| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Completed Research
Forum: Help Conquer Cancer
Thread: Discrimination in Work Units based on Operating System of Client |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 14
|
|
| Author |
|
|
RT
Master Cruncher USA - Texas - DFW Joined: Dec 22, 2004 Post Count: 2636 Status: Offline Project Badges: 
|
Discrimination in Work Units based on Operating System of Client
----------------------------------------I have been running several machines on different operating systems and fairly casually looking at the results in the BOINC Results pages. It seems to me, in my less than scientific analysis, that the HCC work units run faster…much faster in my Ubuntu 64 bit machine than they do in my Windows Vista and Windows 7 64 bit machines. Now some of that is likely attributable to the differences in the processors (I7s) (860,920 and OC870). But that does not account for all of it. What I have observed is that the O/C870 (at 3.3Ghz) using Ubuntu runs HCC WUs (8 at a time) in about 1.25 hours whereas the other machines seem to take twice that. I can explain away some of that because the other machines are used on occasion for other things but I can’t explain such a dramatic delta. One of my first stops, to help explain this difference, is to check and make sure that there is no difference between the work units sent to Ubuntu and Windows machines. So…that is why I ask. Thank You in advance for your consideration of this, perhaps naive, question. |
||
|
|
uplinger
Former World Community Grid Tech Joined: May 23, 2005 Post Count: 3952 Status: Offline Project Badges: 
|
There technically are differences between the work units. But the differences are not because of machines type or OS.
The runtime differences you are seeing is correct and that is due to compiler differences between the two operating systems. -Uplinger |
||
|
|
RT
Master Cruncher USA - Texas - DFW Joined: Dec 22, 2004 Post Count: 2636 Status: Offline Project Badges:
|
Thank You very much for your clear and prompt response. I am taken aback a bit by it. And leads me to two additional questions in that regard:
----------------------------------------1) Would it be appropriate for me to extrapolate that into other projects? 2) Are there expected to be any significant compiler changes in the forseeable future to address this delta? Hope I am not imposing... Kind Regards |
||
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
One of my first stops, to help explain this difference, is to check and make sure that there is no difference between the work units sent to Ubuntu and Windows machines. So…that is why I ask. Thank You in advance for your consideration of this, perhaps naive, question. Actual the queue is one single file and the client/OS that happens to pick up the _0 copy determines where the wingman goes to respect the homogeneity of the results come validation time i.e. a quorum 2 needs both tasks to be crunched on Linux/Windows/Mac. Between the OSses/platforms there are slight differences in produced results. The exception is HPF2 with an init distro of 19 and a quorum of 15. These results can have the small statistical differences.
WCG
Please help to make the Forums an enjoyable experience for All! |
||
|
|
RT
Master Cruncher USA - Texas - DFW Joined: Dec 22, 2004 Post Count: 2636 Status: Offline Project Badges:
|
One of my first stops, to help explain this difference, is to check and make sure that there is no difference between the work units sent to Ubuntu and Windows machines. Soâ¦that is why I ask. Thank You in advance for your consideration of this, perhaps naive, question. Actual the queue is one single file and the client/OS that happens to pick up the _0 copy determines where the wingman goes to respect the homogeneity of the results come validation time i.e. a quorum 2 needs both tasks to be crunched on Linux/Windows/Mac. Between the OSses/platforms there are slight differences in produced results. The exception is HPF2 with an init distro of 19 and a quorum of 15. These results can have the small statistical differences. That is interesting Sekerob. Thank You. Your answer helps a good deal, but does not get at the heart of my question. I suppose that what I am trying to decide is whether to convert one or more of my crunchers to Ubuntu in order to crank out more WUs and be more effective in solving the problems we are all her to help solve. Since the crunchers sit there doing practically nothing but WCG, it would seem to make sense to convert them to Ubuntu if this delta Uplinger and I spoke of is A) expected to be real for the forseeable future and B) exists not only in HCC but in the other projects as well. Again Thanks Sekerob.  |
||
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
RT,
----------------------------------------It's a long standing and frequently mentioned known that Ubuntu 64 bit crunches HCC 2x faster (from the day the beta of the last version launched). C4CW is about 60% faster and all the other sciences are marginally to substantially faster on Ubuntu 64. Think I've even made a mention of this in 1 of the FAQs. Some have profiled the various sciences and think that the root of this is the way Ubuntu handles Integers. The compilers too seem to be a substantial factor. One time the techs made mention of having a choice amongst a hundred+. cheers
WCG
Please help to make the Forums an enjoyable experience for All! |
||
|
|
RT
Master Cruncher USA - Texas - DFW Joined: Dec 22, 2004 Post Count: 2636 Status: Offline Project Badges:
|
RT, It's a long standing and frequently mentioned known that Ubuntu 64 bit crunches HCC 2x faster (from the day the beta of the last version launched). C4CW is about 60% faster and all the other sciences are marginally to substantially faster on Ubuntu 64. Think I've even made a mention of this in 1 of the FAQs. Some have profiled the various sciences and think that the root of this is the way Ubuntu handles Integers. The compilers too seem to be a substantial factor. One time the techs made mention of having a choice amongst a hundred+. cheers I did not realize that I was blissful, but I certainly must have been since I was completely ignorant of that fact.  Thanks Again. For now at least, it saves me some money on Windows and will increase the usercount on Ubuntu. Now off to figure out the best way to reconfigure machines to keep the databases and backups safe/secure while converting another I7 to Ubuntu -- I have shot myself in the foot before and am not interested in an encore.  I appreciate your and Uplingersâ help as always. EDIT:> Since the leverage that the compiler has on this and other projects it would seem that since there is a sea of windows and a pond of Unix, a lot of emphasis would be put on the ability of the Windows compiler to produce more efficient code -- since the Operating System should not be involved in the actual computing..only the system functions. Again Thanks for the Info and the ear. ---------------------------------------- [Edit 1 times, last edit by RT at Oct 26, 2010 4:41:37 PM] |
||
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
The Integer (Dhrystone) part of the Q6600 BOINC benchmark for comparison is on Linux Ubuntu 12,800 and in dual boot under Windows 7, all 64 bit 6,700. The Floating point is practically identical, both around 2,250 (Whetstone). That in itself is a sign I'd think.
----------------------------------------
WCG
Please help to make the Forums an enjoyable experience for All! |
||
|
|
RT
Master Cruncher USA - Texas - DFW Joined: Dec 22, 2004 Post Count: 2636 Status: Offline Project Badges:
|
The Integer (Dhrystone) part of the Q6600 BOINC benchmark for comparison is on Linux Ubuntu 12,800 and in dual boot under Windows 7, all 64 bit 6,700. The Floating point is practically identical, both around 2,250 (Whetstone). That in itself is a sign I'd think. I agree completely and yet this implies something else I did not expect -- at all. That is a much higher than (I) expected proportion of Integer work. Wow…what an informative morning this has been. Thank You Sir. |
||
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
RT,
----------------------------------------There was a reason why HCC was chosen as the prime target for GPUfication even before the scientists optimized the application and removed the external loops to make it run 2x faster on Windows and 4x faster from the old Linux version to the new Linux version.
WCG
Please help to make the Forums an enjoyable experience for All! |
||
|
|
|