| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Completed Research
Forum: Discovering Dengue Drugs - Together - Phase 2 Forum
Thread: c-types arrived |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 93
|
|
| Author |
|
|
knreed
Former World Community Grid Tech Joined: Nov 8, 2004 Post Count: 4504 Status: Offline Project Badges: 
|
You do not need to worry about that error and you will be granted credit. It turns out that there is a valid condition that occurs and the application exits prior to a file being created. However, the output file in question is currently marked as required so BOINC treats its non-existence as an error.
We are going to be modifying the behavior so that it is not marked as an error and that it goes through normal validation. It will just be a very short workunit. |
||
|
|
knreed
Former World Community Grid Tech Joined: Nov 8, 2004 Post Count: 4504 Status: Offline Project Badges:
|
About the DCF and flops estimates. We are currently having to manually estimate the flops for these workunits. This has always been error prone with estimates differing substantially from reality. Because it is so hard to estimates the flops, we have scripts in place that estimate the flops based on how long it has taken to compute results recently returned. This is dramatically more accurate and results in better consistency between our projects. Unfortunately, this mechanism doesn't start to operate until there is a steady stream of work for the project. We are getting closer to having that steady stream but it will be a couple of more weeks before we get there.
|
||
|
|
Hypernova
Master Cruncher Audaces Fortuna Juvat ! Vaud - Switzerland Joined: Dec 16, 2008 Post Count: 1908 Status: Offline Project Badges: 
|
AT LAST!!! YESSSS!!!
----------------------------------------I received three WU's  .erlc_ a204_ sr78a1_ 1-- Ceres In Progress 23.03.10 09:59:04 02.04.10 09:59:04 0.00 0.0 / 0.0 erlc_ a210_ pr78b0_ 0-- Jupiter In Progress 23.03.10 07:18:58 02.04.10 07:18:58 0.00 0.0 / 0.0 erlc_ a216_ pcb006_ 1-- Saturn In Progress 23.03.10 05:48:35 02.04.10 05:48:35 0.00 0.0 / 0.0 The first one is being crunched by the 980X. Cross fingersall goes well  P.S. How strange.I crunch hundreds of WU's per day, and the simple fact of receiving those three so much awaited WU's made me cry   of joy. |
|||
|
||||
|
NixChix
Veteran Cruncher United States Joined: Apr 29, 2007 Post Count: 1187 Status: Offline Project Badges:
|
I got 4! [hop - skip] I got 4! [hop - skip]
---------------------------------------- Cheers   [Edit 1 times, last edit by NixChix at Mar 23, 2010 3:35:01 PM] |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
About the DCF and flops estimates. We are currently having to manually estimate the flops for these workunits. This has always been error prone with estimates differing substantially from reality. Because it is so hard to estimates the flops, we have scripts in place that estimate the flops based on how long it has taken to compute results recently returned. No kidding. My 38 minute WU was estimated at 46 hours originally [Edit 1 times, last edit by Former Member at Mar 23, 2010 3:46:04 PM] |
||
|
|
Mathilde2006
Senior Cruncher Germany Joined: Sep 30, 2006 Post Count: 269 Status: Offline Project Badges: 
|
No kidding. My 38 minute WU was estimated at 46 hours originally 84 hours here  And it was no A-type.   [Edit 1 times, last edit by Mathilde2006 at Mar 23, 2010 3:50:09 PM] |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
I finally got 2 WUs - erlc_a149_pqb010 and erlc_a149_pqb002 - but I still have a bunch of work in front of them so I don't know when they'll be run.
