Send WUs to 3 users instead of 2?

Send message
Joined: 11 Jul 11
Posts: 2298
ID: 105020
Credit: 8,396,846,380
RAC: 6,446,404

I have completed a few SOB tasks and time and again, I see my wingman not coming through with the goods (it times out or has an error) and the work unit is passed onto the next person - another long wait.

I think if a work unit is initially sent to 3 people (for a quorum of 2) and all 3 can get the credit (if its the same result) Seventeen Or Bust will have a noticeable increase in momentum.

Its worth a try?
-----------------------------------------------------------------------------------
My lucky number is 429*2^1301821+1

ID: 69479 |

Volunteer moderator
Project administrator

Send message
Joined: 21 Jan 10
Posts: 14011
ID: 53948
Credit: 433,157,676
RAC: 1,017,804

I have completed a few SOB tasks and time and again, I see my wingman not coming through with the goods (it times out or has an error) and the work unit is passed onto the next person - another long wait.

I think if a work unit is initially sent to 3 people (for a quorum of 2) and all 3 can get the credit (if its the same result) Seventeen Or Bust will have a noticeable increase in momentum.

Its worth a try?
-----------------------------------------------------------------------------------
My lucky number is 429*2^1301821+1

SoB has no mass, so it can never have any momentum. :)

This has been discussed and while I won't rule it out, the reason it hasn't been done is that it will probably have a negligible effect, or, worse, the opposite of what we want.

There's two problems. The first is a problem for volunteer crunchers: the long wait for a wingman is frustrating, especially when it's the last task you need for a badge. We've all been there. The second problem is a problem for the project: Progress through SoB is slow.

Increasing the SoB initial task distribution would help the first problem somewhat, but possibly not as much as we would hope. On average, there are 4.7 tasks per SoB workunit. Looking at it from that perspective, an initial distribution of 3 might not be enough, perhaps 4 or 5 would be better.

However, the real problem isn't just the number of tasks, it's the tasks that go out and never come back, which incurs a 40 day delay. Off the top of my head, I would guess that this is a very common occurrence, happening at least once, if not several times on each and every workunit. But it's not as common as one would think. Only about 0.35 tasks per SoB workunit time out. The bottom line is that increasing the initial distribution wouldn't help the long waits as much as one would think. (Being human, we notice and agonize over the long waits and tend to ignore the tasks that validate quickly.)

On the other side of the coin, doing this would significantly hurt the SoB project. Inevitably, many tasks will be wasted as three tasks are sent out and all three are returned correctly, which wastes a full third of the available computing power.

Generally speaking, the only time it makes sense to increase the initial distribution above 2 is when getting a quick response is vital to the science. There's only one project I know of that is like that (GPUGRID), and even they don't use an elevated initial distribution.

Put another way, this is a tradeoff between latency and throughput. The "momentum" you spoke about is the throughput. Increasing the initial distribution will lower latency (that's good), but will also lower the throughput (that's bad.) The latency is more noticeable (and annoying), but it's the throughput that determines how much work gets done.

(The other long project, GFN-WR, has an average of 17 tasks per workunit and 1.25 timeouts per workunit, with a deadline of 21 days.)

____________
My lucky number is 75898⁵²⁴²⁸⁸+1

ID: 69485 |

Send message
Joined: 11 Jul 11
Posts: 2298
ID: 105020
Credit: 8,396,846,380
RAC: 6,446,404

Thankyou for your excellent reply.

I have been looking over the 11 SOB results still available to me (not a huge sample to work with)

2 tasks are 'completed, conclusion inclusive'
1 task has 3 users validate llr_sob_71263621
8 tasks timed out once (2 of which timed out on an average of 4.5 days otherwise 40 days)

For completed tasks (7) of which 3 timed out for 40 days:
The days to do this are: 34, 51, 15, 19, 13, 41, 58

For non-completed tasks (4) of which 3 timed out for 40 days:
The days so far: 54, 54, 44, 76.

If a solution could be found to encourage people to not let a SOB work unit time out - this will get an increase in momentum ;)

ID: 69518 |

Send message
Joined: 11 Jul 11
Posts: 2298
ID: 105020
Credit: 8,396,846,380
RAC: 6,446,404

It is just a coincidence that the following add up to 11 :
2 tasks are 'completed, conclusion inconclusive'
1 task has 3 users validate llr_sob_71263621
8 tasks timed out once (2 of which timed out on an average of 4.5 days otherwise 40 days)

And its inconclusive (not inclusive) My typo bad.

