On Sep 25, 2009, at 5:20 PM, Lynn W. Taylor wrote: > > > Paul D. Buck wrote: >> On Sep 25, 2009, at 8:34 AM, Lynn W. Taylor wrote: >>> Even I can figure out that returning work late would have a >>> negative effect on credit. >> And I never suggested that it would not. Variety in the queue and >> in the work selected to be run on the processor has nothing to do >> with deadlines. It has to do with Resource Share a bad allocation >> system that has no meaning to a person that is single project >> oriented and we are now throwing under the bus in a variety of ways >> and pretending we aren't. > > Variety in the cache at any given instant is simply not workable. > > If I have two projects, and a 10000:1 resource share, there is no > way that my 0.01% project can always have a work unit in the cache.
Another straw-man argument. Yes, if you have a completely unbalanced resource share I agree with you. I did not cover all cases and permutations because people complain too much when I do so. But BOINC does not support even reasonably balanced resource shares... even cases as simple as even shares on 8 projects with an 8 core processor. Again, today, you are more than likely to have as many as 6 cores working on project A, the other two on B and little or no work from half of the remaining projects ... and this is even if you have selected the projects that virtually always have work. In this case the logical expectation is that I would pretty much see one core working on one project of the eight all the time ... we have seen nothing close to this for at least a year and it gets worse it seems with each generation ... > It only gets worse when you have more than two. It should get better, or at least, as I stated, I never saw these kinds of issues before with versions in the 4 and 5 generation and even the early 6... and I was just as broad in my project attachment as I am now ... > In my opinion, the problem comes from trying to always keep the > cache full. If people set their cache to four days, and BOINC has > to go to the most-overworked project to get work, then we've got > what we've got. > > You simply cannot have everything. You can't have a perennially > full cache, and have all projects represented in that cache all at > the same time. Again, you misinterpret and try to bend my words. I did not say all projects all of the time. And, yes, I have used cache settings from 0 to 10 days and the project selection in a larger cache should naturally be easier to spread than a smaller cache. But you are correct, larger caches are not any better than smaller for variety now these days ... BOINC is agnostic in its ability to screw up the selection. Next, with an i7 920 which is not even the fastest processor burns through tasks like no one's business ... the problem is that BOINC is still pretty well tied to a model of a single processor and it has never been looked at with today's faster and "wider" systems with an eye towards making the system more efficient with these systems. Add in GPU support and some of my systems are running upwards of 16 tasks at the same time. When I got my first four processor system (a Dell HT dual Xeon) John was still working on the Resource Scheduler and when I found an anomaly in the way it was scheduling he did come up with a fix and a private build which I proved solved the issue ... that fix was never approved and we still suffer those kinds of consequences today ... But based on my experience there is almost always, with a cache of a day or two, more than enough time to process work from each and every project one is attached to, in accordance with Resource Share were the work fetch pulling work appropriately and the Resource Scheduler running it likewise. As I stated in my most egregious bad example there is no circumstances where 6+ CPDN tasks should have been pulled were my Resource Shares and recent run times being used appropriately. Understand I basically use BOINC as a screen saver and watch the tasks flow across the system in my secondary screen ALL DAY LONG ... my autistic bent gives me a sensitivity to patterns ... BOINC's systems are bent, have been bent for some time, and getting worse ... but even when proven with numbers ... well ... The saddest part is that I am not the only one that sees these problems but the pervasive attitude at UCB and the fan club is that "if I squint hard enough so I don't see it there cannot be a problem". Now, if you like running one project I have no complaints with you ... but BOINC, based on comments (historically) from the SaH boards where there are a lot of people that are SaH only types are none to thrilled with how BOINC supports them. I tend most of the time (I am in an exceptional mode these last couple months only running I think only 5 projects) to be on the opposite end usually with 50 plus project attached ... if the project has work I am there running it ... and I can assure you that it does not support my needs either ... Lastly, why cannot I have a full cache and tasks from more than one or two projects? There is quite literally no conceptual reason. Again, I will agree that as the number 170 something in world standings with only 5 systems that I am running slightly more capable systems than many ... but that also means I have a perspective that might be useful were it heeded in any manner on how BOINC does work in this arena ... > You have to let go of something. In BOINC design it sure seems to be that we let go of everything for expediency's sake and very seldom look to see if we can solve problem "X" without compromising anything at all ... instead we junk rules and policies that have served us well on whims ... > In