| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Support
Forum: Website Support
Thread: Have we hit an internal bottleneck? |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 7
|
|
| Author |
|
|
David Autumns
Ace Cruncher UK Joined: Nov 16, 2004 Post Count: 11062 Status: Offline Project Badges: 
|
Here's a graph of member numbers and number of years contributed
----------------------------------------http://www.wcgwiki.org/membersandyears.cfm all things being equal as the numbers go up in the top graph the bottom one should show as an upward curve as more crunching gets done by more people every day but it is straight as a die - a steady plod not showing the benefits of the extra crunchers. Is the work being completed being throttled by IBM in some way such as the bandwidth available to dish out and collect the work units? Just increasingly curious Dave  |
||
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
Work Trottled.... As you know the work for BOINC counts on the day of validation and for UD immediately when returned. We discussed your graph before so wont comment on that again.
----------------------------------------
WCG
Please help to make the Forums an enjoyable experience for All! |
||
|
|
David Autumns
Ace Cruncher UK Joined: Nov 16, 2004 Post Count: 11062 Status: Offline Project Badges:
|
Sek I know these things but the question remains unanswered
----------------------------------------as I say I'm increasingly curious |
||
|
|
knreed
Former World Community Grid Tech Joined: Nov 8, 2004 Post Count: 4504 Status: Offline Project Badges: 
|
We are actually experiencing the largest difference between 'work returned' and 'work validated' that we have ever had. The website reports 'work validated'. I have some internal reports that show the daily work returned.
----------------------------------------Normally the trailing 7 day difference is only 2-3 years of runtime. At the moment however we have over 55 years of runtime differences. Some of this is due to the increased length of HPF2 workunits (and the longer durations). Some of this is due to over-assignment of work due to incorrect estimates for FAAH and HFP2 (those are largely resolved now but the clients still are working through the disruption). Part of me suspects that some of this might be some folks queuing up work to return when the bunny race starts on Saturday. We shall see :-) Regardless, we will start to see the catch up between these numbers and we will start to post some record dailly stats when that happens. [Edit 1 times, last edit by knreed at Feb 28, 2008 5:39:53 PM] |
||
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
Membership rose by 7.71% over the measured period whereas the CPU years rose by 15.51 percent. It continues on it's path of the rate of CPU year growth being 2x that of membership. If CPU years are following same behavior as climate we're in for a hot spring.
----------------------------------------(think jal2 computed last time 5 % members and 10% cpu time increase)
WCG
Please help to make the Forums an enjoyable experience for All! |
||
|
|
David Autumns
Ace Cruncher UK Joined: Nov 16, 2004 Post Count: 11062 Status: Offline Project Badges:
|
Thank knreed we'll wait and see
----------------------------------------Sek the membership growth and the CPU time growth do not bear comparison between these two graphs, they are not related. What I would expect though is that with the 25,000 extra members would show the cumulative crunch time to bend upwards instead of going on straight. It might just be my monitor or my heavily corrected eyesight but I think it might even be tailing off  |
||
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
Sorry for not getting back sooner da, it's exactly that. On a flat panel screen it bends up, assuming my ruler is not bend too. Suggest to degauss your CRT and use those buttons to calibrate the display to show 'straight' lines. If that does not do the trick, print and use a known item of which you're sure it's not crooked..... or if you program allows, insert a trend line.... this one is a polynomial.
----------------------------------------So, the only significant bottleneck was getting the matching work units back to achieve quorum particular for way longer HPF2 jobs with now 18 day deadline. The record was hitting on Monday with 231 CPU years and a running 7 day total record of 1487.34 CPU years went into the logbook of achievements.
WCG
Please help to make the Forums an enjoyable experience for All! |
||
|
|
|