| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Support
Forum: Suggestions / Feedback
Thread: Request for better distribution of large and small tasks |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 7
|
|
| Author |
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Please consider a system where large, complex folding, tasks only run on fast PC's and small tasks run on slower PC's.
I have a mix of slow and 'normal' speed PC's. Currently only large tasks seem to be distributed. So on my slow PC I have task running that will take more than 100 hours to complete. I know of someone who has a task running that is expected to complete in 2400!!! hours. I suggest that you create a system where, once the speed of a device is know, it gets task that are appropriate for it. How can you expect to keep people interested in this project if their PC's have to work on such endless tasks? Rgds, HendrikJo |
|||
|
||||
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
True, it is a good idea an I do agree with you, however, to my knowledge, the rosetta program uses random numbers, so they can't really predict which WU's will be long or short or inbetween... yep. See ya around.
  |
|||
|
||||
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
 Hi HendrikJoI can appreciate your frusteration, however, WCG advised us that it is not possible to estimate the length of time a Work Unit will take to complete, as there is no way to calculate the complexity of the folding process in advance Here is Rick Alther's Answer OK. I'll try to answer most of the questions in this thread. * We need to calculate thousands of structures for each protein. * Each workunit contains a full protein sequence. * Each workunit calculates only a fraction of the total structures necessary. Thus, the same protein is "split" into some number of workunits. Each calculates n number of structures. We can make the workunits, on average, take a shorter amount of time by having them each calculate fewer structures. However, there's a tradeoff. We are tuning it. We have to get some measurements back to see what we have. Also, remember this: * The time it takes to calculate each structure is non-deterministic. Thus we can't predict how long a workunit will take. Even workunits from the same protein sequence can vary greatly (some taking hours, others days). * This is a HARD computational problem. That's why we put it on the grid. This is unlike SETI or prior UD projects, and for you veterans of these projects, this project probably seems "strange". With SETI and Cancer the amount of work that had to be done with each workunit was more or less predictable. On this project it is not. As for how we know how far along the workunit we are, we know how many structures a particular workunit is supposed to calculate. When a structure is complete, we store the results, checkpoint, and increment the progress bar. Finally, the grid.org workunits are generated exactly the same way ours are. ---------------------------------------- Rick Alther World Community Grid Application Developer I hopes this helps to answer your question Regards |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Also, what is so bad about having a big task? I'm currently having a task that will take me about 75h (even though I have a 3.4 GHZ P4), but there is no difference between 3 tasks that take 25h or 1 task that takes 75h. They will both give the same amount of points.
|
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Hello all out there.
This is my 1st post ever, so please correct / help me if I’ve got this wrong. I’ve been with this program a while, and now that I’ve reduced the noise of my computer, I’m “allowed” to run it 24/7 ! Even so a WU sometimes takes a couple of days, while at other times I have done 4 x WU in a day. This doesn’t really bother me, as we get the points for “CPU time taken” but I can well understand a lot of people running it as a screensaver or on a slow computer can be very disillusioned when work units take weeks to complete. Although I haven’t seen this mentioned, if a WU takes more than 3 or 4 weeks to return I can quite imagine that when it finally is returned, it will not contribute to ‘solving the problem’ as the same WU will have been sent out to another person in the meantime who will possibly have processed / returned it sooner. That’s 2 x reasons why it must be of benefit to send different WU to fast computers and slower computers, BUT how ?? THIS IS MY IDEA. As all the computers that are on the grid are scored against a standard, the grid already knows which computers are faster than others. (I think I read somewhere that there is also a maximum allowed speed of processing, so some ‘over clocked’ systems are not getting the full benefit of there faster effort) SO; The grid should treat these faster computers as a 1st level group and send ALL WU to them, but to no other computers yet. After processing each WU it is returned to the grid and the time taken for THAT WU is logged. (It doesn’t matter to these computers if it took ‘an hour’ or ‘a week’ they have carried out the processing and will be allocated the points due and the computation of the WU has been of benefit.) NOW, when the grid comes to send out that SAME WU it knows better who it should be sent to. SLOW WU goes to a fast computer and FAST WU goes to the slow computers. This should give all completed WU a better chance of been counted and give members a real sense that they have contributed. I hope my idea holds water, and if it does, that it can be incorporated into a practical solution. PLEASE let me know. SKIPSOFT1 |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Hi Graham,
these kind of projects need the help of a lot of all kind of people. By assigning scorepoints to the amount of work they let their PC's do, WCG creates a element of competition. I think that this is OK. People like to be rewarded for their contribution. However, if a large, slow task is holding back you from getting more point in an overseeable amount of time, they may stop their contribution. I wonder if it is possible to have the client returning the amount of time spend each n hours run time, or at least once a day, especially for large tasks.  Rgds, HendrikJo |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Hi Hendrikjo,
As my team mate already recalled : * The time it takes to calculate each structure is non-deterministic. Thus we can't predict how long a workunit will take. Even workunits from the same protein sequence can vary greatly (some taking hours, others days). Since the WCG is not aware how long a Wu will take, I do not think it is possible to do a "once a day update" for larger tasks.....  Besides the fact of unawareness of the chrunching time needed I do not think that the WCG is willing to solve the additional statistic workload with all questions/remarks we will come up then..... for instance what should be done if a WU becomes outdated/cancelled? No result is returned, should the added runtime be reversed or not?I'll think this would make the system unnescesary complex  At the moment I got two Big ( 100+ hrs ) and one Killer WU ( 396 hrs ) on my four machines so I know where I talk about......  |
||
|
|
|