I don't recall if I saw it here on the BOINC messageboards or whether it was on my native Einstein project boards, but I'm pretty sure that I read that there exists a way to devote one "virtual core" to a single project.

There is a non-BOINC Windows utility that can say basically "SETI can run only on that core"... I also have no idea how it works, where to get it, etc., but I've seen it discussed.

---

@Michael, if "no new work" on all, changing pref on one then update it then update others will propagate in minutes... otherwise, I think it'll take 1 WU each; I would hope that CPDN does it on trickle! :-)

Probably hasn't touched the Climate WU yet, so at this early stage, after the preference alterations and update on one site, that WU could just be "culled", which I think would then force the update through CPDN with "no harm done", no wasted CPU cycles, and another CPDN WU acquired. Of course, I am out on a limb here, assuming that CPDN doesn't have a drastic penalty for a WU errored-out due to abortion.

---

"The arc of history is long, but it bends toward Justice"

SETIClassic burns through 18-24 workunits per day. Its about 5 days total.

LHC has 12 workunits plus the 2 its working on and is taking about 5 hours per unit, 2 at a time (2 get completed every 5 hours). There are no download dates in the work tab. It hasn't downloaded any new work units yet (according to the messages tab).

I'm getting my alienware laptop back on friday. It has an identical processor and it is 90% as fast (crunching seti) as the desktop. Its SETICommander has over 300 workunits in the cache because I export workunits to my 2 older computers that are not on the internet (6-8 and 3-4 work units per day). I also stashed 999 workunits onto a laptop i gave to my mom. I'm guessing they're going to be wasted as i'll see her again next summer. She's not very computer litterate. That machine does about 3 work units per day.

---

LHC has 12 workunits plus the 2 its working on and is taking about 5 hours per unit, 2 at a time

If you "no new work" all projects in BOINC, and let the LHC work finish, you should then be able to let Classic have the computer long enough to finish off that work, without missing any deadlines. Then come back to BOINC with a smaller "connect to every" setting, say 3 days; hit "update" on all projects so that the new pref gets propagated; and all should be good within a few days.

I'd still recommend going even smaller on the cache, even down to 0.01... but that's up to you. I just know that 7 is going to give you grief.

---

I've just read this entire thread right through and quite quickly. There are a couple of points I'd like to add to the discussion.

@GreatInca
I greatly respect the calm way you have continued to ask questions and listen to advice, even although, from your point of view, there have been a number of people severely "beating" on you. You have handled the whole thing very professionally. Normally, if BOINC and classic Seti are running simultaneously, BOINC gets nothing which is what many of the other posters were telling you. I've never used Seti Commander but from what you are saying, it is able to lower the priority of the classic seti executable so that it's even lower than that of the BOINC controlled science apps. I understand your predicament.

The other posters are all giving you sound advice. Because they are all eager to help, a lesser person than yourself would probably have felt quite attacked by all the strong "attention" you were being given. However you have stuck around and your calm replies lead me to believe that you are a very worthy addition to the "BOINC family". So welcome aboard!! :).

Matt Lebofsky posted not too long ago about what is currently going on with classic Seti. He stated that the classic work being returned right now is not going to make it into the science database as the work has already been done by BOINC. The clear message was that the only real reason for continuing to crunch classic was for personal stats only and not for science.

Now, I do regard the stats as being important so if you want to boost your classic score a bit more then please go right ahead. However, if you also think the science contribution is important, I'd seriously suggest that you put your effort into BOINC/Seti rather than classic. Science will not suffer if you were to choose to ditch your remaining classic WUs.

If you convert entirely to BOINC, we can show you how your machine, even with the 50% Seti resource share, can do the same or more Seti work than you were previously doing for classic. You say you can do around 20 classic WUs per day. Under BOINC I wouldn't be surprised to see that figure go up to around 40 by using the marvellously optimised science apps that some very smart users have developed. BOINC doesn't count the actual WUs. It awards credit based on the actual work done, albeit not very satisfactorily at the moment, but that's entirely another story for another time :).

If you choose to finish off the classic work that you have, Bill's plan is your best option. Bill is a very dedicated and very knowledgeable person and I've yet to see him give bad advice. LHC deadlines vary from about 5 to 8 days so let your machine finish off LHC. Set your "connect to network" interval to 0.01, set "No new work" for all BOINC projects and "suspend" all except LHC. As soon as LHC finishes (probably less than a day) shut down BOINC and fire up classic. As soon as classic clears, start BOINC again.

