General problem with BOINC design

Message boards : Questions and problems : General problem with BOINC design
Message board moderation

To post messages, you must log in.

AuthorMessage
Raistmer

Send message
Joined: 9 Apr 06
Posts: 302
Message 26207 - Posted: 22 Jul 2009, 7:46:17 UTC
Last modified: 22 Jul 2009, 7:46:45 UTC

Always when software pretends to be wiser than its users it rather become unusable.
Attempts to restrict user making program untunable lead to program degradation.

Yes, BOINC supposed to run w/o user intervention - excellent goal, indeed.
Can it run in such way now not losing great share of host processing resources? - No it cant.
Can user assist BOINC by tuning it to be more effective and then left it alone? - No, he/she can't.
If one wants to get best performance from his host he damned to constantly do so called "micro management": disable/enable separate tasks, disable/enable network access on so on and so forth.
Why he should do so (and, please, don't argue that BOINC intended to run standalone and no such micromanagement required. Intended != can in real life!)? Because there is no means for user to tune BOINC's scheduling policy. BOINC "knows better". He can only affect on BOINC's decisions in indirect (and often very ineffective and time consuming) way.

What I propose:
Much more tunable options. Sure each such option should have reasonable default value and some easy way to revert to default should be provided, but all such parameters should be tunable by user, and not hardwired in code.

For example:
Current situation with apparent flaw in BOINC scheduler logic.
Can user turn off this "enhancement" to revert to older variant? - No, he can't. If he needs some other bug fixes he should "buy" all these new issues w/o any choice. In best M$ traditions ;)

Another example:
Duration correction factor. Correct time estimations are viable to do proper sheduling, but BOINC sometimes can't correctly estimate this factor. Ok, no problem in that, there are really hard situations (AP task in SETI, hard to predict behavior of current MW GPU opt app, that DCF varies from ~1 to ~20). It's Ok, BOINC has no AI.
But can user use his knowledge to tell BOINC how long this task will run in truth? Only by indirect way via fpops tag AFAIK. Any DCF correction will be owerwritten by BOINC after next task completion.

Each new feature should have separate switch that governs its use.
Only then BOINC will be really powerful and useful instrument, as it intended to be, not collection of bugs as it is perceived now :/
ID: 26207 · Report as offensive

Message boards : Questions and problems : General problem with BOINC design

Copyright © 2024 University of California.
Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.