06 GHz to 2.74 GHz by dropping the frontside bus speed from 133 to 119 after awhile running at 100% CPU utilization. This, of course decreased performance beyond the .32 GHz since the memory bandwidth also decreased. Now why yours would go five trickles at high speed and then slow down, I don't know. This is all just a theory... :)

Interesting; I'm just seeing the opposite with my current WU (<A HREF="http://climateapps2.oucs.ox.ac.uk/cpdnboinc/trickle.php?resultid=424410">424410</A>). It first starts around 3.8 sec/TS in phase 1, then gets a bit faster at 3.7 in phase 2, and is now down to 3.6 in phase 3.

Could it be because I regularly defrag my partitions?

---

My first thought was: hd needs defragging.
Second thought: laptop; gets moved around; heat.
So does the laptop get used / left in some places that are hotter than others, e.g. in the sun?
And have you tried the often suggested trick of placing some objects under the computer to lift it
a little, and thus increase air flow? 4 erasers are a simple, cheap and easily carried solution.

Les

I've got one system (host 20652) that degrades in a similar way. It only started behaving that way when I doubled its memory, and a reboot invariably speeds it up - have a look at its trickle history for <a href="http://climateapps2.oucs.ox.ac.uk/cpdnboinc/trickle.php?resultid=142009">result 142009</a>

Unfortunately the slowdown doesn't noticably affect any of the work I do on that system, so I've no concrete evidence that there's a problem with the memory and I can't really dedicate the system to running memory tests for a week to prove my theory. As long as I remember to reboot it a couple of times a week the secs/TS stays reasonably constant.

---

"The ultimate test of a moral society is the kind of world that it leaves to its children." - Dietrich Bonhoeffer

Have you had a look at task manager to see how much memory is used?
Have you noticed any increase?
Is your machine on 24/7?
Do you use any screensaver...?

---

Forum search Site search

&gt; Have you had a look at task manager to see how much memory is used?
&gt; Have you noticed any increase?

Memory use is exactly the same as on my other P4 HT hosts (steady at just over 50MB for each hadsm3um_* and 7.5MB for each hadsm3_*).

&gt; Is your machine on 24/7?

Yes.

&gt; Do you use any screensaver...?

No. I run boinc_cli as a service and have idle screen blanking. All hosts are set up exactly the same and 20652 is the only one that degrades, but only since its memory "upgrade".

---

"The ultimate test of a moral society is the kind of world that it leaves to its children." - Dietrich Bonhoeffer

You could have leaky memory. Under task manager/processes enable the non-paged pool and see if one of the processes is eating up NP memory with respect to time.

Sorry, Thyme Lawn, I meant the question for John Price.
About task manager, I meant total memory use.
When I started running BOINC CPDN, I noticed total memory use was increasing regularly.
I have an ADSL connection and an unused 56K modem. When I deactivated the modem, that memory use stopped increasing.

---

Forum search Site search

Just an update. The slowing is still going on.

1. It is defraged on a monthly basis, disc is 47% used.
2. The model is usuing 92-98% of CPU time. The PC is largely just processing the model with little else running. Only when I travel is it used more heavily with office, perhaps two days a month.
3. There is 145k of physical memory unused, with the model occupying 56k.
4. Screensaver set to "blank".
5. Machine not running 24/7, just 8-10hrs reboot each day.
6. I am a little puzzeled by what I should be looking for in Task Manager with regard to the NP. Is now enabled, but the numbers in k. are very low. The largest number is 61k with a System type .exe program.
7. I have now moved to Boinc 4.19 with no change. the slowing continues.
8. Tryed CPU Z and everything seems to be OK. CPU running at 2,658.8 Mhz. the multiplyer x20 and the FSB at 132.9 Mhz.

Time Sent (UTC) Host ID Result ID Result Name Phase Timestep CPU Time (sec) Average (sec/TS)
12 Feb 2005 19:30:30 82859 459211 2u6t_000153774_1 1 194436 947387 4.8725
10 Feb 2005 06:28:27 82859 459211 2u6t_000153774_1 1 183634 883654 4.8120
05 Feb 2005 09:17:16 82859 459211 2u6t_000153774_1 1 172832 820246 4.7459
02 Feb 2005 08:13:50 82859 459211 2u6t_000153774_1 1 162030 756501 4.6689
31 Jan 2005 05:47:34 82859 459211 2u6t_000153774_1 1 151228 693152 4.5835
29 Jan 2005 23:46:19 82859 459211 2u6t_000153774_1 1 140426 629454 4.4825
28 Jan 2005 08:14:43 82859 459211 2u6t_000153774_1 1 129624 565763 4.3646
23 Jan 2005 01:58:06 82859 459211 2u6t_000153774_1 1 118822 501772 4.2229.

Look forward to any other ideas. Or hear from others with a similar problem.

---

&gt; Time Sent (UTC) Host ID Result ID Result Name Phase Timestep CPU
&gt; Time (sec) Average (sec/TS)
&gt; 12 Feb 2005 19:30:30 82859 459211 2u6t_000153774_1 1 194436 947387
&gt; 4.8725
&gt; 10 Feb 2005 06:28:27 82859 459211 2u6t_000153774_1 1 183634 883654
&gt; 4.8120
&gt; 05 Feb 2005 09:17:16 82859 459211 2u6t_000153774_1 1 172832 820246
&gt; 4.7459
&gt; 02 Feb 2005 08:13:50 82859 459211 2u6t_000153774_1 1 162030 756501
&gt; 4.6689
&gt; 31 Jan 2005 05:47:34 82859 459211 2u6t_000153774_1 1 151228 693152
&gt; 4.5835
&gt; 29 Jan 2005 23:46:19 82859 459211 2u6t_000153774_1 1 140426 629454
&gt; 4.4825
&gt; 28 Jan 2005 08:14:43 82859 459211 2u6t_000153774_1 1 129624 565763
&gt; 4.3646
&gt; 23 Jan 2005 01:58:06 82859 459211 2u6t_000153774_1 1 118822 501772
&gt; 4.2229.
&gt;
&gt; Look forward to any other ideas. Or hear from others with a similar problem.
&gt;
&gt;
All trickle durations are about 64000 sec, which is about what they were when I calculated it before. So the computer/model is not gradually slowing down, it went along nicely for the first 6 trickles, then something happened, the last 10 have all been about the same. Since the Average (Sec/TS) is cumulative, i.e. CPU time divided by total timesteps, it will continue to increase as long as the trickle durations average about 64000 sec, which is about 6 sec/TS.