When you restart BOINC, because you have 0.01 for your connect interval, you will be in the best position to save the remainder of your BOINC work without BOINC going into panic mode. You will leave LHC and CPDN suspended but you will unsuspend EAH and Seti. Because they both have 2 week deadlines and because you should have the best part of a week before deadline, BOINC should be able to run down the stored work you have for those two. And it should just about be able to do it in a nice round-robin fashion. Particularly if you use the time from now until then to find out about the optimised Seti app which will plow through the Seti work very quickly. Once everything settles, it will be time to unsuspend all projects, allow new work and gradually start increasing your connect setting, probably to around 0.5 days which should give you a nice manageable supply of work and true round-robin scheduling.

Have a good think about all this and reply here to let us all know what you would like to do. We'd love to help you get going quickly. Once you have made that decision, we can walk you through, step by step if necessary, handling any specific questions you would like to throw at us.

---

Cheers,
Gary.

GreatInca,

I've just looked you up on LHC, EAH, and Seti. I see you've already finished all the LHC. Good job there. You've got 16 EAH and 41 Seti with one of the Seti results finished and validated. Crunch time was just over 2 hours so you will be able to do 20+ per day with an unoptimised science app and a lot more if you choose to run an optimised app.

In any case, you have less than 5 days of short deadline BOINC work so you should have no problem finishing the classic stuff first. I imagine you are probably doing that right now :).

Because things aren't as critical as I imagined they might be, you don't really need to go to 0.01 for your connect interval. Set it to about 0.1 to 0.5 days and BOINC will be fine. Most of your deadlines are 13 December but 20 of your Seti results are actually 16 December so you actually have even more time than I first thought.

---

Cheers,
Gary.

OK I hope BOINC's weightings are longterm. I wanted SETI to get exclusive use of 1 CPUCore and The rest to share the 2nd (50/25/25). If SETI runs out of stuff and the server is broken then Einstein gets to monopoize a CPU while LHC/Climate shares the 2nd 50/50. Definitly looks like it don't work work that way.

One way to force BOINC to run SETI only is to suspend temporarily all of the other projects. As there is nothing else it can run, it will then process SETI. I'm not sure whether you can do that on a per CPU basis though. If not, you can (say) on alternate days let everything run, or let SETI only run. I had to do that on a PC which had decided that it must run CPDN only for the next ten years :)

As others have said, you should get rid of Classic. It will die anyway later this month.

---

---[One way to force BOINC to run SETI only is to suspend temporarily all of the other projects]

My 2 cents.

Ignoring the classic SETI problem, I completely agree with GreatInca's assessment of the BOINC scheduler problem.

IMO, the BOINC scheduler should simply cycle through my projects and hand over my machine for the prescribed amount of time. (In my case, I have it set to 120 minutes). If the project finishes a WU during that time it should start up the next one, if it has one available. When the 120 minutes are up, BOINC hands my machine over to the next project.

Deadlines are project specific. BOINC should not care about deadlines, and it should not allow this to interfere with the other projects.

BOINC needs to do 3 things:

- time slice my machine equally, and fairly, and consistantly, between the many projects, based on MY preferences.

- maintain an adequate cache of WU's for each project so that the projects are not sitting idle on my machine.

- upload results to the project web sites, in a timely manner.

The deadlines for an individual project should only come into play when deciding how many WU's to download for THAT project. BOINC can keep stats on how long the WU's typically take to run for THAT project. A simple calculation will determine how many it can safely cache. If my machine is time-sliced in a constistant manner, this calculation is easy to do and is extremely accurate. To this end BOINC should have a new preference: "How many days worth of WU's do you want to cache ?"

I have 7 computers running BOINC (5.2.13), with 7 projects. It is amazing to me (astounding actually) how radically different the schedules are on each of my machines. They should all be identical !! IMO, the BOINC scheduler is too smart for it's own good.

To work around this problem I have gone to ONE project on each machine. The time slicing is now trivial !!

--- bt

IMO, the BOINC scheduler should simply cycle through my projects and hand over my machine for the prescribed amount of time. (In my case, I have it set to 120 minutes). If the project finishes a WU during that time it should start up the next one, if it has one available. When the 120 minutes are up, BOINC hands my machine over to the next project.

How do you balance _your_ desire for this behavior, with the desire of others to never switch projects until a result is complete? Many people are pushing just the opposite - do away with timeslicing altogether. Do one full result, then decide which project to go to next based on the debts. (There are definitely problems with what they're asking for, *coughCPDNcough* but there are also advantages...)

Deadlines are project specific. BOINC should not care about deadlines, and it should not allow this to interfere with the other projects.

