Message boards :
Number crunching :
a strange phenomenon
Message board moderation
Author | Message |
---|---|
Send message Joined: 8 Oct 08 Posts: 4 Credit: 4,833 RAC: 0 |
rencently I noticed a starnge phenomenon about milkyway home.Everytime when I turn up my machine,the WU that run first can run very fast,like completing 1 percent in 2 seconds,so,within 5 minutes,it will finish.after the first WU is completed,the second WUs run much slower,somthing like 0.002% per second.I wonder why?it seems not because those are different WUs,I remember I had a WU which took me 4 hours to complete the first 30% and less than 5 mins to finish the last 70%.do any of you have the phenomenon? is there anyway to have milkyway maintain its initial speed? looking forward to answer。 |
Send message Joined: 21 Aug 08 Posts: 625 Credit: 558,425 RAC: 0 |
There are 3 different lengths of tasks here. Tasks that start with "gs_371" are the shortest running tasks. Tasks that start with "gs_373" are the "medium length", although typically on the order of 30X the short tasks. Tasks that start with "gs_372" are the longest running tasks and are about 50X the short tasks. One additional thing to bear in mind is that the computation on a task is really "complete" here at what you will see in your BOINC client as "50% complete". Once a task gets to that point, it will complete very quickly after that... Also, there may be stages of a task that are faster than others. The efficiency of the application is a sore spot right now. The performance may increase and the computational efficiency may level out across the entire workunit with a newer version of the science application, but that is all up to the project... Brian |
Send message Joined: 8 Oct 08 Posts: 4 Credit: 4,833 RAC: 0 |
thanks for answering me.your answer is very clear,I got it~~ |
Send message Joined: 8 Oct 08 Posts: 4 Credit: 4,833 RAC: 0 |
there is still one thing I dont get it,is the credits awarded vary according to the different length of the tasks? I have just finish a task started with 373,medium length, in 7 hours,if 373 is 30X of 371,732 is 50X 371,then 372 should take about 13hours to complete.is this a bit too long or is this normal?thanks for replying! |
Send message Joined: 15 Aug 08 Posts: 163 Credit: 3,876,869 RAC: 0 |
there is still one thing I dont get it,is the credits awarded vary according to the different length of the tasks? I have just finish a task started with 373,medium length, in 7 hours,if 373 is 30X of 371,732 is 50X 371,then 372 should take about 13hours to complete.is this a bit too long or is this normal?thanks for replying! It's normal for your computers my friend. 372 wu's are awarded with 260 credits. (sorry for my bad english) Best regards. Logan. BOINC FAQ Service (Ahora, también disponible en Español/Now available in Spanish) |
Send message Joined: 21 Aug 08 Posts: 625 Credit: 558,425 RAC: 0 |
there is still one thing I dont get it,is the credits awarded vary according to the different length of the tasks? I have just finish a task started with 373,medium length, in 7 hours,if 373 is 30X of 371,732 is 50X 371,then 372 should take about 13hours to complete.is this a bit too long or is this normal?thanks for replying! Also, something to bear in mind is that the credit/hour rates are generally consistent. For the single 373 result that he reported, the cr/hr was around 19. This is generally the same cr/hr ratio that his Celeron will get no matter if it was a 371, 372, or 373... The moral of the story is that just because it runs longer here doesn't mean you're getting less cr/hr... |
Send message Joined: 4 Oct 08 Posts: 1734 Credit: 64,228,409 RAC: 0 |
This is generally the same cr/hr ratio that his Celeron will get no matter if it was a 371, 372, or 373... The moral of the story is that just because it runs longer here doesn't mean you're getting less cr/hr... Out of interest, Brian, I looked at the current records fopr my small farm of 4 rigs, running from quads to a dual P3. Right across them your statement on credit per hour was correct. My Penny gives 71.4 per hour per core on both 373 and 372 WU. The figures for the core 2 (65nM) was 45.2 per hour per core for both 372x and 373s. The dual Prestona was 19.7 per core per hour for both 372s and 373s. The dual P3 was more productive than I anticipated at 12.0 per CPU per hour on both WU types. NOTE: I have yet to see a 371 WU listed in my BOINC Manager cache. All machines are using the stock client. But I would love to get my hands on an efficiently coded client. |
Send message Joined: 21 Aug 08 Posts: 625 Credit: 558,425 RAC: 0 |
This is generally the same cr/hr ratio that his Celeron will get no matter if it was a 371, 372, or 373... The moral of the story is that just because it runs longer here doesn't mean you're getting less cr/hr... Yeah, I mentioned it because I saw several people aborting tasks like crazy after asking how long they ran. It would indeed be nice to have a more efficient application, but if people were aborting them because they felt like they were going to get less credit/time, then at least as for right now with this project, that's not the case... |
Send message Joined: 8 Oct 08 Posts: 4 Credit: 4,833 RAC: 0 |
thanks.then mine is normal~~ |
©2024 Astroinformatics Group