| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Completed Research
Forum: FightAIDS@Home Phase 2
Thread: Fight AIDS@Home, Phase 2 - project update, Sept. 2016 |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 22
|
|
| Author |
|
|
ErikaT
Former World Community Grid Admin USA Joined: Apr 27, 2009 Post Count: 912 Status: Offline Project Badges: 
|
In the latest project update, the FightAIDS@Home team explains how they are working with the World Community Grid technical team to create a new sampling protocol. This new protocol will more closely predict the binding strengths of potential drugs to their HIV protein targets as determined in real-life experiments.
Thank you to everyone who is contributing to this project! ErikaT |
||
|
|
SekeRob
Master Cruncher Joined: Jan 7, 2013 Post Count: 2741 Status: Offline |
Plainly Superb
----------------------------------------On the acceleration and what this means for the trickling, the below quote infers there will be feedback to the client on how to proceed with a task. How to understand this, and will trickling at all be continued or abandoned? "Asynchronous replica exchange allows information from the different copies to be shared and exchanged among all copies dynamically after short periods of simulations, and this process yields the correct equilibrium statistical physics needed for our analysis." [Edit 1 times, last edit by SekeRob* at Sep 30, 2016 8:23:59 PM] |
||
|
|
supdood
Senior Cruncher USA Joined: Aug 6, 2015 Post Count: 333 Status: Offline Project Badges: 
|
A great development! Thanks for the update!
---------------------------------------- |
||
|
|
Papa3
Senior Cruncher Joined: Apr 23, 2006 Post Count: 360 Status: Offline Project Badges: 
|
Let's hope that trickling dies a quick yet painful death!!
I see from the linked article that although we crunchers are altruistically donating CPU cycles, the write-ups of the results are behind Elsevier & Wiley paywalls. Hasn't anyone here ever heard of open access (https://www.plos.org/open-access/)? Much scientific and medical research is paid for with public funds. Open Access allows taxpayers to see the results of their investment. |
||
|
|
Sgt.Joe
Ace Cruncher USA Joined: Jul 4, 2006 Post Count: 7574 Status: Offline Project Badges: 
|
Let's hope that trickling dies a quick yet painful death!! I see from the linked article that although we crunchers are altruistically donating CPU cycles, the write-ups of the results are behind Elsevier & Wiley paywalls. Hasn't anyone here ever heard of open access (https://www.plos.org/open-access/)? Much scientific and medical research is paid for with public funds. Open Access allows taxpayers to see the results of their investment. The complete paper for the first one can be found here The complete paper for the second one can be found here. The other paper referenced at the end of the article can be foundhere All of them are way over my head. Enjoy your reading. Cheers Edited 3 times to get all the papers
Sgt. Joe
----------------------------------------*Minnesota Crunchers* [Edit 3 times, last edit by Sgt.Joe at Oct 2, 2016 1:42:04 AM] |
||
|
|
Papa3
Senior Cruncher Joined: Apr 23, 2006 Post Count: 360 Status: Offline Project Badges:
|
Let's hope that trickling dies a quick yet painful death!! I see from the linked article that although we crunchers are altruistically donating CPU cycles, the write-ups of the results are behind Elsevier & Wiley paywalls. Hasn't anyone here ever heard of open access (https://www.plos.org/open-access/)? Much scientific and medical research is paid for with public funds. Open Access allows taxpayers to see the results of their investment. The complete paper for the first one can be found here The complete paper for the second one can be found here. The other paper referenced at the end of the article can be foundhere All of them are way over my head. Enjoy your reading. Cheers Edited 3 times to get all the papers They are talking about doing computation for b-Cyclodextrin and ABL kinase for anywhere from 4 minutes to an hour, then contacting the BOINC server to report results and receive new parameters for the next unit of computation. This is all asynchronous, fortunately, but the computation periods for these smaller molecules are so short that it's still trickling-esque. Fortunately, FA@H is studying yuuuge molecules which should require much longer computation periods. So there is good reason to think that crunchers would be able to crunch offline for at least several days at a time!  Thanks for the free links to the papers, Sgt Joe, but I'd still rather see them just dump Elsevier and Wiley completely by doing all their publishing either in one of the PLOS journals instead or in some other open-access journal(s). Not sure why they aren't doing that.  |