You're going to have trouble getting _any_ of the projects to agree with this one. The purpose of using BOINC is to get work done for the project. NOT to make the participants happy at the expense of getting work done. Deadlines are, for _some_ projects, absolutely necessary for the science. For other projects, they're necessary to be able to get credits issued in a reasonable time. And for some, they're irrelevant.

The deadlines for an individual project should only come into play when deciding how many WU's to download for THAT project. BOINC can keep stats on how long the WU's typically take to run for THAT project. A simple calculation will determine how many it can safely cache. If my machine is time-sliced in a constistant manner, this calculation is easy to do and is extremely accurate. To this end BOINC should have a new preference: "How many days worth of WU's do you want to cache ?"

The deadline for a result is not known to BOINC Manager until the result has already been downloaded, as it's contained within the result. Projects can change deadlines every ten minutes if they so desire. The rest of your statement is already in place! Especially including the cache! If you didn't know that, then the entire rest of your argument becomes irrelevant, as you don't have enough knowledge of the existing system to be qualified to suggest changes. If you _did_ know it, but consider the current handling inadequate, please explain...

I have 7 computers running BOINC (5.2.13), with 7 projects. It is amazing to me (astounding actually) how radically different the schedules are on each of my machines. They should all be identical !! IMO, the BOINC scheduler is too smart for it's own good.

All identical machines? Same CPU, same motherboard, same memory, same "100% BOINC 24/7 no other use at all" rules? If so, and the schedules are different by very much, then you may have a reasonable complaint. But, just for one example, computer "x" is better at SETI while computer "y" is better at Einstein. So one computer may do a SETI result quicker on average, and thus need to do _more_ SETI results in a given week than another computer which does Einstein better, in order to give the same resource share. And certainly, if you suddenly play a CPU-intensive game on one machine while it was running SETI, then it's going to have to shift everything around later to recover that time...

The scheduler may well be "too smart", but if anything, MORE logic needs to be added to it. You're asking for it to be removed altogether, to go back to the days of BOINC 4.19, where people frequently missed deadlines. Sorry, not gonna happen.

---

Bill,

Thank you for a wonderfully-stated, informative post, not to mention an excellent point-by-point rebuttal.

"The purpose of using BOINC is to get work done for the project, NOT to make the participants happy at the expense of getting work done." - I'm saving this classic, if you don't mind.

Regards,

Michael

---

"The arc of history is long, but it bends toward Justice"

That was a great rebuttal Bill............


Yes Michael, it seems that many forget that they are here for the projects not the other way around......

---

BOINC Wiki

---[The scheduler may well be "too smart", but if anything, MORE logic needs to be added to it]

Yikes !! I was afraid SOMEONE was going to say that !! Eventually BOINC will be more complex than any of the real projects and they will have to set up a BOINC project to distrubute that work. [grin]

---[The purpose of using BOINC is to get work done for the project]

Nooo...the purpose of the PROJECT is to get work done for the project. The purpose of BOINC is to provide a common medium for users to SHARE their unused cpu resources among many projects.

---[Deadlines are, for some projects, absolutely necessary for the science]

These projects should NOT be running in a shared environment. They should come with a warning: "I need to take over your entire computer. I do not share well with others".

Of the 7 projects that I'm running, PrimeGrid has the shortest deadline and CP.NET has the longest. Are you trying to tell me that the "science" of factoring 200 digit numbers is more important than predicting the effects of climate change ?

I repeat ... the deadlines are purely arbitrary and BOINC should not even consider them. And if BOINC won't ignore the deadlines, I will, by droping the project(s).

---[if you did know it, but consider the current handling inadequate, please explain...]

I was simply reiterating the ENTIRE process, for completeness sake.

---[You're asking for it to be removed altogether]

Yes ... but ... I would like to have the OPTION of using SIMPLE time-slicing, or COMPLEX arbitray-deadline based scheduling.

---[go back to the days of BOINC 4.19]

Like I said ... I've already SOLVED this problem by assigning ONE project to ONE machine. I have NO scheduling problems what-so-over. This is as simple as it gets.

--- bt

---[it seems that many forget that they are here for the projects not the other way around]

Yikes. I'm running 7 projects full out, 24/7, on 7 machines. Don't tell me that I'm not here for the projects. I'm here for fair play for ALL projects, equally. Not to give preference to projects with arbitrary tight deadlines.

You should, perhaps, read my post, understand my concerns, and provide some constructive feedback (much like Bill Michael did), instead of simply patting him on the back for HIS "job well done".

--- bt

---[Thank you for a wonderfully-stated, informative post, not to mention an excellent point-by-point rebuttal...]

Michael Roycraft, you are contributing to this discussion how ?