|
||
|
|
Ingleside
Veteran Cruncher Norway Joined: Nov 19, 2005 Post Count: 974 Status: Offline Project Badges: 
|
tluehr, There could be a risk of overcommitment only if the techs would suddenly underestimate the C-type WUs as much as they have overestimated them and if you have pushed your cache size to the max in hope of getting more of the overestimated ones. Hmm, it seems you're overlooking the real problem... So, to maybe make things a little clearer... User configures to only run DDDT2-tasks, and crunches maybe 50 C-tasks, each with initial estimate of 20 hours cpu-time. But, the actual cpu-time of these C-type-tasks is, let's say, only 1 hour... So, after finishing the collection of 50 C-type-tasks, the DCF has dropped to... 1 hour / 20 hours = 1/20 = 0.05 User configures to run other WCG-work besides DDDT2, and example gets a bunch of FAAH-tasks that let's say takes 6 cpu-hours to run... But, due to the very low DCF due to all the C-type-work, each FAAH-task has now estimated cpu-time of ... 6 hours * 0.05 = 0.3 hours... Even if user has a fairly small cache-size of 0.5 days, he'll get 40 FAAH-tasks per core... With each in reality taking 6 hours to run, he's suddenly got 10 days worth of FAAH-work per core... With a little larger cache-size of 1 days, the same computer will suddenly have 20 days worth of FAAH-work, and obviously will have no hope to return all this before the 10-day deadline... I trust the techs for not doing this second mistake, and you, on your side, you would not have done this big cache increase, would you? As long as the cache is kept to a reasonable size there is no such risk. What will happen if/when the weight of type C jobs is corrected is that we will slowly get more until their number is in synch with reality. The real effects of C-type work is something like this: 12-Mar-2010 01:06:04 [World Community Grid] Computation for task erlc_a107_pcb007_1 finished 12-Mar-2010 01:06:04 [World Community Grid] [dcf] DCF: 1.167979->1.075098, raw_ratio 0.239167, adj_ratio 0.204770 12-Mar-2010 01:06:06 [World Community Grid] Computation for task erlc_a107_pcb008_1 finished 12-Mar-2010 01:06:06 [World Community Grid] [dcf] DCF: 1.075098->0.991512, raw_ratio 0.239235, adj_ratio 0.222524 12-Mar-2010 01:06:13 [World Community Grid] Computation for task erlc_a107_pcb014_0 finished 12-Mar-2010 01:06:13 [World Community Grid] [dcf] DCF: 0.991512->0.916332, raw_ratio 0.239721, adj_ratio 0.241773 12-Mar-2010 01:06:25 [World Community Grid] Computation for task erlc_a106_pca001_0 finished 12-Mar-2010 01:06:26 [World Community Grid] [dcf] DCF: 0.916332->0.848755, raw_ratio 0.240553, adj_ratio 0.262518 12-Mar-2010 01:06:43 [World Community Grid] Computation for task erlc_a106_pca000_0 finished 12-Mar-2010 01:06:43 [World Community Grid] [dcf] DCF: 0.848755->0.788052, raw_ratio 0.241728, adj_ratio 0.284803 12-Mar-2010 01:07:29 [World Community Grid] Computation for task erlc_a108_pca012_0 finished 12-Mar-2010 01:07:29 [World Community Grid] [dcf] DCF: 0.788052->0.733731, raw_ratio 0.244841, adj_ratio 0.310692 12-Mar-2010 01:07:29 [World Community Grid] Computation for task erlc_a109_pcb014_0 finished 12-Mar-2010 01:07:29 [World Community Grid] [dcf] DCF: 0.733731->0.684842, raw_ratio 0.244841, adj_ratio 0.333694 12-Mar-2010 01:07:36 [World Community Grid] Computation for task erlc_a109_pcb013_1 finished 12-Mar-2010 01:07:36 [World Community Grid] [dcf] DCF: 0.684842->0.640888, raw_ratio 0.245304, adj_ratio 0.358191 (snip) 12-Mar-2010 02:21:07 [World Community Grid] Computation for task erlc_a111_pca013_0 finished 12-Mar-2010 02:21:07 [World Community Grid] [dcf] DCF: 0.140888->0.239497, raw_ratio 0.239497, adj_ratio 1.699909 12-Mar-2010 02:21:36 [World Community Grid] Computation for task erlc_a110_pca009_0 finished 12-Mar-2010 02:21:36 [World Community Grid] [dcf] DCF: 0.239497->0.239738, raw_ratio 0.241906, adj_ratio 1.010060 12-Mar-2010 02:21:44 [World Community Grid] Computation for task erlc_a110_pca010_0 finished 12-Mar-2010 02:21:44 [World Community Grid] [dcf] DCF: 0.239738->0.239984, raw_ratio 0.242202, adj_ratio 1.010281 So, as the log shows, the DCF kept on dropping, and dropped 4.87x from the pre-C-type-DCF. Meaning, if downloads any non-C-type work afterwards, the computer can suddenly get roughly 5x more work than should have got, making it easy to become hopelessly overcommitted.  "I make so many mistakes. But then just think of all the mistakes I don't make, although I might." |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Yes, Ingleside, that was exactly what I had in mind. A good example!
|
||
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
Let's see, you want to be rated reliable? If so you need to have a cache setting of under 2 days [for WCG ergo this then applying to all attached projects]. I run 1 day cache. Standard deadlines at WCG is 10 days (including C types). The 4.87 times 2 is still less then 10 and we can reasonably expect that such managed switching is not done on a perpetual basis. Even for those that play the
----------------------------------------I'm sure Ingleside can come up with some constructed schema to have that fall over too, but those that have this are usually those that have brought it upon themselves. Can't possible counter program BOINC for all eventualities and have it always outsmart the outsmarter, without fail. Of course it's always the clients fault or the projects fault, and never the customer who's king, but not always. edit: cache
WCG
----------------------------------------Please help to make the Forums an enjoyable experience for All! [Edit 1 times, last edit by Sekerob at Mar 23, 2010 7:45:55 PM] |
||
|
|
|