my opinion, half of the cache should be set aside for the project > that is owed the most. That means if you're crunching for LHC > (which is perenially out of work) and all of the other projects have > been overworked, your four day cache is a two-day cache. I proposed something of this sort to replace Resource Share and well, let us just say that the support was less than tepid. As part an parcel of *THAT* proposal it was intended to begin to move us in the direction where support for the single project class and the single project class with safety projects would start to be able to configure and forget ... something you cannot really do now ... and it would not screw up those of us on the other end of the spectrum ... > Maybe it shouldn't be half, maybe it's a formula based on average > debt, but there has to be room for the "most needed" work, or you > run the risk of not being able to get "x" when "x" has work. > > ... and unless I've missed something pretty major, it all works off > of actual CPU time, and benchmarks aren't a factor in the long-term. > > At this point, that's my OPINION. > > Until I have time to sit down and code something, to demonstrate > that I'm right, I'm not going to berate those who are working very > hard to keep BOINC functional -- and actually doing the work. And what would you have me do? I cannot code C or C++ that well... those rare code segments that I do send in are ignored ... when I spent up to 18 hours a day for a couple years documenting BOINC, well ... let us not go there ... Dr. Anderson pleaded with us to look at a proposal last year and to make suggestions ... so 8 or 9 of us did ... and I summarized and consolidated the input and made changes to his proposal, changes of which he accepted none ... The consequences to his, ahem, intransigence is with us today in the problems we are seeing with ATI integration ... which could have been a slam dunk done already ... His latest proposal on credit won't work but he is silent when asked if this is a serious effort and he is going to actually listen to input ... his silence, to me if no one else, is message enough ... And I was a systems engineer with a specialty in Db design ... this is what I used to do for a living ... can I get my advice listened to on any topic? ... history says no ... Back in BOINC Beta when I was heavily into PHP and web work I offered to completely rewrite the BOINC web code to make it CSS and W3C compliant, I was turned down, emphatically ... now there is a standing plea to have the web site code worked on ... I send in bug reports that are derided for being too long except when they are derided as being too short ... for having logs that are too long because that is what it takes to see the problem ... to being told I don't know what I am doing ... when I do ... that a benign error message is the cause of the problem when it isn't ... which I might observe one could take to indicate that the UCB development team really does not use BOINC that much and does not know that much about the projects and how they can interact with each other ... oh, yeah, that was another set of bug reports that we don't want to talk about from mid-year ... how Drug Discovery or IBERCIVIS tasks were causing problems with other project's tasks ... but I digress ... So, what would you have me do? If it seems I am on the sidelines shouting ... well, Dr. Anderson has deemed that that is the only place I can be ... he won't let me be anywhere else ... and no, I did not start out this disgusted, i'm not angry, just sad and disgusted ... and if I were the only one that is pretty consistently shut out it is very likely I would shut up and go away ... but I am one of the few that can't quite seem to convince myself that maybe a change might not be impossible... It only takes a few days on anyone of a number of project sites to see that there a lot more disenchanted users out there than just me ... some so much so they consider even reporting bugs that are killing them a waste of time. And the 800 pound question no one wants to touch is that with however many years do we not have more than 300,000 active participants out of 1.7 million that have signed up ... could it be that BOINC does not work the way people find acceptable? Could it be the lack of responsiveness on the part of UCB? Why did we lose so many from SETI Classic? I actually have answers to those questions ... so does UCB if they would read their own survey ... did you know about the survey? Most people don't ... As to Berate? The closet I have come to berating some one anytime in recent history is the somewhat parallel discussions with John who is changing his argument so suit what appears to be some strange agenda that I cannot figure out. As soon as he stops being a hypocrite on efficiency I will stop calling him out on it. He cannot stand on a platform of being against one proposal because he fears it will lower "efficiency" (it really won't) and being against others when it will improve efficiency and stability of the BOINC client... I have more respect for him and his contributions than you can imagine ... but he is sure making it hard to maintain that respect. So, for the last time what would you have me do? Where do you think Dr. Anderson will let me make a contribution to BOINC? So far he has been real consistent in preventing anything I do from becoming part of the system... Are your really suggesting that you want to take the roll of the one that answers "Who will rid me of this meddlesome priest?" What would you have me do? I really want to know ... _______________________________________________ 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.
