Message boards :
Number crunching :
Linux Screwing Windows for Credit on HT machines.
Message board moderation
Author | Message |
---|---|
The Gas Giant Send message Joined: 22 Nov 01 Posts: 1904 Credit: 2,646,654 RAC: 0 |
I've been looking through my work unit list for my 3.2GHz HT machine and have noticed that I am getting screwed for credit by Linux hosts. Is this something that other folks are seeing? Is this solely related to benchmarking on HT machines (my benchmarking is lower with HT on than with HT off)? Paul (S@H1 8888) And proud of it! |
Darrell Send message Joined: 14 Mar 03 Posts: 267 Credit: 1,418,681 RAC: 0 |
My observations from the beta, alpha and live is that the linux clients have always requested less credit than the windows clients. It has alot to do with the benchmarks and also do to the fact that linux is a more efficient OS. |
Toby Send message Joined: 26 Oct 00 Posts: 1005 Credit: 6,366,949 RAC: 0 |
There is a known problem with the benchmarks on the linux client. They obviously aren't using very good optimizations when compiling the "stock" core client for linux. Here you can find instructions on how to recompile the core client to help with this problem. It will very nearly double the benchmark scores and bring the claimed credit much closer to the windows client. How "efficient" the OS is really has very little to do with it. No matter what OS you run, seti is going to get somewhere between 95 and 99% of your unused CPU time. Unless of course you include all the spyware that many windows installations are infected with... But that only has a little to do with the OS and a LOT to do with the intelligence of the user - my windows box has never had any spyware :) Also worth noting is that with seti-classic, windows actually processed work units faster than linux. This has nothing to do with the OS and everything to do with the way the client was compiled. A member of The Knights Who Say NI! For rankings, history graphs and more, check out: My BOINC stats site |
Trane Francks Send message Joined: 18 Jun 99 Posts: 221 Credit: 122,319 RAC: 0 |
Screwed for credit only matters if you're in it for the credit. In looking at my own credits for stuff crunched by my Linux box, I'm often surprised to find that I get more credit than I've claimed. |
The Gas Giant Send message Joined: 22 Nov 01 Posts: 1904 Credit: 2,646,654 RAC: 0 |
Look..if your giving credit out then you might as well do it fairly! If a machine running linux takes 19,000 seconds to complete a wu but the same spec'd machine running doze take 21,000 seconds to complete the same wu, then I would expect the same amount of credit to be claimed by both. As the linux machine completed the wu more quickly than the doze machine then overall it will have a higher rac if they claim the same credit. CPDN have it right...each trickle gives the same amount of credit and different machines complete the trickle in different times (depending on GHz and wu complexity) so overall the better machines/os combinations end up with a higher rac. But with SAH and LHC it ain't necessarily the case. Happy crunching, Paul (S@H1 8888) And proud of it! |
Arm Send message Joined: 12 Sep 03 Posts: 308 Credit: 15,584,777 RAC: 0 |
Take a look at this WU - one of the results is made under Slackware 10.0 (9 000 seconds) and the other one under XP (17 275 seconds). Claimed credits are 26 vs. 54. It is fair I think. |
1mp0£173 Send message Joined: 3 Apr 99 Posts: 8423 Credit: 356,897 RAC: 0 |
... and when credit is granted all three will get the "middle" score. > Take a look at <a> href="http://setiweb.ssl.berkeley.edu/workunit.php?wuid=3277716">this[/url] WU - > one of the results is made under Slackware 10.0 (9 000 seconds) and the other > one under XP (17 275 seconds). Claimed credits are 26 vs. 54. It is fair I > think. > > |
Toby Send message Joined: 26 Oct 00 Posts: 1005 Credit: 6,366,949 RAC: 0 |
If everyone would just recompile the liux client we wouldn't be having this discussion. Or even better if Berkeley optimized the default one! Check out this work unit. My linux box 'got screwed' by 2 windows boxes! Here a windows box and an un-recompiled linux both claimed less credit than my recompiled linux client. It seems that HT processors screw up the credits no matter what. And there is no easy way to fix that. BOINC sees 2 CPUs so it runs 2 benchmark threads. But on HT processors they don't play well together and totally trash the benchmark score. From the look of things I would say HT on windows is about as bad for credits as an un-recompiled linux client. I blame intel! Often my linux machine ends up being the middle score that is granted to everyone and it usually isn't too much lower than the windows claimed credit. So I'm doing my part! :) A member of The Knights Who Say NI! For rankings, history graphs and more, check out: My BOINC stats site |
Hans Dorn Send message Joined: 3 Apr 99 Posts: 2262 Credit: 26,448,570 RAC: 0 |
> It seems that HT processors screw up the credits no > matter what. And there is no easy way to fix that. BOINC sees 2 CPUs so it > runs 2 benchmark threads. But on HT processors they don't play well together > and totally trash the benchmark score. I've also got that impression. The benchmarks seem to hog the CPU and give low results, but two seti clients run like a breeze on the Prescott P4 (with HT enabled) Regards Hans |
Toby Send message Joined: 26 Oct 00 Posts: 1005 Credit: 6,366,949 RAC: 0 |
> I've also got that impression. The benchmarks seem to hog the CPU and > give low results, but two seti clients run like a breeze on the > Prescott P4 (with HT enabled) Well this is pretty much just a problem with the HT design. Well... not so much a problem as 'not its best suit'. It is intended to speed things up by being able to run 2 processes at once. But It only has one CPU to do it on so it works best if the 2 processes that are running are doing things in different parts of the CPU. The benchmarks are doing the exact same thing at the same time so the CPU essentially starts behaving like a single CPU. It might help if the 2 benchmarking threads did things in reverse order. One runs the integer tests while the other does the floating point tests and then they swap. Not sure if that would help all that much but it might be worth a shot. *pokes developers* I personally just go with AMD. Get about the same performance, less cost and no HT headaches :) Except for laptops of course - as should be obvious from my avatar :) A member of The Knights Who Say NI! For rankings, history graphs and more, check out: My BOINC stats site |
Petit Soleil Send message Joined: 17 Feb 03 Posts: 1497 Credit: 70,934 RAC: 0 |
> I personally just go with AMD. Get about the same performance, less cost and > no HT headaches :) Except for laptops of course - as should be obvious from > my avatar :) Some P4 HT are doing 20+ WU per day. The AMD FX53 is not capable of processing as many WU in that time. For the same price I'd rather have two or three P4 HT then one FX CPU. You can buy 2 complete DELL P4 2.8 GHz HT boxes for the price of the AMD FX 53 CPU alone. For $300 you have the choice between an AMD64 3400 or a P4 3.4 GHz HT. I would bet on the P4 HT. What do you think ? |
©2024 University of California
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.