Of the 11 work units I have these events :
6 * aborted
18 * error
8 * time out (6 of which took 40 days)
2 * completed, validation inconclusive
17 * completed, validated
4 * in progress

ID: 69525 |

Send message
Joined: 11 Jul 11
Posts: 2298
ID: 105020
Credit: 8,396,846,380
RAC: 6,446,404

I am really sorry everyone, what a tangled web I weave.
I hope everyone can have a laugh at my previous errors:

Of the 11 work units I have these events :
6 * aborted
18 * error
8 * time out (6 of which took 40 days)
2 * completed, validation inconclusive
7 * completed and validated
17 * results that need another result
4 * in progress

ID: 69526 |

Volunteer moderator
Project administrator

Send message
Joined: 21 Jan 10
Posts: 14011
ID: 53948
Credit: 433,157,676
RAC: 1,017,804

8 tasks timed out once (2 of which timed out on an average of 4.5 days otherwise 40 days)

Any timeouts that take less than the correct amount of time to time out happen because the system clock on the host computer is set incorrectly. The BOINC client will abort and return any unstarted task that is already past its deadline (as defined by the clock on the host computer), and the server then marks it as expired even though the server has the correct time and knows it's not really expired yet.

____________
My lucky number is 75898⁵²⁴²⁸⁸+1

ID: 69527 |

Send message
Joined: 11 Jul 11
Posts: 2298
ID: 105020
Credit: 8,396,846,380
RAC: 6,446,404

So its an anomaly? A true time out will only be 40 days?
We can voom SOB up by at least 20% by encouraging people to do this:
If you take the WU you do it, preferably in half the time. Focus on only 1 or 2 projects - preferably just 1 project.

ID: 69528 |

Send message
Joined: 11 Jul 11
Posts: 2298
ID: 105020
Credit: 8,396,846,380
RAC: 6,446,404

I just had the 'I've got an idea moment' It could be an exaggerated idea as a bottle of red was involved.

Why not give the bonus to people that have the same projects running at once?
It applies to SOB, PSP.

(I am thinking of solutions that are easy to implement)

Alternatively, the users need to get their chops up.
So the computer can be tested - and see if its able to run for 240 hours plus.
Just one test of SOB. (much easier to code?)

The first SOB WU for a new user should be a test to see how they do it. If they get it.

ID: 69529 |

Volunteer moderator
Project administrator

Send message
Joined: 21 Jan 10
Posts: 14011
ID: 53948
Credit: 433,157,676
RAC: 1,017,804

So its an anomaly? A true time out will only be 40 days?
We can voom SOB up by at least 20% by encouraging people to do this:
If you take the WU you do it, preferably in half the time. Focus on only 1 or 2 projects - preferably just 1 project.

We certainly try to encourage that behavior, but understand that many, many people never visit the website or read the notices or anything like that. Many don't speak English at all. Also, sometimes the task is abandoned due to computer problems (e.g., a disk crash or other failure) rather than a willful act by the user.

____________
My lucky number is 75898⁵²⁴²⁸⁸+1

ID: 69530 |

Send message
Joined: 11 Jul 11
Posts: 2298
ID: 105020
Credit: 8,396,846,380
RAC: 6,446,404

I want to finish this project. I think many people would like to understand how many work units are involved. It is finite? Or are we building castles in the air?

ID: 69532 |

Send message
Joined: 10 May 09
Posts: 804
ID: 39821
Credit: 1,935,039,845
RAC: 1,392,602

SOB Wiki Article will help you understand it more. SOB is finite, whereas Sierpinski and Proth are infinite.

____________
My lucky numbers are 121*2^4553899-1 and 3756801695685*2^666669±1
My movie https://vimeo.com/manage/videos/502242

ID: 69533 |

Gary Craig
Volunteer tester
Avatar

Send message
Joined: 30 Dec 09
Posts: 3213
ID: 52890
Credit: 1,005,618,748
RAC: 0

I'm currently running 7 tests: 1 vs. 10223, 2 vs. 21181, 2 vs. 24737, and 2 vs. 55459. A few should finish late tomorrow, the others, not until this weekend.

