Agreed.
And what seems to get overlooked is that BOINC is open source.
Whoever really wants a particular feature is free to get out the C++
reference manual and implement it.
And if it's done in a modular way that doesn't mess up
the rest of the code, I'd be happy to check it in.

-- David

Lynn W. Taylor wrote:
> I think we're more than a bit off-topic for the developers' list.  This 
> whole discussion probably belongs in Alpha, and not here.
> 
> I did not say that "power users" weren't of value.  In fact, I agree 
> that people who use and care enough about BOINC to devote the time and 
> study are a major asset.
> 
> Where I have a problem is with "power user features."
> 
> For example, the idea of selecting the mix of projects that can run at 
> any one instant seems problematic.
> 
> Let's take one scenario:
> 
> I have a four core system, and I'm signed up for four projects.  Let's 
> assume for a moment that I have equal resource shares.
> 
> ... and let's assume one of those (call it LHC, for historical reasons) 
> has work sometimes, and not at others.
> 
> Strict interpretation of the rule says that when LHC is out of work, and 
> my queue is empty, I can't use that core for other processing because 
> that would mean two instances of the same project running at the same time.
> 
> So, idle cores and debt that will run away because the rule doesn't 
> allow more than one work unit per project -- and that means no way to 
> make up for projects that are overworked.
> 
> Sure, there are users who could tune that to work, but there are also 
> ways to create a whole lot of edges and impossible situations.
> 
> Martin raised questions about a "Computer Science" view vs. "Computer 
> Engineering" and frankly, as an applications programmer, I'm not sure 
> how to tell how large the cache is, or what other programs are running 
> that could bump my data or code from the cache.
> 
> I do know how to arrange loops to minimize the odds of a cache fault, 
> but that's the limit.
> 
> So, I think it's good to listen to power users (and anyone else) when 
> they start pointing out problems.
> 
> What I worry about are the solutions that many power users "demand."
> 
> While I don't believe a "mom and pop" scenario like you suggested exists 
> (I don't think the common set-and-forget cruncher is more than dimly 
> aware, or even dimly interested, in what BOINC is doing on a day to day 
> basis), is that user better served by more buttons and knobs, or are 
> they better served by a "per processor" DCF?
> 
> Limiting the total number of work units probably makes a lot of sense 
> (as a safety in case of a poor DCF), but how many power users would 
> accept that?  It doesn't seem like a popular idea.
> 
> ... and personally, I'm just a little bit offended by the "We're power 
> users, they'd better cater to us or else!" attitude.  Milo Bloom said it 
> best when he said "The first sign of a nervous breakdown is when you 
> start thinking your work is terribly important."
> 
> But that's my personal opinion, it's not a technical position, nor does 
> it represent anyone but myself.  It does not represent BOINC, SSL, or 
> the Regents of the University of California.
> 
> Anyone who sees a problem and reports it is an asset.  If it's possible 
> to change BOINC so that it solves the problem without help, that's the 
> best possible result.
\
_______________________________________________
boinc_dev mailing list
[email protected]
http://lists.ssl.berkeley.edu/mailman/listinfo/boinc_dev
To unsubscribe, visit the above URL and
(near bottom of page) enter your email address.

Reply via email to