| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Community
Forum: Chat Room
Thread: CPU Time Production Back on Track |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 18
|
|
| Author |
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
June 29,06 was the first workday, where cpu time production was equivalent to a pre HPF2 project start day.... global 122 years CPU time.
----------------------------------------Speculating, that we still got piles of 'inconclusives' to resolve as well as having to recoup the initial erroring and aborting WU's, AND this much discussed 60% CPU de-throttle for UD Agent Crunchers (80+ percent of total crunching done at WCG), we seem to be getting there.... Curious is, one week after closing HPF1, still 1 year CPU time was turned in for this finished project yesterday. Any other interesting WCG stat analysis...you know stats can be sliced and diced every which way.
WCG
----------------------------------------Please help to make the Forums an enjoyable experience for All! [Edit 2 times, last edit by Sekerob at Jun 30, 2006 1:43:50 AM] |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
 Well, my analysis of your analysis tells me it's bunk, nothing more than a mix of ambiguities and meaningless comparisons intended to give the impression of A when B is far more likely. The global 122 years CPU time? Where did you get that? Does it mean anything? I think not because UD reports wall clock time. So, OK, let's say it's actually wall clock time. Then what does it mean? Nothing! It's the FLOPs that count, Sek. So are the FLOPS back to normal? Well, if all other factors have remained the same and the only difference is the throttle down to 60% then there is absolutely no way the FLOPS are equal to what they were. Have all other factors remained the same? I dunno, we don't see the right kinds of stats to assess that. Convince me they've changed significantly. [Edit 1 times, last edit by Former Member at Jun 30, 2006 2:43:56 AM] |
||
|
|
David Autumns
Ace Cruncher UK Joined: Nov 16, 2004 Post Count: 11062 Status: Offline Project Badges: 
|
LOLThe Throttle only affects UD and only on HPF2 so like Dagorath says UD measures wall clock time so "Run Time" won't be affected. Work Units at the mo seem a tad longer than the norm hence the fall in results generated And then there's the Summer Holiday effect as most of the world pop live in the Northern Hemisphere and also colleges and Uni's wind down The Throttle in UD is designed to get more folk onboard and not to loose those that go "Why is my CPU at 100% that can't be good" and switch off. We'll keep more laptop crunchers this way too. A couple of days of teething troubles but a lifetime of crunching especially with the honesty shown by IBM in known issues. We're in it for the long term Sekerob Keep your eyes peeled for the next project Dave (p.s the "my CPU at 100% that can't be good" statement I can vouch makes no difference to your CPU, it's designed to do this and with over 10 years of run time to my name I haven't lost a single CPU to crunching. With a constant temperature it's likely to last longer than with the temperature cycling in normal operation as it speeds up and down. Over time this will have a physical wear and tear as the CPU die contracts and expand with each heat cycle. Just my humble opinion)  |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
when you refer to having to recoup "initial" errors, as far as i know many of these errors have still not been solved.
if and when they do get resolved it would be nice to hear about what happened. the impression is that HPF-2 was started too soon. |
||
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
halfcard, i think if you read the messages of knreed, alther and the like, you'd appreciate more the issues....there's no way testing this for all conditions in the 6 weeks WCG done it "in-house"...you'd need a supercomputer for that...the grid is one! In a few days with thousands of units processed, 99% very likely came to surface. Without doing this, there'd never be progress in this science......
----------------------------------------.....BUT, this was not the topic of discussion, its dicing stats and reading the teeleaves., whether off-the-wall-clock or CPU-Clock
WCG
Please help to make the Forums an enjoyable experience for All! |
||
|
|
retsof
Former Community Advisor USA Joined: Jul 31, 2005 Post Count: 6824 Status: Offline Project Badges:
|
Curious is, one week after closing HPF1, still 1 year CPU time was turned in for this finished project yesterday. BOINC is set to a one week deadline, but the UD interface still has a 3 week deadline. When that expires, that should be the end of points for HPF1.Well, even that may not be the end. If the workunit is completed within the deadline, I think it can be sent in later. WCG might have to make a decision to close out HPF1 workunits later and that will finally be the end. grid.org is tight on server space, and they close out their cancer workunits every week, and only give a few days to turn them in. That is a penalty for slow computers that have several workunits lined up to work on.
SUPPORT ADVISOR
----------------------------------------Work+GPU i7 8700 12threads School i7 4770 8threads Default+GPU Ryzen 7 3700X 16threads Ryzen 7 3800X 16 threads Ryzen 9 3900X 24threads Home i7 3540M 4threads50% [Edit 2 times, last edit by retsof at Jun 30, 2006 11:37:43 AM] |
||
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
retsof, is this serious....BOINC crunchers since about may 19 on a 3 copy quorum for HPF1/FAAH with 1 week return requirement and UD crunchers still being on 3 week return digestive tract......oh well, i never understood that to be the case....missed that entirely....wont keep me awake
---------------------------------------- 
WCG
----------------------------------------Please help to make the Forums an enjoyable experience for All! [Edit 1 times, last edit by Sekerob at Jun 30, 2006 1:12:11 PM] |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
I use FLOPS here to mean floating point operations rather than floating point operations per second.
If we accept the notion that FLOPS is the only meaningful measure of production and forget about CPU hours, wall clock time, returned results and the higher than normal error rate for a moment then I don't believe production in the past week is what it could have been. With the hosts that do 80% of the work throttled back to 60% speed, it's ludicrous to think we've produced the FLOPS we could have produced in the past week. On the matter of whether or not WCG should have throttled UD to 60%, my position is YES they should have BUT they should have emailed all the UD hosts and informed them. I am firmly convinced there are thousands of UD hosts that are now running at 60% when they could be running 100% without overheating. I believe thousands of hosts aren't even aware they are at 60%. While I can accept the fact that we had to sacrifice some production for the greater good, we have lost more than was necessary to achieve that greater good. One simple email might have prevented that. It's really hard for me to justify spending time recruiting new members when WCG dumps production in the trash bin so carelessly. |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Hello Dagorath,
I do not know the reasoning behind this decision, but I will make a guess. So far, WCG has made a habit of starting things before the big announcement. This allows a lot of bugs to be detected early. This time we are certainly finding a lot of bugs. I expect that the next newsletter will contain full information about the throttle. Lawrence |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
David, I fear you are spreading misinformation. WCG have retrofitted the throttle to FAAH, and it will be implemented on all future UD projects.
As an interesting aside, I tried running UD throttled to 60% side by side with BOINC. Result? A 30/70 split. UD took exactly half of its target, and BOINC took the rest. I was impressed. (Note: this isn't a normal configuration; I just wanted to test that bad things wouldn't happen if someone did it by accident.) |
||
|
|
|