I concur with the reasoning to keep the initial quorum as 2. But I wonder if the deadline is a bit on the high side? It does look proportionate w/r/t the other long tasks when considering recent run-times, but still, 40 days is quite a while. Maybe reduce to 31? That's a month, and prime. :-) Just a thought. I'm curious as to whether there are many valid SoB WU's returned in the 31-40 day range.

Cheers,
--Gary

ID: 69535 |

Send message
Joined: 10 May 09
Posts: 804
ID: 39821
Credit: 1,935,039,845
RAC: 1,392,602

Deadlines have the extra time for many reasons. Most users do not dedicate crunch, meaning that they crunch less than 24 hours a day (some just 8 or so hour a day), and many crunch more than one project, so it gets swapped out for other work. If you bring the time down you are asking people to be dedicated. That's not what BOINC is. It is shared services. Many of us who do crunch things do a lot of dedicated crunching as yourself, but there are those casual crunchers that send back just as good work as anyone else, just not as fast.

There are dozens of projects, many with multiple sub-projects (like PrimeGrid). The combinations could be astronomical. People want choice. They do not want to be strapped into one project.

The moderators do listen to people. Your comments do not fall on deaf ears. As you see, one of the moderators has been very helpful in answering questions. PrimeGrid is one of the best projects for this type of service.

It might be time, with the systems getting faster (although the numbers crunched keep getting larger) that a cut back on deadline may be looked at again. Do not expect anything immediately, but it could happen.

____________
My lucky numbers are 121*2^4553899-1 and 3756801695685*2^666669±1
My movie https://vimeo.com/manage/videos/502242

ID: 69538 |

Send message
Joined: 11 Jul 11
Posts: 2298
ID: 105020
Credit: 8,396,846,380
RAC: 6,446,404

Maybe I didn't word that well.
Thankyou pooh bear to remind me of the math.

ID: 69539 |

Send message
Joined: 11 Jul 11
Posts: 2298
ID: 105020
Credit: 8,396,846,380
RAC: 6,446,404

ok now you are patronising me Pooh Bear
I do not appreciate this.
I started this thread to find ways of speeding up SOB
Maybe I am not brilliant enough for your magic?

ID: 69541 |

Volunteer moderator
Project administrator

Send message
Joined: 21 Jan 10
Posts: 14011
ID: 53948
Credit: 433,157,676
RAC: 1,017,804

I want to finish this project. I think many people would like to understand how many work units are involved. It is finite? Or are we building castles in the air?

The answer to your question is "we don't know".

If the conjecture is true, then the project has a finite length, although how long it will run is impossible to estimate. It could end tomorrow, or it could take longer than the age of the universe. Or anywhere in between.

On the other hand, if the conjecture is false, then the project is infinite because we're searching for something that does not exist and we will search forever without finding what we're looking for.

This answer applies to all of the conjecture projects: SoB, PSP, TRP, SR5, and ESP.
____________
My lucky number is 75898⁵²⁴²⁸⁸+1

ID: 69543 |

Send message
Joined: 10 May 09
Posts: 804
ID: 39821
Credit: 1,935,039,845
RAC: 1,392,602

How did I patronize you, if you do not mind telling me. There was another person who recommended changing deadlines just before my conversation about it. I just spelled out what goes through admins minds as they adjust project times, deadlines, etc. Being an old admin myself, I know what goes into thinking about a project and why PrimeGrid has set things as they have. Mostly common sense really. I knew you know most of it, but sometimes there is in need of a reminder. We all sometimes find a tunnel vision and we need to break through it and see the larger picture.

The answers I gave were not only for you, but what if someone new comes onto PG, starts reading this forum and wondered some of those answers?

I am sorry if you felt like I may have been taking down to you by what I wrote, but I've been in the computing arena since the mid 1970s, I have done many jobs in computers throughout the years. I've even taught at the Technical/Jr. College level. It's in my nature to pass on information and try to help others understand, and I try to do it at a level that allows even less learned understand. Most people might think it is too much, but knowing that your target audience is much bigger than just you and I having a conversation is where I put myself.
____________
My lucky numbers are 121*2^4553899-1 and 3756801695685*2^666669±1
My movie https://vimeo.com/manage/videos/502242

ID: 69544 |

Send message
Joined: 11 Jul 11
Posts: 2298
ID: 105020
Credit: 8,396,846,380
RAC: 6,446,404

Sorry dude,

I think I was out of line. I was just a tad paranoid.

