knreed, thanks for that explanation about the throttle. It is and was obvious that the system is used for fine-tuning the distribution system in place. It's nice to know some extra background about the system used.

Since the thread was started, I did some thinking about what could be done to improve the system. Nothing more. Just adding my two Eurocents. If it is possible, and if some of the stuff as suggested gets done, then I think it would be a nice addition.

As I know I don't have to program anything of what I suggested I allowed myself to suggest whatever came to mind. It gets read by you and the other people at WCG, and it's up to you to consider whether it's feasible or not. There was no intention to bugger anyone on a daily basis to ask for the implementation. Maybe sometime there will have to be changes made, and some of the stuff can be integrated easily later on. Maybe it will never happen. Whatever way, I don't know the extent of the to-do list that you guys have and I'm pretty happy sitting on this side of the grid

As user it's nice to have a project like WCG that offers multiple projects, with the possibility to select 1, more or all whilst knowing that the computers don't begin to run idle because of a full disk, ill scientists, end of funding and stuff like that. Everything extra is just that: extra.

Thanks for taking the time to explain!

I recently worked out the most efficient tasks to run for each of my systems, in terms of points granted. The optimal WCG Projects differed from one system to another. I have setup 4 profiles to try to accommodate each computer, but there are 6 projects and I have 5 different projects I wish to run on different machines, so I am one Profile short.
If we could allocate WCG Projects to individual systems it would make this easier.
While most systems seem best at CMD2 and then HPF2 (I a€™m not presently running NRW), my i7 was best at HFCC and my Opteron 1354 was best at FA@H.

Would it be possible to have a "benchmark" project that would run whenever new projects were added? Something that would run a "representative subset" of each active project, and report the efficiency of each task type to the server. These "results" could then be fed into the database and used to update the device profiles of users who had opted in to an "Allow WCG to optimise project selection" policy. In fact, the benchmark project wouldn't even have to be available to those who hadn't.

That way one might be able to bolt it on to the existing system, without fiddling with the internals of the server.

Could that work or am I just being naive here?

Fair enough. The discussion was about improving throughput by distributing the tasks to the hosts which could be expected to finish them fastest. If that doesn't benefit the scientists, then I don't know what would.

Personallly, I'm happy to provide whatever spare cycles I have and I couldn't care less about badges and points. While I have some personal reasons to dislike cancer in particular, I'm happy to contribute to whatever project the WCG deems worthy of inclusion. Most of my CPU time seems to be spent on muscular dystrophy, which is fine with me. I don't even know what it is, but I'm sure understanding it better and possibly curing it would benefit mankind.

As it appears that certain configurations crunch certain projects more efficiently, I think it would make perfect sense to take advantage of this. Whenever someone donates something, they like to think it is being put to the best possible use.

The question was whether it would be possible to retrofit the current system with a mechanism to take advantage of these differences. It was suggested that it would require major rewrites to the server software, but I was wondering whether something a bit less ambitious could still provide enough of an improvement to make it justifiable.

My humble suggestion was to create a "benchmark" project that could be made available to new clients (and to older clients whenever new projects are added). This would generate a ranking of each project's suitability for the individual client and this could in turn be used to generate preferences tailored to the client. Assuming the preferences are stored in a database, the server wouldn't even know anything was going on. It would distribute work units and receive results, just as it does all day.

As for "benefitting the crunchers"... Well, I guess it's their forum. I find quite a few arguments here downright silly, but to each his own... They are contributing resources that would otherwise be wasted, and that should be appreciated regardless of their personal motivations for doing so.

The concept of specialisation has served mankind well for quite some time now. I fail to see why it could not be applied to our computers.



I'm not sure I follow you here, but that probably makes us even... It wouldn't benefit us as computer owners in any way whatsoever. It would, however, benefit the scientists in their work to rid the world of certain diseases. And, of course, in the long run it would benefit those suffering from them.

And i refuse to accept that it would somehow be "unfair" to cure AIDS just because cancer had not yet been cured. That line of reasoning just makes no sense. Unless, of course, you're in Scandinavia...

Now, my computer is happily crunching away at whatever task the server hands it. It would continue to do so even if the server could distribute the work according to where it got the best throughput. Of course, if the server ran out of work to distribute, things would be different. But that, incidentally, is one of the reasons behind WCG: to ensure a continuous flow of work from the different (sub-) projects.

I do realise that the (human) resources available are limited. As far as I can tell they are doing a sterling job, and long may they do so. I was merely proposing a more lightweight alternative to some of the approaches suggested earlier (which typically involved rewriting the scheduling mechanism). I still think the idea of specialisation has a good deal of merit, but not if it requires rewriting BOINC. So, I was wondering whether any thought had been put into an implementation that could be bolted on to (or into) the existing system.

I don't know if it qualifies as a "really, really good idea", but it's certainly not as bad as one might think after reading your objections. It is one way of increasing throughput without requiring major changes to the server internals. That's all. I'm not trying to sell you anything here, just trying to be a bit constructive.