The question is...what happened between trickles 6 and 9 that caused it to slow down? I'm out of ideas.

&gt; &gt; Time Sent (UTC) Host ID Result ID Result Name Phase Timestep
&gt; CPU
&gt; &gt; Time (sec) Average (sec/TS)
&gt; &gt; 12 Feb 2005 19:30:30 82859 459211 2u6t_000153774_1 1 194436 947387
&gt; &gt; 4.8725
&gt; &gt; 10 Feb 2005 06:28:27 82859 459211 2u6t_000153774_1 1 183634 883654
&gt; &gt; 4.8120
&gt; &gt; 05 Feb 2005 09:17:16 82859 459211 2u6t_000153774_1 1 172832 820246
&gt; &gt; 4.7459
&gt; &gt; 02 Feb 2005 08:13:50 82859 459211 2u6t_000153774_1 1 162030 756501
&gt; &gt; 4.6689
&gt; &gt; 31 Jan 2005 05:47:34 82859 459211 2u6t_000153774_1 1 151228 693152
&gt; &gt; 4.5835
&gt; &gt; 29 Jan 2005 23:46:19 82859 459211 2u6t_000153774_1 1 140426 629454
&gt; &gt; 4.4825
&gt; &gt; 28 Jan 2005 08:14:43 82859 459211 2u6t_000153774_1 1 129624 565763
&gt; &gt; 4.3646
&gt; &gt; 23 Jan 2005 01:58:06 82859 459211 2u6t_000153774_1 1 118822 501772
&gt; &gt; 4.2229.
&gt; &gt;
&gt; &gt; Look forward to any other ideas. Or hear from others with a similar
&gt; problem.
&gt; &gt;
&gt; &gt;
&gt; All trickle durations are about 64000 sec, which is about what they were when
&gt; I calculated it before. So the computer/model is not gradually slowing down,
&gt; it went along nicely for the first 6 trickles, then something happened, the
&gt; last 10 have all been about the same. Since the Average (Sec/TS) is
&gt; cumulative, i.e. CPU time divided by total timesteps, it will continue to
&gt; increase as long as the trickle durations average about 64000 sec, which is
&gt; about 6 sec/TS.
&gt;
&gt; The question is...what happened between trickles 6 and 9 that caused it to
&gt; slow down? I'm out of ideas.

Does this mean that it calcs the sec/ts by averageing over the whole of the phase, or the complete model. That is when it move to phase two it may start to show the the real calc rate of about 6 sec/ts? This of course is twice what it should be and starts to point the finger at thermal throttling.

---

&gt; Does this mean that it calcs the sec/ts by averageing over the whole of the
&gt; phase, or the complete model. That is when it move to phase two it may start
&gt; to show the the real calc rate of about 6 sec/ts? This of course is twice what
&gt; it should be and starts to point the finger at thermal throttling.
&gt;
It does it over the complete model. <a href="http://climateapps2.oucs.ox.ac.uk/cpdnboinc/trickle.php?resultid=422111">Here</a> is an example where I upgraded a system from a 800 Duron on a KT133A chipset motherboard to an AthlonXP 3000+ on a NForce2 board (with two different processors intermediate to those two in between). The entire last phase was run at 2.45 sec/TS but it just kept dropping the average sec/TS slowly.

Your processor may do thermal throttling through the insertion of idle cycles in which case you won't see a CPU speed decrease in CPU-Z. There's a program called <a href="http://www.panopsys.com/throttlewatch.php">Throttlewatch</a> that can check for both types of throttling.

Uhhh. . . you mean this isn't normal? I've run two complete models and begun a third one. I've used two different computers (P4 and A64), three different hard drives, three different types of RAM, and several versions of BOINC. The sec/TS has always crept upwards at a regular rate throughout each phase of each model. It had never occurred to me before reading this that this wasn't what was supposed to happen. Current system is a brand new Athlon64 3500+ with top of the line Corsair XMS RAM, running four projects with BOINC 4.19. I don't think there's anything wrong with my hardware.

Did this ever get resolved, then?

---

&gt; Uhhh. . . you mean this isn't normal? I've run two complete models and begun
&gt; a third one. I've used two different computers (P4 and A64), three different
&gt; hard drives, three different types of RAM, and several versions of BOINC. The
&gt; sec/TS has always crept upwards at a regular rate throughout each phase of
&gt; each model. It had never occurred to me before reading this that this wasn't
&gt; what was supposed to happen.

It's normal for it to gradually creep up, maybe several hundredths to a tenth of a sec/TS. However, it is not normal to go up from 3.1 to 4.3 in ten trickles. Since the sec/TS reported on the trickles under the results is cumulative, a sudden slowdown in running the model will still appear as a "gradual" slowdown. It would show up as more gradual if this sudden slowdown occurred later in the run since there would already be a large accumulated time where it was running at the faster speed. The same slowdown in the 2nd trickle would show up as a sudden slowdown.

A few hundredths to a tenth is normal over the course of a phase or model run. One sec/TS is not.