Have a great day
Nick

ID: 69545 |

Send message
Joined: 10 May 09
Posts: 804
ID: 39821
Credit: 1,935,039,845
RAC: 1,392,602

Nick,

I want to say you made some good points. There is nothing wrong with examining the projects from new perspective. As I said, these admins at PG are very willing to listen. There have been many changes made here because people like you and I havimg asked questions. So no not hesitate to ask.

Glad to have had a nice conversation with you.

ID: 69548 |

Jay

Send message
Joined: 27 Feb 10
Posts: 132
ID: 56067
Credit: 64,433,752
RAC: 17,531

As I understand it, a task is initially sent to two machines. If one of those two machines doesn't return the task before the deadline then the task is sent to a third machine. This happens until two matching results are received at which time credit is given to the two people who returned matching results and the task is considered finished. This all makes sense to me.

However, after a task is complete, if one of the machines that couldn't meet the deadline finally does return the task (with matching results) then that machine also gets the same credit. In my mind this doesn't make as much sense. The machine that couldn't make the deadline (for whatever reason) is likely the cause of the task being sent to a third machine, which as Michael described, "wastes a full third of the available computing power."

I think it may eliminate some of the frustration described by Nick (and many others) to encourage users who have machines that won't meet the deadlines to look at other projects. This might be done by reducing the credit awarded to a machine that returns results after a timeout that causes a task to be sent to a third machine (wasting resources). Initially, I think it makes sense to reduce the credit awarded for a late result by one-third reflecting the amount of resources wasted. Maybe a sliding reduction amount based on how much past the deadline would be possible?

I hate to see less users working on SOB as it's the one project I really want to see finished, and the only one I've dedicated to working on. And I know that at this time my machine is on the slower side of average. I've also failed to restart crunching after manually stopping which has caused me to miss deadlines so I know my suggestion would impact my own credits received, but I do understand the frustration of being the third machine to receive a task, and seeing the task validated a day later because someone missed a deadline. It's extremely frustrating to feel that the work you're doing is the wasted resources.

ID: 69570 |

DaveB

Send message
Joined: 20 Jun 09
Posts: 351
ID: 42198
Credit: 11,898,570
RAC: 0

Initially, I think it makes sense to reduce the credit awarded for a late result by one-third reflecting the amount of resources wasted. Maybe a sliding reduction amount based on how much past the deadline would be possible?

I remember a UNI lecturer on late work telling first lecture to class "You are adults now. If at work you told boss your presentation to the board would be a couple of days late you would probably get the sack." His idea was late work gets 100% reduction.

As a mature student for now I can say that 5 to 10 % per day is normal here (at JCU http://www.jcu.edu.au/)

As one who has waited for months for a wingman many times I would think that 10% per day would mean better returns as those who are not up to it would go off and find more suitable tasks, but the task would need to be held from purging until all the untimed-out work either is returned or times out.

The idea seems to be one of the better sugestions of the many I have seen over the years.
____________
Member team AUSTRALIA
My lucky number is 9291*2^1085585+1

ID: 69599 |

Volunteer moderator
Project administrator

Send message
Joined: 21 Jan 10
Posts: 14011
ID: 53948
Credit: 433,157,676
RAC: 1,017,804

Initially, I think it makes sense to reduce the credit awarded for a late result by one-third reflecting the amount of resources wasted. Maybe a sliding reduction amount based on how much past the deadline would be possible?

I remember a UNI lecturer on late work telling first lecture to class "You are adults now. If at work you told boss your presentation to the board would be a couple of days late you would probably get the sack." His idea was late work gets 100% reduction.

As a mature student for now I can say that 5 to 10 % per day is normal here (at JCU http://www.jcu.edu.au/)

As one who has waited for months for a wingman many times I would think that 10% per day would mean better returns as those who are not up to it would go off and find more suitable tasks, but the task would need to be held from purging until all the untimed-out work either is returned or times out.

The idea seems to be one of the better sugestions of the many I have seen over the years.

It's not worth the effort. At this moment, there are 1,257,928 results in the database that have credit, and only 18,253 of those were returned beyond the deadline. That's about 1.5%. The vast majority of those are on the short tasks where you're not really worrying too much about missed deadlines. Of those 18 thousand, 6 thousand are PPS sieve, 3 thousand are SGS, 2 thousand are PPS, 1 thousand are TRP sieve, and 3 thousand are PPSE. Everything else is in the low hundreds (or less).

