Index  | Recent Threads  | Unanswered Threads  | Who's Active  | Guidelines  | Search
 

Quick Go ยป
No member browsing this thread
Thread Status: Active
Total posts in this thread: 4
[ Jump to Last Post ]
Post new Thread
Author
Previous Thread This topic has been viewed 792 times and has 3 replies Next Thread
Former Member
Cruncher
Joined: May 22, 2018
Post Count: 0
Status: Offline
Reply to this Post  Reply with Quote 
@techs: On mis-estimating the expected runtime of project tasks.

There is an interesting discussion on Primegrid v.v. their SoB tasks (No, the acronym does not mean that ;>). One line by the project developer says this
"It's a long-standing problem with the way BOINC estimates run times, but recently a second algorithm was added to BOINC which works much better. I've enabled this on the server side, but it won't be active on your computer until either you reset or re-attach to PrimeGrid or we release new versions of the apps."

Does not say which client version, or server version is involved [WCG is per log at 701], but suspect you'd need later 7.4.n clients to have the benefit, the WCG 7.2.47 not supporting this.

The question is: Could you enable this for those clients that do benefit the additional runtime estimator?

There's still the unanswered question how one member ended up with 99 tasks per core on the device, when officially there's a limit of 35 [total WCG]. At any rate, mis-estimating, which is easy with OET particularly has serious buffering effects, the larger a minimum cache is set. (Albeit, WCG has essentially broken the caching function, particularly for faster devices... you can hardly go a day away to crunch offline, and find your host(s) idling.

Oh and reading on the answer is given on the estimation experience..."Yes, the estimates are definitely better in 7.4 build after it runs for a bit..."

And more reading at Primegrid gives the developers answer, and may explain why my 7.4 clients usually have 25-30% more 'elapsed' estimated buffer [per BOINCTasks] than the cache setting has been set at, though the answer says differently.
If you have a relatively new BOINC client, you may notice that BOINC suddenly got better at estimating how much longer your PrimeGrid tasks will take to finish. BOINC will now ignore its internal estimate of how long the task should take and will rely only on the elapsed time divided by the percentage completed (i.e., what we always tell you to do if you want a good estimate.) For this to work, you need (I believe) BOINC client 7.3.18 or later. The 7.2.xx clients that many still use do not support this feature. Please note that this only affects the "Remaining (estimated)" display for tasks that have already started. Tasks that have not started are not affected and may still show erroneous (or wildly erroneous) times. It also won't have any affect on BOINC deciding to download too much or too little work because its time estimates are off. Only started tasks are affected by this new feature. (The is the same functionality you can manually set yourself by using the <fraction_done_exact> tag in app_config.xml.)

Well as Will E found out, we need no WCG to benefit on 7.4... simply add a line in the app_config for the relative app, yes at app level, not WCG wide
fraction_done_exact
if set, base estimates of remaining time solely on the fraction done reported by the app. List-add.pngNew in 7.3.13

Off to do just that for OET, no dang it, for MCM and CEP2 too, just to see if this makes a useful difference to the TTC on latter in particular [have doubts :]
[Mar 12, 2015 1:14:45 PM]   Link   Report threatening or abusive post: please login first  Go to top 
deltavee
Ace Cruncher
Texas Hill Country
Joined: Nov 17, 2004
Post Count: 4890
Status: Offline
Project Badges:
Reply to this Post  Reply with Quote 
Re: @techs: On mis-estimating the expected runtime of project tasks.

<fraction_done_exact/> Works perfectly. Thanks for the info.

edit: Doesn't seem to work on 7.2.47, does work on 7.4.27
----------------------------------------
[Edit 1 times, last edit by deltavee at Mar 12, 2015 2:23:58 PM]
[Mar 12, 2015 2:06:57 PM]   Link   Report threatening or abusive post: please login first  Go to top 
Former Member
Cruncher
Joined: May 22, 2018
Post Count: 0
Status: Offline
Reply to this Post  Reply with Quote 
Re: @techs: On mis-estimating the expected runtime of project tasks.

shock As the last quote in the OP includes "New in 7.3.13" wink

(It certainly gave some rampant CEP2 times, but then when running, the TTC dropped minutes per few seconds update).

Enjoy
[Mar 12, 2015 2:32:08 PM]   Link   Report threatening or abusive post: please login first  Go to top 
Former Member
Cruncher
Joined: May 22, 2018
Post Count: 0
Status: Offline
Reply to this Post  Reply with Quote 
Re: @techs: On mis-estimating the expected runtime of project tasks.

This morning did some calc checking on MCM, OET and UGM, all running for a little, and the (remaining)TTC is indeed to the minute exact when using the progress percent against the elapsed time [the seconds would not interest anyone ;]. Even CEP2 was quasi correct at 54%.

Also, where I noted the buffer total elapsed was always 25-30% above cache setting [per BOINCTasks, projects view time remaining per core], it was now just 1 hour over the 1 day setting. It says just the running tasks TTCs are affected, so it could be a coincidence... more monitoring, on this [developers only] 7.4.42 build.

Of course, WCG Techs can help any member on 7.4 get better projections if they'd implement this server side, if they would confused
[Mar 13, 2015 9:28:34 AM]   Link   Report threatening or abusive post: please login first  Go to top 
[ Jump to Last Post ]
Post new Thread