Optimizing an Intel Dual Core E6300 1.8Ghz Machine

Jeff
Jeff
Joined: 17 Jan 06
Posts: 19
Credit: 150,640
RAC: 0

So how can I fix this

Message 61648 in response to message 61646

So how can I fix this issue?

Quote:

WCG is coming through just fine. If you're not seeing it in your stats you probably have a CPID problem.

http://www.boincstats.com/stats/project_graph.php?pr=wcg


Alinator
Alinator
Joined: 8 May 05
Posts: 927
Credit: 9,352,143
RAC: 0

RE: Programs that are

Message 61649 in response to message 61640

Quote:

Programs that are loaded but not actually doing anything will be swapped out to virtual RAM if the physical memory is needed for an active process. As long as you have enough virtual memory you shouldn't notice anything significant.

You often see people giving the reason for slow execution speed as "other (perhaps rogue) processes stealing your cpu cycles". As an example let's assume an EAH result normally takes 10K cpu seconds on your machine. Suddenly you notice that the latest results are now taking 14K cpu seconds. Would you interpret this as a sign that other processes are consuming a significant proportion of your available cpu cycles? If you do then you would be wrong :). Many people seem to fall for this. It's not the cpu time that would change under these circumstances but rather the wall clock time - something that isn't easily visible without a bit of work on your part. You would need to record from your messages tab, the actual wall clock starting and finishing times for every time any crunching was done on a particular result and see how this total elapsed time compared with the reported cpu time.

Fortunately, BOINC does this hard work for you and reports (on each computer's page on the website) the actual cpu efficiency which is the ratio of cpu time to wall clock time. Alinator has asked you a couple of times already to have a look at this value but you don't seem to have noticed since you haven't reported the figure :). Please have a look and let us know.

Agreed, there's a big difference between have a result which took more wall time to complete and when all of a sudden your CPU time for equivalent work goes up or is significantly more than for similar machines.

Unfortunately, the latter case can be a lot more difficult than the former to troubleshoot.

Alinator

archae86
archae86
Joined: 6 Dec 05
Posts: 3,156
Credit: 7,178,114,931
RAC: 741,395

RE: You often see people

Message 61650 in response to message 61640

Quote:
You often see people giving the reason for slow execution speed as "other (perhaps rogue) processes stealing your cpu cycles". As an example let's assume an EAH result normally takes 10K cpu seconds on your machine. Suddenly you notice that the latest results are now taking 14K cpu seconds. Would you interpret this as a sign that other processes are consuming a significant proportion of your available cpu cycles? If you do then you would be wrong :). Many people seem to fall for this. It's not the cpu time that would change under these circumstances but rather the wall clock time - something that isn't easily visible without a bit of work on your part.

Granted that folks are often wrong in saying that--but they are not _always_ wrong.

Running Hyperthreaded on a Gallatin processor (Northwood relative with an added 2M L3 cache), I've noted on reproducible experiments that the process running on the "other" virtual CPU can appreciably raise or lower the reported CPU time for running a benchmark or a BOINC result. I've seen effects both way up to at least 20%, if memory serves.

That much is repeated careful observation, and I'll stand by it (by the way the reported bonus efficiency a year or so ago of running the then current SETI and Einstein science aps on the two virtual CPUs of a typical hyperthreaded P4 was one excellent and widely observed example of such an effect).

Reasons why is more a matter of informed inference and uninformed speculation. The intent of the timer-reporting is to attribute each clock cycle to _one_ of the two currently active processes. But inside the machine, both are often doing something (in fact, that is the whole point of the exercise). So some reporting bag-squeezing can come into play, if the "wrong" process is charged time for a cycle, it pushes the time up for one and down for the other. Anti-efficiency is also sometimes apparent (one of Akos Fete's Einstein code revisions suddenly poisoned HT utility to a remarkable degree). Probably this is due to interlocks required to preserve state integrity giving the effect of more time lost to switching overhead than is gained by extra parallel resource use.

Less malevolently, the ability of the hardware usefully to share its resources (in effect, really to make useful progress in parallel on work required by both processes) clearly varies a lot with the code. Hence real work productivity can go up _and_ reported time go down when a good pairing is active.

I'm not saying that any of this applies in the case of Jeff's situation. Just nit-picking the sweeping dismissal of such effects by a post from Gary Roberts.

Jeff
Jeff
Joined: 17 Jan 06
Posts: 19
Credit: 150,640
RAC: 0

I take back what I said

Message 61651 in response to message 61650

I take back what I said before about 900 in a day. It is really still only 300-400 because the climate project added an extra 500 as it does not update everyday. It seems even with turning everything off aside from the essentials still doesn't do the trick. I am going to install that program that was mentioned and see if there is something else eating up the cpu time.

Alinator
Alinator
Joined: 8 May 05
Posts: 927
Credit: 9,352,143
RAC: 0

You never did tell us what

You never did tell us what BOINC is reporting for the performance metrics. They're one of those host items the project only displays to the owner.

Alinator

Jeff
Jeff
Joined: 17 Jan 06
Posts: 19
Credit: 150,640
RAC: 0

Hmmm, where can I find that

Message 61653 in response to message 61652

Hmmm, where can I find that info?

Alinator
Alinator
Joined: 8 May 05
Posts: 927
Credit: 9,352,143
RAC: 0

On the Computer Summary page

On the Computer Summary page at the bottom just above the Location selection drop down.

Alinator

Jeff
Jeff
Joined: 17 Jan 06
Posts: 19
Credit: 150,640
RAC: 0

I am confused now. Where

Message 61655 in response to message 61654

I am confused now. Where would I find this? Is this info in the Boinc client itself or on the Boinc stats page? I have looked in both locations and cannot find it...

anders n
anders n
Joined: 29 Aug 05
Posts: 123
Credit: 1,656,300
RAC: 0

Hi Jeff If you look at

Hi Jeff

If you look at your account here at Einstein and then klick on the line
"Computers on this account". Next klick on the Computer ID you want to se
and the info is there.

Anders n

Jeff
Jeff
Joined: 17 Jan 06
Posts: 19
Credit: 150,640
RAC: 0

Hi thanks, I thought maybe

Message 61657 in response to message 61656

Hi thanks,

I thought maybe this is what you wanted to see, but thought I had posted this information earlier, the site to see it anyway, maybe you guys couldn't access the link...

Ok here is the info only for Einstein.

Total Credit 2,421.45
Recent average credit 104.21
CPU type GenuineIntel
Intel(R) Core(TM)2 CPU 6300 @ 1.86GHz [x86 Family 6 Model 15 Stepping 2] [fpu tsc pae nx sse sse2 mmx]
Number of CPUs 2
Operating System Microsoft Windows XP
Professional Edition, Service Pack 2, (05.01.2600.00)
Memory 2047.11 MB
Cache 976.56 KB
Swap space 2719 MB
Total disk space 117.19 GB
Free Disk Space 108.03 GB
Measured floating point speed 1590.76 million ops/sec
Measured integer speed 3286.61 million ops/sec
Average upload rate 4.43 KB/sec
Average download rate 380.76 KB/sec
Average turnaround time 0.55 days
Maximum daily WU quota per CPU 72/day
Results 24
Number of times client has contacted server 512
Last time contacted server 24 Mar 2007 15:22:48 UTC
% of time BOINC client is running 88.8714 %
While BOINC running, % of time work is allowed 88.7853 %
Average CPU efficiency 0.951668
Result duration correction factor 0.362694

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.