____________
My lucky number is 75898⁵²⁴²⁸⁸+1

ID: 69603 |

Jay

Send message
Joined: 27 Feb 10
Posts: 132
ID: 56067
Credit: 64,433,752
RAC: 17,531

It's not worth the effort. At this moment, there are 1,257,928 results in the database that have credit, and only 18,253 of those were returned beyond the deadline. That's about 1.5%.

I know SOB is one of the less popular projects here. 1.5% of tasks in the entire database being returned beyond the deadline may be correct, but as someone who only runs SOB it seems extremely low. Is there anyway to determine the percentage of SOB tasks that are returned beyond the deadline? Without looking at any of my personal tasks or making any informed estimates I'm guessing it's close to 15-20%.

For me personally, the frustration doesn't come from waiting for a wingman (or two or three) for a huge amount of time, but from being the third or fourth recipient of a task and then seeing one of the initial recipients return it a couple days late and realizing that my time and energy is being spent on repetitive and non-useful work.

Thanks,
Jay

ID: 69729 |

cncr04s
Send message
Joined: 17 Sep 05
Posts: 139
ID: 797
Credit: 3,990,128,691
RAC: 3

I've got a bigger hand then most into the timed-out workunits debate, Due to the nature of my compute resources, There are times where the workunits get lost because the "Instance" Running it disappears into nothingness.

If my workunits are the cause in the case of people here, I am sorry but the tradeoff of some times having to wait for credit is the gain in computational capacity to the project as a whole.
This statement kind of also applies to people trying to help the project with their computer, credit can wait as its not important to the goal of the project(s).

____________

ID: 69733 |

Volunteer moderator
Project administrator

Send message
Joined: 21 Jan 10
Posts: 14011
ID: 53948
Credit: 433,157,676
RAC: 1,017,804

For me personally, the frustration doesn't come from waiting for a wingman (or two or three) for a huge amount of time, but from being the third or fourth recipient of a task and then seeing one of the initial recipients return it a couple days late and realizing that my time and energy is being spent on repetitive and non-useful work.

Thanks,
Jay

For SoB, just under 10% of results with credit awarded were returned after the (40-day) deadline. I can't easily tell how many of those cause a situation such as you describe where work is wasted. While the timeout causes a third task to be sent, that doesn't always translate into wasted work, if, for example, the third task or the original wingman return errors.

But I agree, nobody likes wasting their resources.
____________
My lucky number is 75898⁵²⁴²⁸⁸+1

ID: 69735 |

whynot
Send message
Joined: 9 Mar 11
Posts: 127
ID: 89957
Credit: 46,462,161
RAC: 5,153

For me personally, the frustration doesn't come from waiting for a wingman (or two or three) for a huge amount of time, but from being the third or fourth recipient of a task and then seeing one of the initial recipients return it a couple days late and realizing that my time and energy is being spent on repetitive and non-useful work.

Me seconds this. For my SOBs it's 4.8%. In this one case late-wingman reported ~16h before me and ~4h before me uploading (IIRC). Honestly, it's not that much comparing with two other projects I waste time occasionaly. Weird, it's still pain.

____________
I'm counting for science,
Points just make me sick.

ID: 69941 |

Send message
Joined: 11 Jul 11
Posts: 2298
ID: 105020
Credit: 8,396,846,380
RAC: 6,446,404

My frustration is that I have a wingman's WU time out again! First sent 18th July and now due 3rd December. Yes I'm waiting patiently to go gold, but I am also waiting to see if my computer is able to do SOB reliably. The WU I'm waiting for has validation inconclusive. I need to know before I take on any more SOB tasks.

ID: 70327 |

Send message
Joined: 11 Jul 11
Posts: 2298
ID: 105020
Credit: 8,396,846,380
RAC: 6,446,404

I have 1 computer.
The time out for my SOB is a not confirmed yet scenario. I like to know before crunching for SOB again.

ID: 70474 |

DaveB

Send message
Joined: 20 Jun 09
Posts: 351
ID: 42198
Credit: 11,898,570
RAC: 0

My frustration is that I have a wingman's WU time out again! First sent 18th July and now due 3rd December. Yes I'm waiting patiently to go gold, but I am also waiting to see if my computer is able to do SOB reliably. The WU I'm waiting for has validation inconclusive. I need to know before I take on any more SOB tasks.

