| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Support
Forum: Suggestions / Feedback
Thread: FIghtAids work units seem to dominate? |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 17
|
|
| Author |
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Thanks for the info - good to have.
|
||
|
|
SekeRob
Master Cruncher Joined: Jan 7, 2013 Post Count: 2741 Status: Offline |
It is my reasoning that those weights are heavily influenced by the momentary work availability of each. Fact is, my 8 core client gets thoroughly saturated with FAH2 if running FAH2/OET1/UGM1/Zika. Add MCM to that set and it's truly "make way" against the shorter running project tasks. It may very well be that FAH2 is 10.4 on the scale of 1-100. but that 10.4 represents 7 fold runtime per task compared to OET1/Zika, i.o.w, have one FAH2 and it has taken 7 spots for OET/Zika, so a work fetch is also 7 times farther apart. Have 8 FAH2 running concurrent. Maybe you can set up an 8 core test system and let it run a week to see for yourself how much runtime each accumulates on the FAH2/OET1/UGM1/Zika profile.
BTW, is your 'weight' the same as feeder priority? |
||
|
|
uplinger
Former World Community Grid Tech Joined: May 23, 2005 Post Count: 3952 Status: Offline Project Badges: 
|
Sek,
The weight marks how many slots out of say 1000 that the feeder has for a given project. So I have them to you in percentages, which means that 10.4 is 104 slots in a feeder of 1000 size. Like you said though, say you have 5 slots available. On your first query with BOINC you have a chance of getting 3 oet, 1 zika, and 1 fah2. The zika and oet finish in 2 hours. You request more work. You still have a 20% chance of getting fah2 and the other 3 slots will finish in 2 hours. So at hour 4 shceduler request, you need 3 work units whlie fah2 is still has another 10 hours and 12 hours to run. You can see by the chances jsut from that, you can get over loaded with FAH2 running on your device. There are other factors into scheduling, but from a pretty simple view, FAH2 may overwhelm your runtimes. The runtimes for each project are at the moment stable. We have to make sure we don't build up a backlog on our servers of results waiting to be sent back to the researchers and that we can download plenty of work to keep the queues full. Thanks, -Uplinger |
||
|
|
SekeRob
Master Cruncher Joined: Jan 7, 2013 Post Count: 2741 Status: Offline |
Thanks for making my case too, Keith, and petition you to make the requested location profile augmentation for FAH2/FAHB. Including previous ayes in this thread, 1001 already.
 (There's some other control, how else do you get Beta and CEP2 to come out first by priority  ) |
||
|
|
SekeRob
Master Cruncher Joined: Jan 7, 2013 Post Count: 2741 Status: Offline |
Keith, has the FAH2 drown-out 'issue' been addressed for multi project selection devices? (A simple yes/no will do).
|
||
|
|
SekeRob
Master Cruncher Joined: Jan 7, 2013 Post Count: 2741 Status: Offline |
Well, I'll answer my own question 3 days later: No, so I've set app_config.xml for fahb max 2 concurrent of 8 cores, pumped up the buffer and whence no more work fits in, FAH2 has near entirely monopolized the queue or cores threaten to go idle, select > abort, which gives a brief flood of the other 3 sciences in the profle. Sorry, but this is much much easier for me (the egotistic volunteer), than continuously having to revisit the device profile and flip on and flop off projects of choice.
|
||
|
|
KLiK
Master Cruncher Croatia Joined: Nov 13, 2006 Post Count: 3108 Status: Offline Project Badges: 
|
Greetings, The weights for projects are as follows on the feeder: mcm1 - 41.67% ugm1 - 1.04% oet1 - 28.13% fahb - 10.42% hstb - 1.04% zika - 12.5% (I know this does not add up to 100%, we have beta apps as well as some rounding) So, if you have mcm1 unselected, you may see more from fahb, this is because with shorter tasks from zika and oet1 when someone with that project selected specifically they will pull more results off the feeder at one time. They request 24 hours worth of work, they may pull many results off the feeder. But if they were to try to pull 24 hours for fahb, they may only pull like 4 out of say 40 sitting on the feeder. There is no plan at the moment to limit the number of results per host for FAHB. The reason that cep2 has that option was due to the high IO for those work units. As for our immovable schedule, the schedule changes from time to time. There has been an enhancement that I have been working on to potentially fix the download issues for members. Phase 1 of that was deployed yesterday without any members noticing or having issues. We are about to utilize a beta test to test phase 2 for this enhancement. I will explain more then. These are just enhancements that I was working on, others on our team are working hard on other important tasks as well. Thanks, -Uplinger THANKS, 'cause revealing this makes a new step of WCG to better understand the science & work behind it...as also gives some of advanced users some info about how projects have been run!  Keep up the good work! |
||
|
|
|