some advise about AS code

[vmzy]

I have got WU that I can't finish it in time(in prefer days) twice. P3860 which prefer 8 days, but my old machine(BTW:it runs only about 6 hours eachday) need about 35 days to finish it , it means that I can't finish it in prefer days.

Now I have an idea ,I think you can change your AS code in this way.

At first , You can get client's PPD(points per day) in 50 days through your stat database.
Second , You can get your WU's prefer PPD from project info. Just like (project points)/(project prefer days) = project prefer PPD.
Third , When AS assign WU, AS server can select WU which prefer PPD is lower than requesting client's PPD.

In this way , AS can send WU according to client's calculating speed in 50 days. So client can finish it in prefer days, and it can reduce project's redundancy(can save much more calculating power, too).
And you can realize this just on your server side easily , don't need updating client side.

Do you agree with me?

P.S. Sorry for my bad English .I'm not good at it ,Because my first language is not English.

[dimilunatic]

That's a pretty interesting idea! :O Thumbs up!

[7im]

The preferred deadline for works units are all set the same, except for a few high priority projects. That means all work units will process at the same rate of speed, with a similar deadline. A small project will have a small deadline, and a large project will have a large deadline... see the FAQ...

http://folding.stanford.edu/English/FAQ-main#ntoc52

So there aren't any work units as you describe, so adjusting the AS code would not help. A slower computer (or a faster computer that only folds part time) will miss the small deadline on a small work unit the same as it would miss a similar large deadline on a large work unit. The deadlines are proportional to the size of the WU, as noted in that FAQ entry. About the only option is to fold more hours each day.

[vmzy]

Oh I see!7im thanks for your replay.
So can FAH add an option "do not accept small deadline WU" in new upcoming V7 client?(Or can FAH mod AS code in order to not send 'small deadline WU' to the clients whose PPD were too slow ?)
Because I really don't want to waste my CPU time on redundancy.And I also don't want to shutdown my old machine delay.I'll shut it down when I don't use it ,otherwise it's waste energy ,it's bad for earth(environment protection).

[bruce]

The client already calculates a number similar to your suggested 50 day average called the performance fraction. It's not exactly the same thing, but at one time there were plans to use it to manage assignments. It turned out to be an ineffective method so it was abandoned.

If the deadline is 50 days and you return the results in 5 days, the PF for that WU would be 90% or 0.90, which is then averaged with other recent WUs. You may still see the number reported in FAHlog.txt but the server ignores it.

In this case p3860 has a preferred deadline of 8 days and a final deadline of 60 days. That does mean that you'll make the final deadline, which does mean your donation will be accepted, although it would certainly have been better if you could have met the preferred deadline.

Please tell us about your hardware. (What CPU and what speed?) There's not a lot that the servers can do to assign work to older machines which only operate 6 hours per day, but it may be time for the Pande Group to reconsider doing something about this issue, whether they adopt your suggestion or something else. I can't promise anything, though.

[vmzy]

For this p3860 I have calculated , it took about 2.5 hours to finish one percent .So even though I open my computer for 8*24 hours(But this kind of behavior is against our company's rule, I may lost my job because of this .I can only open it 9 hours on weekdays) ,It can be only finished about 80% ,which means it can't meet the preferred deadline absolutely. And it wasn't the first time that the AS has sended that kind of 'small preferred deadline WU' to me. I really don't want it happens any more .

[anandhanju]

@mrshirts or a Moderator, Can you please take a look at the preferred deadline for the 3860-3864 projects? They have a preferred deadline of 8/ 6 days but final deadlines of 60 days. Something doesn't seem right with that big a difference. I based my calculation on the formulae I used here but I cannot find them in the FAQ anymore.

@vmzy, Can you post the contents of your FAHLog? This project can use SSE2 optimizations, if your system supports it. Sometimes, due to a crash, optimizations can get disabled even if they are available. The messages in the FAHLog.txt file can show if SSE/ SSE2 optimizations are being used. Also,

Please tell us about your hardware. (What CPU and what speed?)

[bruce]

@mrshirts or a Moderator, Can you please take a look at the preferred deadline for the 3860-3864 projects?

I'm waiting on a response from mrshirts.

I suspect that the issue here is SSE2. Those projects do use SSE2 if you have it, but they run significantly slower if you do not.

In the days when the P4 was picked as the benchmark machine, only a fraction of the donor machines had support for SSE2 so both points and deadlines were based on a P4 with SSE2 disabled. In today's world, SSE2 has become very common. I can see how a final deadline based on non-SSE2 hardware could be radically different than the typical return times.

[vmzy]

My computer is in my office which hardware is:
CPU:Intel celeron D 331@2.66Ghz
memory: DDR 704MB
MainBoard chip:VIA P4M800CE
I have added -forceASM parameter on FAH and I also have changed checkpoint interval to minimum 3 minutes .
In FAH.log you can confirm FAH has used SSE2.Here is the log .

Code: Select all

--- Opening Log file [June 24 01:19:33 UTC] 


# Windows CPU Console Edition #################################################
###############################################################################

                       Folding@Home Client Version 6.23

                          http://folding.stanford.edu

###############################################################################
###############################################################################

Launch directory: E:\Program Files\fah
Service: E:\Program Files\fah\fah
Arguments: -svcstart -d E:\Program Files\fah -forceasm 

Launched as a service.
Entered E:\Program Files\fah to do work.