I would suggest crunching another one while you wait. If you have finished then it IS possible for you to get it right and the more you have pending the greater the chance of getting a good wingman for at least one of them, or you may get someone else out of the wait by being the wingman next time. I always do a couple extra for the badge just to minimise the wait.
____________
Member team AUSTRALIA
My lucky number is 9291*2^1085585+1

ID: 70475 |

GDB
Send message
Joined: 15 Nov 11
Posts: 298
ID: 119185
Credit: 4,061,485,731
RAC: 1,942,441

In order to reduce frustration to regular SOB users, how about considering reducing SOB task timeout to 30 days, from 40 days. This has a good chance of reducing SOB task turnaround, without too many extra task starts.

I don't have the SOB completion time info, but a good rule of thumb might be to set SOB timeout to 95th percentile of completed tasks. By doing this, most SOB users should get a faster validation of their task, without too many extra tasks being started.

ID: 70774 |

Volunteer moderator
Project administrator

Send message
Joined: 21 Jan 10
Posts: 14011
ID: 53948
Credit: 433,157,676
RAC: 1,017,804

In order to reduce frustration to regular SOB users, how about considering reducing SOB task timeout to 30 days, from 40 days. This has a good chance of reducing SOB task turnaround, without too many extra task starts.

I don't have the SOB completion time info, but a good rule of thumb might be to set SOB timeout to 95th percentile of completed tasks. By doing this, most SOB users should get a faster validation of their task, without too many extra tasks being started.

No matter how long or short the deadline is, there's always somebody whose computer, running 24/7, is going to take longer than the deadline. With that in mind, anytime you lower the deadline -- and remember, the tasks are getting longer over time -- there are going to be some people who could meet the previous deadline but can not meet the shorter deadline no matter what they do.

So you're trading off, on the one hand, the aggravation of impatient users with fast computers vs. the aggravation of users with slow computers who are no longer able to crunch a project they used to be able to crunch. Essentially, we can either tell one group that they're no longer welcome here, or we could tell the other group that they need to be more patient.

Unless we really have no alternative, we try to be as inclusive as possible. Within reason. Don't expect to see SoB running on Android anytime soon!

____________
My lucky number is 75898⁵²⁴²⁸⁸+1

ID: 70775 |

whynot
Send message
Joined: 9 Mar 11
Posts: 127
ID: 89957
Credit: 46,462,161
RAC: 5,153

Don't expect to see SoB running on Android anytime soon!

Me feels terribly sorry for the green beast.

Anyway, my frustration ratio has been just upped to 6.45% (95.2Ksec wasted, it will decrease soon). And this one makes me icky. Bets?

____________
I'm counting for science,
Points just make me sick.

ID: 70945 |

DaveB

Send message
Joined: 20 Jun 09
Posts: 351
ID: 42198
Credit: 11,898,570
RAC: 0

It is not so bad, your 'this one' should be a nice xmas present, perhaps santa will come early, see my clip :-

501179445 192746 15 Nov 2013 | 14:33:29 UTC 25 Dec 2013 | 14:33:29 UTC In progress --- --- --- Seventeen or Bust v6.17

____________
Member team AUSTRALIA
My lucky number is 9291*2^1085585+1

ID: 70960 |

Tyler

Project administrator
Volunteer tester
Send message
Joined: 4 Dec 12
Posts: 1078
ID: 183129
Credit: 1,378,812,350
RAC: 30,684

Don't expect to see SoB running on Android anytime soon!

Me feels terribly sorry for the green beast.

Anyway, my frustration ratio has been just upped to 6.45% (95.2Ksec wasted, it will decrease soon). And this one makes me icky. Bets?

on this WU (the one that makes you icky) I'm betting on the second person who returned it, hostid=403859. Because hostid=297445 has many inconclusive tasks and already has an invalid SoB
____________

275*2^3585539+1 is prime!!! (1079358 digits)

Proud member of Aggie the Pew

ID: 70966 |

Volunteer moderator
Project administrator

Send message
Joined: 21 Jan 10
Posts: 14011
ID: 53948
Credit: 433,157,676
RAC: 1,017,804

Don't expect to see SoB running on Android anytime soon!

Me feels terribly sorry for the green beast.