--- bt

---[if you did know it, but consider the current handling inadequate, please explain...]

I was simply reiterating the ENTIRE process, for completeness sake.

I'm having trouble reconciling the above with:

To this end BOINC should have a new preference: "How many days worth of WU's do you want to cache ?"

Other than that, your viewpoints and mine on some things in BOINC don't match. *shrug* Okay.

---

Bill:

Don't get me wrong. I LIKE BOINC. I like the concept. I like being able to contribute. I ran the classic SETI for many years and switched over to BOINC a year and a half ago. I'm now contributing to 7 projects. I don't turn my computers off because of BOINC.

My ONLY complaint is the scheduling.

To this end BOINC should have a new preference: "How many days worth of WU's do you want to cache ?"

Yes there's a vague preference of "how often do you plan to contact BOINC", but I'm not sure it's as cut and dry as "keep x days of work cached for each project". Perhaps it is, but it's certainly masked by the bizarre scheduling algoritm.

Other than that, your viewpoints and mine on some things in BOINC don't match. *shrug* Okay.

This is to be expected, but I think we both agree that BOINC is useful and these discussions are useful.

Thx --- bt

My ONLY complaint is the scheduling.

The simpler scheduling was abandoned/fixed/broken (depending on your view :)) when people were missing too many deadlines.

So, the logic was changed to make meeting deadlines the priority. In the case of CPDN, though the deadlines are run times are long, the point is for some work they do need it back in time ...

---

---[Thank you for a wonderfully-stated, informative post, not to mention an excellent point-by-point rebuttal...]

Michael Roycraft, you are contributing to this discussion how ?

--- bt

Excuse me, Massa, Ah go back to da shed now and wait fo' ma whuppin' :-)

Seriously, the deadlines are in place for good reasons, and although the wording on cache size in the preferences could be made easier to understand, that is being addressed.

---

"The arc of history is long, but it bends toward Justice"

To this end BOINC should have a new preference: "How many days worth of WU's do you want to cache ?"

Yes there's a vague preference of "how often do you plan to contact BOINC", but I'm not sure it's as cut and dry as "keep x days of work cached for each project". Perhaps it is, but it's certainly masked by the bizarre scheduling algoritm.

That would be why it says:

Connect to network about every
(determines size of work cache; maximum 10 days)

(Which I have been pushing to drop the first sentence and only leave the second...)

I understand what you're saying, and what you're asking for; I'm just saying that we've "been there, done that, it didn't work". Have you read the Wiki on the scheduling algorithm? Or the page on it here on this site? When I first came over to BOINC from SETI Classic, it took a while to figure out what was going on. Now that I _understand_ the choices that have been made, I (mostly) agree with them. Before I understood them, it was real easy to think "gee that's stupid". I'm just trying to tell you that unless your understanding is solid, you can't reasonably say "this way would be better".

I currently see a problem in the work fetch part of the scheduler. I've been trying for several days to figure out how to describe it in enough detail, and have enough solid examples, that the developer will instantly agree with me - if I fall short, I'll lose credibility, and it'll be harder to "prove" that there's really a problem that needs to be fixed; my argument has to be as perfect as I can make it before I send it in.

As far as deadlines - it may seem stupid to have a short deadline for a mathematics project, and I _did_ say that some deadlines were arbitrary. But from the project's perspective, if some "part" of a project needs to be done before something _else_ can begin, then it's reasonable that they would set short deadlines for each "section" of the work. Otherwise, the "next part" could be held up waiting on a handful of people to return results from the "prior part". And _some_ projects have had to _shorten_ their deadlines, under pressure from participants - who got tired of waiting a month for credit to be awarded, just because one of the people in their quorum hadn't returned the result in a reasonable time. There are many reasons for deadlines. Just saying that you want to ignore them, ignores the expectations that other people have, whether that be the project staff, or other members of your WU quorums.

I don't know of anyone who thinks that any part of BOINC is "perfect", and I realize that you aren't here screaming "BIONIC SUX" - if you were, my response would have been a LOT different. You're seeing ways that you think it could be made better, and that's good. The developers _do_ listen. They'll listen MORE, if you have detailed information on what should be changed, and the side-effects of that change, showing that you've actually investigated all of it, instead of just "I would like it this way". The scheduler is probably the HARDEST thing to convince anyone needs a change, BECAUSE it's so complex - only a few people _really_ understand everything about it, and they aren't going to listen to anyone who doesn't have at least a "better than average" understanding of it. Convincing them that it's not even necessary, as I said, ain't gonna happen.

---

Another pat on the back Bill........

---

BOINC Wiki