| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Completed Research
Forum: The Clean Energy Project - Phase 2 Forum
Thread: CEP2: Notable Efficiency Improvement with W7 32+64 since May'11 patches |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 13
|
|
| Author |
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
First saw this on me W7-32 laptop. Suddenly 97% and better efficiency per BOINCTasks, overnight, when running on 1 of 2 cores, winged by Clean Water, AFTER applying the May patches on top of SP1. Then decided to boot the Linux quad into W7-64 and apply the patches, pre-SP1 **, and low and behold, this OS too with continuous running on 2 of 4 cores, water on the side showing 98.8-99.1% efficiency, when left alone, and water doing 99.4% or better on the side.
----------------------------------------Using the WCGDAWS [one more plug for the pirogue coded tool  ] and now having run a whole bunch of tasks, the analysis says:-Linux-64 Natty had an average of 7.01 hours CPU time per task since May 10 [35 tasks at 96% efficiency] -W7-64 has 7.09 [after 17 tasks at ~98.5% efficiency]. Too small a variation on CPU time with knowing that times are slowly increasing for CEP2, and W7 being the last to be tested. As an aside, Clean Water stats in dual boot and available in 64 bits showed: -Linux-64 Natty: 97 tasks for 4.51 hours. -W7-64: 42 tasks for 4.49 hours Efficiencies not even marginally different for either 2 OSses. Notably, when USING the device, the Windows efficiency drops much more than the Linux. If just web browsing or writing whilst listing to music, Linux remains choice, for me. Based on these figures though, I'll be crunching on W7-64 for these 2 sciences on the dedicated quad cruncher (Q6600, stock 2.4Ghz, no SSD, BOINC Data dirs and VM in separate partitions, 3.3GB usable RAM). 99% efficiency on this old clunker :O --//-- PS, HCC continues to be 10% faster under Linux in it's 32 bit science release on these 64 bit OSses. ** [SP1 upgrade failed as GRUB2 seems to have mangled the MS identifiers] edit: corrected CW stats for Windows. [Edit 1 times, last edit by Former Member at May 20, 2011 7:35:42 AM] |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Hi SekeRob,
thanks for the info - that may be interesting for a number of crunchers. Cheers Your Harvard CEP team |
||
|
|
sk..
Master Cruncher http://s17.rimg.info/ccb5d62bd3e856cc0d1df9b0ee2f7f6a.gif Joined: Mar 22, 2007 Post Count: 2324 Status: Offline Project Badges: 
|
Sounds like SP1 is a worthy upgrade for Win7 CEP2 crunchers.
|
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Let me emphasize as what I wrote in the first post, that SP1 is NOT what kicks it. The quad runs without SP1, which is merely a collection of all patches that MS released previously. It's specifically the May patches that with and without SP1 kicks the file system in [much] higher gear, again for this device!
Getting more audacious, forced 3 CEP tasks to run concurrently with 7 minute intervals at start to get past the Jobs 0/1/2 setup phase and let them run, specifically the session not locking itself, rather signing out i.e. no active sessions. The outcome of these, operating as a pure grid cruncher: World Community Grid 6.40 cep2 E202203_268_C.24.C20H11NOSSe.00726857.3.set1d06_1 07:27:43 (07:24:24) 3Min19Sec Lapse 99.3% Eff. World Community Grid 6.40 cep2 E202200_625_C.25.C20H11N3S2.00254968.1.set1d06_1 08:21:22 (08:16:46) 4Min36Sec Lapse 98.5% Eff. World Community Grid 6.40 cep2 E202200_593_C.25.C21H11NOS2.00285842.4.set1d06_1 07:25:19 (07:21:11) 4Min8Sec Lapse 98.45 Eff. The ratio of CPU time for this device running on W7-64 and the overall project mean time has remained same as what it was with Linux-64 i.e. there's on CPU time no meaningful difference in the actual processing... it's just that the efficiency for this device cannot be matched by Linux, kernel 2.6.38.9. The top job ran at 99.3%, so now increased the number of cacheable CEP2 tasks so that these tasks get to use 2-3 cores, gaps filled with Clean water. The 2 Year CEP2 goal will be reached much quicker :D Looking forward to test a true 64 bit compile, when it comes. :P --//-- |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Interesting.... I've had my quad-cruncher reduced in the number of CEP units it can have at once. (after discovering that letting it have all 4 cores at once wasn't such a hot idea)
I might try increasing it now to see what happens. |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Upped the game a little further:
Cores: 4 Cache: 0.9 days Allowed CEP2 in cache: 9 Shared projects: Clean Water (CW) OS State: No-one signed in! With 3 concurrent getting a consistent efficiency in the range of 98.6 to 99.15%, exclusively with CW on the side. Now will allow it to go hands-free up to inclusive 4 concurrent if queue sequence so makes this happen. This was almost the case today, except the repair job was server aborted before it could get started, making the core free for another CW. Crunching on needing 37 days to reach one 1 year (discounting for present PV jail-mates). --//-- |
||
|
|
kateiacy
Veteran Cruncher USA Joined: Jan 23, 2010 Post Count: 1027 Status: Offline Project Badges: 
|
Hi Sekerob,
----------------------------------------Interesting comparison. What kind of hard drive are you using? I have been running CEP2 on all 4 cores in my AMD Phenom II X4 610e machine since January and am getting around 98.5% efficiency with Ubuntu Lucid. (I just looked at the last 6 WUs uploaded, and the exact average for them is 98.52%.) I'll add more data to my spreadsheet over the next couple days so I have a better average. This is with an SSD, though. I don't have Windows to install on this machine, so I can't make a dual-boot comparison. I'm not running CEP2 on my Atom machine (with a slow HDD) at this point, both because it's running HCC full tilt until I get my Sapphire there, and because the bigger CEP2 WUs don't get very far on it. When I was running CEP2 on it, I noticed that the fewer jobs it got through in a WU, the more "efficiently" they appeared to run. No doubt that was just because there were fewer stops to write output files. The only place where that issue would show up with faster machines that always finish CEP2 WUs is that if one exited early with an RC 100 or whatever, it would look extra efficient. Edited to fix typo in number (98.72 corrected to 98.52)  [Edit 2 times, last edit by kateiacy at Jun 3, 2011 11:56:48 PM] |
||
|
|
Bearcat
Master Cruncher USA Joined: Jan 6, 2007 Post Count: 2803 Status: Offline Project Badges:
|
I think it has to do with using all cores on a PC that degrades the CPU vs actual time. Currently running 4 cores out of 8 real cores with about a 5 to 7 minute difference. When I max out all 8 cores (3Ghz harpertowns), my difference go between 10 to 20 minutes. Since the heat is up, will have to stay with 4 cores, bummer.
----------------------------------------
Crunching for humanity since 2007!
 |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
kateiacy,
----------------------------------------The BOINC datadir sits in its own partition on a 320GB middle of the road HD, the VM has one of its own too [Linux default] to minimize fragmentation. Think it's a Seagate Barracuda... not sure, nearing 5 years in age. Searching found this link http://www.seagate.com/ww/v/index.jsp?vgnexto...0VgnVCM100000dd04090aRCRD The type codes look very familiar. Now, if W7-64 can squeeze this out, no reason to contemplate going the SSD route. [OT]Anything in US prices need multiplication by 2 to buy out here when it gets to techie stuff, but me next Lappie will have it by default. Ogling the new Samsung 9 series [very dear]. Will arrive here in probably 2012 ;>). Has to last at least another 5 years.[/OT] At any rate, with 9 CEP2 in cache still not had a ''coincidence'' where 4 ran concurrent... always CWs doing incursions. :D --//-- [Edit 1 times, last edit by Former Member at Jun 4, 2011 9:08:59 AM] |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Well Well, eventually 4 ran concurrently, but only for 1:08 hours before the first 2 finished near simultaneous. A new CEP2+CW started at that time which was a good load test... still uploading when started the 3rd job in a new task. Both completed on 98.9% efficiency. In meantime the now oldest has climbed to ... 98.9% efficiency at 6:40 CPU hours and rising.
Changes applied: Cores: 4 Cache: 0.95 days Allowed CEP2 in cache: 10 Shared projects: Clean Water (CW) OS State: No-one signed in! Increased the cache numbers to force a sooner backfill of CEP2 tasks, else too many CW arrive in the mix before reporting of completed CEP2 jobs. Maybe I'll suspend all queued CW tasks tonight to force a longer "CEP2 only" environment and see what BOINCtasks logs. If lapse remains under 5-6 minutes I'll let it go full blast till the next new project/phase launches... want that Blue at minimum. For now, I'm pretty swayed on W7-64 being the best OS for this hardware and these 2 sciences. --//-- |
||
|
|
|