Anyway, my frustration ratio has been just upped to 6.45% (95.2Ksec wasted, it will decrease soon). And this one makes me icky. Bets?

on this WU (the one that makes you icky) I'm betting on the second person who returned it, hostid=403859. Because hostid=297445 has many inconclusive tasks and already has an invalid SoB

You're likely correct. The result file (which I can see, but you can't) shows that 297445 had a processing error and restarted from a checkpoint, whereas the other host completed cleanly. Given the history of this host, it's likely that its result is bad.
____________
My lucky number is 75898⁵²⁴²⁸⁸+1

ID: 70967 |

TroubledBunny

Volunteer tester
Avatar

Send message
Joined: 4 Jun 07
Posts: 607
ID: 8931
Credit: 470,669,737
RAC: 0

In order to reduce frustration to regular SOB users, how about considering reducing SOB task timeout to 30 days, from 40 days. This has a good chance of reducing SOB task turnaround, without too many extra task starts.

I don't have the SOB completion time info, but a good rule of thumb might be to set SOB timeout to 95th percentile of completed tasks. By doing this, most SOB users should get a faster validation of their task, without too many extra tasks being started.

No matter how long or short the deadline is, there's always somebody whose computer, running 24/7, is going to take longer than the deadline. With that in mind, anytime you lower the deadline -- and remember, the tasks are getting longer over time -- there are going to be some people who could meet the previous deadline but can not meet the shorter deadline no matter what they do.

So you're trading off, on the one hand, the aggravation of impatient users with fast computers vs. the aggravation of users with slow computers who are no longer able to crunch a project they used to be able to crunch. Essentially, we can either tell one group that they're no longer welcome here, or we could tell the other group that they need to be more patient.

Unless we really have no alternative, we try to be as inclusive as possible. Within reason. Don't expect to see SoB running on Android anytime soon!

This WU:360546843 is a point in case. Mine is the 5th host to be sent this WU and I am 60% through it now. I will leave it to comlpete - estimated for 2 days time - but it was a waste sending it to me. A longer timeout would have been better in this particular case.
Pete

ID: 71020 |

whynot
Send message
Joined: 9 Mar 11
Posts: 127
ID: 89957
Credit: 46,462,161
RAC: 5,153

Check this out -- double time-out! And my condolences to Sam, something terrible has happened at 411647

____________
I'm counting for science,
Points just make me sick.

ID: 71536 |

whynot
Send message
Joined: 9 Mar 11
Posts: 127
ID: 89957
Credit: 46,462,161
RAC: 5,153

Frustration ratio -- 7.50%, time wasted -- ~11day, time secured -- ~46hour. That wasn't fair.

However, that made me thinking. I guess that would be possible to add an e-mail notification (on opt-in basis, for interested parties only) when late WU is reported. That might imply some manual work on admin's side, so future of such notification is moot. Upstream guys, what do you think?

Alternatively, except implementation pending, no manual work -- a possibility to subscribe for notifications about status changes in third party's WUs. Loads of abuse ahead makes it moot too.

Alternatively, does anyone know what debugging output should be enabled in client that would make server's cancel requests visible? (However, in this particular case, there wouldn't be any cancel request anyway.)

____________
I'm counting for science,
Points just make me sick.

ID: 71659 |

TroubledBunny

Volunteer tester
Avatar

Send message
Joined: 4 Jun 07
Posts: 607
ID: 8931
Credit: 470,669,737
RAC: 0

Hi Whynot,
Why did you abort 366222209? I could guess that you have no interest in credit and you're assuming that host 195491 will return a correct result?
Cheers,
Peter

ID: 71688 |

whynot
Send message
Joined: 9 Mar 11
Posts: 127
ID: 89957
Credit: 46,462,161
RAC: 5,153

I didn't assume. I'd weighted chances for ~10hour and got to conclusion that I'm not interested in races who-would-waste-more-time. Thus I forfeit. Besides, it's clear now that I've lost that race.

____________
I'm counting for science,
Points just make me sick.

ID: 71856 |

whynot
Send message
Joined: 9 Mar 11
Posts: 127
ID: 89957
Credit: 46,462,161
RAC: 5,153

UID:59940 is doing it again. However, may be, just may be, that's not his fault. Suspending for a month.

____________
I'm counting for science,
Points just make me sick.

ID: 74241 |

whynot
Send message
Joined: 9 Mar 11
Posts: 127
ID: 89957
Credit: 46,462,161
RAC: 5,153

Frustration level: dropped to 6.15%, 100Ksec wasted, 860Ksec secured. And one more in queue.

