9 Gromacs, each worth 51.4
Moderators: NeilBlanchard, Ralf Hutter, sthayashi, Lawrence Lee
9 Gromacs, each worth 51.4
This seems odd. My 8 processors are working on 9 Gromacs, each and every one worth 51.4 points. And it's not like they are all working on the same protein, but 8 different proteins.
I might not think anything about it except for a thread elsewhere about how work is assigned, and whether it's totally random.
David
I might not think anything about it except for a thread elsewhere about how work is assigned, and whether it's totally random.
David
-
- *Lifetime Patron*
- Posts: 1465
- Joined: Sun Mar 09, 2003 12:27 pm
- Location: Reading.England.EU
That's not impossible: it seems to me you get assigned to a server based on
1) whether or not you have advmethods flag
2) the benchmark score the core calculates for you
If your boxen are set up the same way and similar performance then likely they are all working with server 171.64.122.111 (my 2 Athlons are - both worth 28k points/year) which is currently dishing out the 103x Gromacs. My P3 is slower (about 17k) and is working with server 171.67.89.150 which is dishing 32 point Gromacs to advmethods clients. It all goes wrong when the server is out for a bit and your boxen rotate to another server that only has Tinkers availablle.
[edit]ps - 1 of my boxen just completed a WU. It got its next WU from 171.67.89.149 - project 1024 but still 51.4 points. Figure the bottom line conclusion is their are a lot of 51.4 pointers out there at the moment![/edit]
BTW - that qd utility that is talked about in the finstall script for Linux you aimed me at also has a Windows version. It calculates points/year based on your last 10 WU in queue.dat - I remember a while back you were looking for a more reliable throughput measure than EM etc.
1) whether or not you have advmethods flag
2) the benchmark score the core calculates for you
If your boxen are set up the same way and similar performance then likely they are all working with server 171.64.122.111 (my 2 Athlons are - both worth 28k points/year) which is currently dishing out the 103x Gromacs. My P3 is slower (about 17k) and is working with server 171.67.89.150 which is dishing 32 point Gromacs to advmethods clients. It all goes wrong when the server is out for a bit and your boxen rotate to another server that only has Tinkers availablle.
[edit]ps - 1 of my boxen just completed a WU. It got its next WU from 171.67.89.149 - project 1024 but still 51.4 points. Figure the bottom line conclusion is their are a lot of 51.4 pointers out there at the moment![/edit]
BTW - that qd utility that is talked about in the finstall script for Linux you aimed me at also has a Windows version. It calculates points/year based on your last 10 WU in queue.dat - I remember a while back you were looking for a more reliable throughput measure than EM etc.
How does one tell what project his box is currently working on? I have noticed a wide variation in the wu's that are downloaded to the same machine. Some of them, my boxes can whip through in just a few hours, but others of the same size (ie, 500) will take a couple of days.
Is there a way to get some consistency?
Wendell
Is there a way to get some consistency?
Wendell
It's true that all of my machines would benchmark about the same. The slowest is an Athlon 1800+, the most powerful a P4 2.6.
My Linux machines are assigned to two servers:
My Windows machines are assigned to five servers:
I am looking at Stanford's Server Status page, but it doesn't tell me very much other than they assign fah, bet, and adv.
David
My Linux machines are assigned to two servers:
Code: Select all
# grep "assigned to" /home/fah/ws*/FAHlog.txt
/home/fah/ws000/FAHlog.txt:[13:08:44] - Successful: assigned to (171.67.89.149).
/home/fah/ws001/FAHlog.txt:[04:22:59] - Successful: assigned to (171.67.89.149).
/home/fah/ws001/FAHlog.txt:[21:38:48] - Successful: assigned to (171.67.89.149).
/home/fah/ws002/FAHlog.txt:[16:43:18] - Successful: assigned to (171.64.122.111).
/home/fah/ws002/FAHlog.txt:[08:21:10] - Successful: assigned to (171.64.122.111).
/home/fah/ws002/FAHlog.txt:[23:40:26] - Successful: assigned to (171.64.122.111).
/home/fah/ws002/FAHlog.txt:[03:48:10] - Successful: assigned to (171.67.89.149).
/home/fah/ws002/FAHlog.txt:[06:24:29] - Successful: assigned to (171.67.89.149).
/home/fah/ws002/FAHlog.txt:[13:28:26] - Successful: assigned to (171.67.89.149).
/home/fah/ws002/FAHlog.txt:[05:49:52] - Successful: assigned to (171.64.122.111).
/home/fah/ws003/FAHlog.txt:[14:36:26] - Successful: assigned to (171.67.89.149).
/home/fah/ws003/FAHlog.txt:[11:01:39] - Successful: assigned to (171.67.89.149).
/home/fah/ws003/FAHlog.txt:[07:39:19] - Successful: assigned to (171.64.122.111).
/home/fah/ws004/FAHlog.txt:[22:18:55] - Successful: assigned to (171.67.89.149).
/home/fah/ws004/FAHlog.txt:[07:33:15] - Successful: assigned to (171.64.122.111).
Code: Select all
171.67.89.149
.151
.64.122.111
114
116
David
I specified during Folding@Home "configuration" that I only want to work on FAH (Folding@Home), not GAH (Genome@Home). I have further specified -advmethods on the command line when I start up FAH, saying I prefer to work on "advanced methods", which means Gromacs, one of the two FAH "cores," the other being Tinker.ColdFlame wrote:I must admit I'm still in the dark. What do you mean?
In the past, I have gotten mostly Gromacs, but also an occasional Tinker. The Gromacs have ranged in "value" from 20 points to 70 points. Lately I have been getting 100% Gromacs, each and every one worth the same 51.40 points.
There's nothing sinister about this, I just find it "interesting". The assignment used to be somewhat random, but I seem to have gotten "stuck" on 51.4 point Gromacs. I'm not complaining since these units are worth 600 to 737 PPW on my P4 boxes, 429 to 570 on my Athlon boxes.
With a steady diet of these WU's, my total PPW according to F@H Log Stats is 4578, an average of 572 points per box.
David
Wendell, I think if you answer that question you will have answered my question. I seem to have achieved remarkable consistancy, I just don't know why. Perhaps it's because all my machines are folding 24@7 and therefore producing consistant results, and perhaps I am being "rewarded" for that consistancy by the Stanford servers? I don't know what criteria they use for assigning work so I can't say.wgragg wrote:How does one tell what project his box is currently working on? I have noticed a wide variation in the wu's that are downloaded to the same machine. Some of them, my boxes can whip through in just a few hours, but others of the same size (ie, 500) will take a couple of days.
Is there a way to get some consistency?
Wendell
David