Warning:
 By using the -forceasm flag, you are overriding
 safeguards in the program. If you did not intend to
 do this, please restart the program without -forceasm.
 If work units are not completing fully (and particularly
 if your machine is overclocked), then please discontinue
 use of the flag.

[01:19:33] - Ask before connecting: No
[01:19:33] - User name: vmzy (Team 3213)
[01:19:33] - User ID: 353435FE66DBAAA8
[01:19:33] - Machine ID: 1
[01:19:33] 
[01:19:33] Loaded queue successfully.
[01:19:33] 
[01:19:33] + Processing work unit
[01:19:33] Core required: FahCore_7c.exe
[01:19:33] Core found.
[01:19:33] Working on queue slot 09 [June 24 01:19:33 UTC]
[01:19:33] + Working ...
[01:19:34] 
[01:19:34] *------------------------------*
[01:19:34] Folding@Home Double Gromacs Core C
[01:19:34] Version 1.00 (Thu Apr 24 19:12:09 PDT 2008)
[01:19:34] 
[01:19:34] Preparing to commence simulation
[01:19:34] - Assembly optimizations manually forced on.
[01:19:34] - Not checking prior termination.
[01:19:36] - Expanded 1266074 -> 3572756 (decompressed 282.1 percent)
[01:19:36] 
[01:19:36] Project: 3860 (Run 573, Clone 10, Gen 12)
[01:19:36] 
[01:19:38] Assembly optimizations on if available.
[01:19:38] Entering M.D.
[01:19:44] Will resume from checkpoint file
[01:19:51] Working on p3860_fkbpabsolute_complex
[01:19:51] Completed 0 out of 300000 steps  (0%)
[01:19:52] Extra SSE2 boost OK
[01:20:32] Resuming from checkpoint
[01:20:33] Verified work/wudata_09.log
[01:20:33] Verified work/wudata_09.edr
[01:20:33] Verified work/wudata_09.xvg
[01:20:34] Verified work/wudata_09.trr
[01:20:34] Verified work/wudata_09.xtc
[01:20:34] Completed 63518 out of 300000 steps  (21%)
[01:23:38] Timer requesting checkpoint
[01:27:03] Timer requesting checkpoint
[01:30:06] Timer requesting checkpoint

Bad news .My computer at home ,which is slightly slower than the computer in my office (and it only runs 3 hours a day) got a p3860 last night .I just don't have more words on AS code now.

My computer at home ,hardware is:
CPU:AMD Sempron 3000+@1.6Ghz(support SSE2 too)
memory: DDR2 498MB
MainBoard chip:NVIDIA C61s

[vmzy]

BTW: Can FAH remove '[01:23:38] Timer requesting checkpoint' information from default ouput?Although it need to update FAHcore. For my experience core8x ,Core7c all have this problem .

[anandhanju]

Can you check Task Manager to see if FAHCore_7c.exe is using 100% of your CPU? Also, can you double-check the actual processor speed with a utility like CPU-Z to ensure that a power-saving scheme is not throttling the CPU?

The checkpoint messages are due to the 3 minute setting. As a side note, a checkpoint interval of 3 minutes may not be the best choice as some users have reported errors/ corrupt data with such frequent checkpoints.

[7im]

The benchmark computer only needs to run 3 hours a day to meet the very lax deadlines. However, the benchmark computer is a 2.8 GHz P4. If you computer is half as fast, it would need to fold twice as many hours per day to meet the deadline. If that computer will not be running fah more than 6 hours per day, it would be best to stop running fah on that computer. Sorry. FAH is a time sensitive project, unlike many of the other @home projects, which are very lax about deadlines.

[vmzy]

@anandhanju
My celeron CPU doesn't has a power-saving scheme .
I have checked with cpu-z 1.51 it's real speed was 2680.2 MHz (20.0 x 134.0 MHz) higher than Stock frequency 2666 MHz.
In Task Manager FAHCore_7c.exe is using 100%, and it used 3.3 hours(cpu time)in 4 hours(wall clock) ,about 82.5% in average .As you know ,I must work with my computer ,so FAH can't use 100% .

@7im
I can meet the final deadline .So I'll go on with FAH.I have took part in FAH science 2003( I also translated your website into Simplified Chinese in 2004 ,introduced FAH to as many chinese people as possibile ), I won't give up.If I can't meet the final deadline I'll update my cpu at home(I have done it once in 2007 , replace my Duron 1.1Ghz with Sempron 3000+) or I'll apply my boss to update my offical computer .
I just want FAH team to consider my opinion to avoid something we can avoid .For example ,avoid sending a special small preferred deadline WU to slow machines .Because there is no need to waste cpu time on redundancy.We should send these special WUs to faster CPUs .

[7im]

Good plan. Keep on folding.

[bruce]

Thank you for your suggestions, vmzy.

One thing that should be mentioned . . . when a server has plenty of WUs, there is no guarantee that a duplicate WU will be issued immediately when the Preferred Deadline passes. The Pande Group doesn't like to duplicate work any more than you do and when the server has a choice of which WU to issue, the way of making that choice can change depending on the conditions on the server. They've never given any details about how the server optimizes the workload, but I know they've considered more than one option.