Posts by Peter Ingham

1) Message boards : Number crunching : granted credit < 10% of claimed credit (Message 55517)
Posted 4 Sep 2008 by Peter Ingham
Post:
Evilsizer, the credit system compensates for your time spent, because you will have completed more models. And rather then waste the machine power of having two machines do the same task, just to get credit straight, Rosetta averages all of the related (but unique) tasks and grants you credit based on that.

But there does seem to be a problem with credit of late. DK agreed to look in to it. But change or correction will not be immediate.


I have also found a very low granted to credit ratio.

I had a small farm of Q6600 based systems available, and these all consistently got very low granted ratios for most WU\'s (around 10%). Suffice to say, these systems no longer process Rosetta WU\'s, which is a shame.

2) Message boards : Number crunching : new to rosetta using q6600 (Message 51092)
Posted 30 Jan 2008 by Peter Ingham
Post:
I think it\'s from here:

http://boincstats.com/stats/host_cpu_stats.php?pr=rosetta&st=0

Although the figure currently given is 0.001784. Doesn\'t seem right.


My posting http://boinc.bakerlab.org/rosetta/forum_thread.php?id=3890&nowrap=true#50960 gave the source as http://boincstats.com/stats/host_cpu_stats.php?pr=rosetta&st=200&or=12


As at now it appears at rank 316 in http://boincstats.com/stats/host_cpu_stats.php?pr=rosetta&st=300&or=12 with 0.001823


Cheers
3) Message boards : Number crunching : new to rosetta using q6600 (Message 51091)
Posted 30 Jan 2008 by Peter Ingham
Post:
Thanks for the advice, I\'d love to be contributing more, but the credits seen for other projects are much, much higher.


I know some people do distributed computing for credits, but I encourage you to also consider the scientific and medical value of your computer\'s work. In the end, credits are just numbers; medical breakthroughs can improve the lives of millions.


As an experiment, 1 of the Q6600\'s I have was set to run Rosetta 100% share, EIST was disabled & the system conf\'d for 4 cpus, 100% usage etc. CPU-Z confimed ongoing 2.4Ghz Clock rate & BoincMgr often showing >99% on all 4 active running cores, even during interactive usage. After a week or so it was consistently getting a daily credit of about 500.

The Q6600 systems I have access to (currently 6 of them, a few more on the way) are now primarily processing SIMAP. I figure this is contributing (indirectly) to the same scientific and medical benefits as Rosetta. I\'d hate to support something with virtually senseless aims!! Chasing credits is not my primary motivation, but it does add a little interest.

Why not support Rosetta, regardless of the credits? The stats seem to indicate that there is something fundamentally askew with Rosetta running on Q6600\'s. The averages across all Q6600\'s on Rosetta are way down compared to other projects where Q6600\'s rate much better on average. Based on my own claimed vs granted, many WU\'s are getting 10% of claim, which means a lot of wasted resource. If other projects are more suitable, then in terms of contribution to mankind\'s scientific knowledge it makes sense NOT to use them for Rosetta.

Cheers
4) Message boards : Number crunching : new to rosetta using q6600 (Message 50960)
Posted 25 Jan 2008 by Peter Ingham
Post:
Looking into this from another perspective, looking at http://boincstats.com/stats/host_cpu_stats.php?pr=rosetta&st=200&or=12 shows that for Rosetta, the Q6600 is ranked as 274th for credit per cpu-second (much higher ranking in other projects - see some earlier posting of mine in this thread).

Using the quoted average of 0.01945 Rosetta credits per cpu second, we find that a Q6600 @2.4 giving 100% of 24hrs to Rosetta should average 672 credits per day.


on my calculator, using your number 0.01945*3600*24, I get 1680 credits


Sorry, my mistake, it should read:

Using the quoted average of 0.001945 Rosetta credits per cpu second, we find that a Q6600 @2.4 giving 100% of 24hrs to Rosetta should average 672 credits per day.

0.001945 * 3600 * 24 * 4 = 672
5) Message boards : Number crunching : new to rosetta using q6600 (Message 50935)
Posted 24 Jan 2008 by Peter Ingham
Post:
Looking into this from another perspective, looking at http://boincstats.com/stats/host_cpu_stats.php?pr=rosetta&st=200&or=12 shows that for Rosetta, the Q6600 is ranked as 274th for credit per cpu-second (much higher ranking in other projects - see some earlier posting of mine in this thread).

Using the quoted average of 0.01945 Rosetta credits per cpu second, we find that a Q6600 @2.4 giving 100% of 24hrs to Rosetta should average 672 credits per day.
6) Message boards : Number crunching : new to rosetta using q6600 (Message 50933)
Posted 24 Jan 2008 by Peter Ingham
Post:

i know a few people running multiple Q6600s with credit as expected. you should get something in the region of 1700 RAC from a 2.4GHz Q6600 on 24/7. If you\'re getting less then it might be because the CPU is throttling down... is it definitely running at 2.4GHz when Rosetta is the only thing running?


I\'d love to get 1700RAC from a single Q6600!! I practice I\'ve been seeing < 400 with Rosetta on 66% share.

For an example of low returns, see
http://boinc.bakerlab.org/workunit.php?wuid=121649462 It looks like somehow the credit system thinks some serious cheating is going on and scaling the granted credit severely.

I\'ll try setting Rosetta to 100% share on a Q6600 box & see what happens.


Yikes! you are really getting the shaft. Here is a link to an Acer Aspire only running on one core.
http://boinc.bakerlab.org/rosetta/workunit.php?wuid=122537142


Thanks for the advice, I\'d love to be contributing more, but the credits seen for other projects are much, much higher.