||
|
|
wflynny
FightAIDS@Home Scientist Joined: Jun 17, 2014 Post Count: 10 Status: Offline Project Badges: 
|
SekeRob and Papa3 re: trickle messaging -
Trickle messaging may or may not stay when asynchronous replica exchange (AsyncRE) goes live. It is tied to optimal workunit lengths which will be decided upon future conversations with the IBM team. We currently use trickles to know whether a volunteer is making progress towards the 100k step mark and let's us minimize repeated computations. Depending on how long AsyncRE workunits run, trickle messages may still be needed to keep wasted computing time to a minimum. Reducing workunit runtimes significantly will remove the need for trickle messaging but comes at the cost of reduced efficiency of the entire multi-workunit computation. We are still working out what the optimal runtimes will be for AsyncRE workunits; they will definitely be shorter (likely less than 10 hours on average), and cumulative runtime will likely be shorter as well, but I can't say whether trickles are here to stay or not right now. I will update once we get closer to AsyncRE going live. Also, despite the shorter runtimes, all the other aspects of AsyncRE workunits should appear the same to volunteers; the "information sharing" between workunits occurs during the time between when a workunit is validated/assimilated and the next workunit is created and pushed to the queue. Information is not passed during workunit execution via trickle messages. The (simplified) way we are proposing AsyncRE goes like this:
I hope that makes sense. We will likely put out either another news item or a forum post with more details once they are hammered out closer to AsyncRE's release. -wflynny |
||
|
|
Papa3
Senior Cruncher Joined: Apr 23, 2006 Post Count: 360 Status: Offline Project Badges:
|
It seems to me that there are better ways to know if a volunteer is making progress toward the 100k step mark. The BOINC server could easily assign reliability scores to volunteers based on their prior performance. Those who reliably return fully crunched results on time (before the WU's deadline) are scored as highly reliable & they are not to be hit with all the problems that come with trickling.
A opt-in checkbox on the WCG My Projects page could be added so that people can decide for themselves whether or not they want to enable trickling. If they don't, well, it's their contribution and their decision; WCG needs to accept that & move on. |
||
|
|
deltavee
Ace Cruncher Texas Hill Country Joined: Nov 17, 2004 Post Count: 4843 Status: Offline Project Badges: 
|
wflynny,
----------------------------------------Thanks for the explanation. I'm looking forward to crunching the AsyncRE WUs whether trickling or not. And thank you for your work on this very important project.  |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
If the only purpose for the trickles is to provide "status" information regarding WU progress, the WU shouldn't fail if the trickles are received out of order or not at all. Based on the explanation, it sounds like the real interest is in the completed WU and the trickles are actually superfluous. So why does the WU fail if the trickles are not sent in correctly?
----------------------------------------On another subject, reading the papers on the AsyncRE computing architecture, it seems the AsyncRE Manager runs on the same server as BOINC. It raises the concern around backend resources as we see messages from BOINC today regarding deferring scheduler requests due to high load. I would imagine that the AsyncRE Manager will have to handle input from maybe 100,000 cores and this would just add to the existing load. Additionally, it seems like communication between BOINC and the AsyncRE Manager is a potential bottleneck. In my opinion, beta testing will, most likely, not expose these bottlenecks due to it's limited involvement relative to the production environment. Then, " the "information sharing" between workunits occurs during the time between when a workunit is validated/assimilated and the next workunit is created.." What happens when the validator gets stuck and needs the proverbial kicking. [Edit 1 times, last edit by Doneske at Oct 4, 2016 3:13:43 PM] |
||
|
|
|