One of the Q6600\'s has been set so Rosetta is getting 100%, XP power setting was changed to always on (BIOS still has EIST enabled) & I have confirmed via CPU-Z that the clock is not dropping below spec 2.4GHz. It has been running like this with 4 cpus enabled for several days.

Still seeing a lot of WU\'s with low ratio of claimed to granted. Credit over last 24Hrs for this system 482. A WU returning 1 decoy took 9843.75 cpu seconds.

I will try setting EIST off in BIOS.

The only slightly strange characteristic of these systems is that they have 4GB physical RAM of which only 3 is usable (due to PCI memory mapping etc).
7) Message boards : Number crunching : new to rosetta using q6600 (Message 50793)
Posted 18 Jan 2008 by Peter Ingham
Post:

i know a few people running multiple Q6600s with credit as expected. you should get something in the region of 1700 RAC from a 2.4GHz Q6600 on 24/7. If you\'re getting less then it might be because the CPU is throttling down... is it definitely running at 2.4GHz when Rosetta is the only thing running?


I\'d love to get 1700RAC from a single Q6600!! I practice I\'ve been seeing < 400 with Rosetta on 66% share.

For an example of low returns, see
http://boinc.bakerlab.org/workunit.php?wuid=121649462 It looks like somehow the credit system thinks some serious cheating is going on and scaling the granted credit severely.

I\'ll try setting Rosetta to 100% share on a Q6600 box & see what happens.
8) Message boards : Number crunching : new to rosetta using q6600 (Message 50749)
Posted 17 Jan 2008 by Peter Ingham
Post:
In theory, there is nothing you need to do to take advantage of all 4 cores.

However, I have discovered that Rosetta gives unusually poor credit to systems running Q6600\'s (in some cases around 10% of what would be expected). Due to this, I\'m only running Rosetta (at low share) on one Q6600 until this is resolved.
9) Questions and Answers : Web site : Granted Credit < Claimed Credit (Message 50192)
Posted 30 Dec 2007 by Peter Ingham
Post:
One of the \"Interesting\" WUs is http://boinc.bakerlab.org/workunit.php?wuid=118012427.

Note that there are 2 different values shown for Granted: 4.11 and 46.52

It would appear that the actual granted credit was 4.11


Also of interest (but possibly not related to the low Granted figures):

The Q6600, across all cpu types covered by boincstats is #49 of 3583 in points per cpu-second (http://boincstats.com/stats/host_cpu_stats.php?pr=bo&st=0&or=12)

For Rosetta, it is ranked at 140 of 1722 http://boincstats.com/stats/host_cpu_stats.php?pr=bo&st=0&or=12

This may be related to some interaction between the q6600 internal architecture and the rosetta application resulting in lower relative performance.
10) Questions and Answers : Web site : Granted Credit < Claimed Credit (Message 50156)
Posted 29 Dec 2007 by Peter Ingham
Post:
At the time of the runs described, Rosetta was getting ~70% of available CPU on essentially idle systems.

The Claimed credit seems to be in line with expectations for these CPUs, but the Granted is much, much lower.

Observed behaviour consistent as described across 3 x Q6600\'s and 3 x P4 2.8-3.2 (running with HT off).

P4\'s have been running for many months, Q6600\'s for a week or so.


Due to minimal credit, the Q6600\'s have now been adjusted to contribute primarily to other projects.
11) Questions and Answers : Web site : Granted Credit < Claimed Credit (Message 50100)
Posted 27 Dec 2007 by Peter Ingham
Post:
I have a curious issue with Granted...

I have 3 x Quad Core Q6600\'s, 2 x P4 2.8 & 1x P4 3.2.

The Quad\'s are often struggling to achieve the same RAC on Rosetta as the P4\'s.

The Quads have Result Duration Correction Factors in the region of 2.9, the P4\'s in the region of 0.5


On other projects, the quads are getting several times the RAC of the P4\'s, which seems in line with the published performance comparisons.

Any ideas as to what is going on??
12) Message boards : Number crunching : Report stuck & aborted WU here please (Message 11092)
Posted 21 Feb 2006 by Peter Ingham
Post:
FYI, I\'ve just aborted a WU () stuck at 1% after 175K seconds

Name: PRODUCTION_ABINITIO_RANDOMFRAG_1vcc__309_441
WU: 9337995
ResultID: 11582797
13) Message boards : Rosetta@home Science : How Is Rosetta Different Than Folding@home? (Message 759)
Posted 29 Sep 2005 by Peter Ingham
Post:
At the risk of triggering defensive counter-attacks, I'll posit another suggestion for encouraging users to take up the project (or at least not have them abandon inviolvement at a later date).

My suggestion is to try and keep RAM and Disk requirements as low as practical. Excessive use of these will impact the machine owners use (enjoyment) of the systems they have purchased. Using lots of unused CPU time is fine, slowing a system down (especially if no warning is given) is not. Yes I am aware of the stated requirement for 512MB RAM to run the R@H project, perhaps the implications (and symptoms) of running it on systems with overcommitted RAM could be more clearly set out.

Boinc will get a bad name and will be banned from more installations if it becomes demonstrated that clients are, in fact, causing other than trivil system degradation. When this happens, no one will win.

I'm NOT taking a swipe at R@H, I AM trying to offer suggestions to make your project one of thse that users will prefer to donate their resources to.

14) Message boards : Rosetta@home Science : How Is Rosetta Different Than Folding@home? (Message 589)
Posted 27 Sep 2005 by Peter Ingham
Post:
There are a number of index web sites - get your project listed on these (google for "Boinc" and "distributed computing")

The BOINC home page http://boinc.berkeley.edu/ does not currently list your project!

The Boinc Wiki should get an entry as well!

Best wishes






©2020 University of Washington
http://www.bakerlab.org