____________
I'm counting for science,
Points just make me sick.

ID: 75057 |

Volunteer moderator
Project administrator

Send message
Joined: 21 Jan 10
Posts: 14011
ID: 53948
Credit: 433,157,676
RAC: 1,017,804

Whynot,

The main reason why computers miss the deadline is not because the computers are slow, or because the owners don't care about the deadlines, or because the owners are doing anything wrong.

The main reason results get returned late is because BOINC is inherently bad at estimating how long it will take tasks to run, and this causes the BOINC client to sometimes underestimate how long a task will take. It therefore starts tasks too late to finish in time.

The problem is mainly with BOINC, not the computers, and not the users.
____________
My lucky number is 75898⁵²⁴²⁸⁸+1

ID: 75058 |

composite
Volunteer tester
Send message
Joined: 16 Feb 10
Posts: 1149
ID: 55391
Credit: 1,095,582,095
RAC: 734,071

Has anyone considered the idea of breaking a long WU into 2, split half-way? The first returns a checkpoint at completion, and the second starts from validated checkpoints. Half the BOINC credit to each, shared attribution on found primes. This could apply to all long-running subprojects.

ID: 75060 |

Volunteer moderator
Project administrator

Send message
Joined: 21 Jan 10
Posts: 14011
ID: 53948
Credit: 433,157,676
RAC: 1,017,804

Has anyone considered the idea of breaking a long WU into 2, split half-way? The first returns a checkpoint at completion, and the second starts from validated checkpoints. Half the BOINC credit to each, shared attribution on found primes. This could apply to all long-running subprojects.

I've considered it. There's a lot of logistical problems that make it difficult and undesirable. It's not impossible, but I doubt we'd ever do it.

A lot of the problems stem from the fact that the checkpoint files are huge, so that involves a lot of bandwidth and a lot of disk storage.

____________
My lucky number is 75898⁵²⁴²⁸⁸+1

ID: 75061 |

whynot
Send message
Joined: 9 Mar 11
Posts: 127
ID: 89957
Credit: 46,462,161
RAC: 5,153

The problem is mainly with BOINC, not the computers, and not the users.

(that's me putting things clear here) That's me live-blogging.

(that's me thinking here) I believe, that would ever frustration-ratio reach 10% then I would consider wishing theirs kernel catching cancer. But, that's important: _considering_ isn't _wishing_. And frustration-ratio never was that high.

Besides. This time I've discovered that *aborted* WUs aren't added to totals (what a surprise). I've adjusted frustration-ratio accordingly and now I feel ~0.5% better.

p.s. I'm old enough to get tired. You don't have to baby-sit me.

____________
I'm counting for science,
Points just make me sick.

ID: 75179 |

whynot
Send message
Joined: 9 Mar 11
Posts: 127
ID: 89957
Credit: 46,462,161
RAC: 5,153

Three month has gone and here it is, another blog entry. Frustration level ~5.75%, 19.4h wasted, 33.3day secured. Those huges provide added sense with that hunting. I like them :)

____________
I'm counting for science,
Points just make me sick.

ID: 77879 |

gazzyk1ns
Send message
Joined: 2 Aug 11
Posts: 1329
ID: 107047
Credit: 367,379,246
RAC: 0

If you want to maximise your chances of not ending up as a "third checker" and, as it seems to be in your case, having to abort at that stage - then check your WU as soon as it downloads. If it's already got one result returned and waiting for validation, and someone else with an equally fast PC has already got a head start on you for the double-check result, then you could abort it straight away, thus wasting almost no time. Hopefully after a couple of tries you'll get a WU with no returned results at all, and so as long as you don't max out your CPU (core-wise), you've got a good chance of not ending up as a "third checker".

ID: 78009 |

whynot
Send message
Joined: 9 Mar 11
Posts: 127
ID: 89957
Credit: 46,462,161
RAC: 5,153

Actually, I apply other strategy -- as soon as I found a problematic WU I suspend it for 10 days. However, I suppose one day I'll have to increase that tolerance gap because SOBs got bigger. I was overrun by ten days late WU (I've blogged about this already). Proper statistics would provide material to compute, let's name it, avoided frustration period. Right now I'm still unskilled to do it myself and I don't have the statistics anyway.

As of "checking", unfortunately it's manual and it's going to stay manual until I have time to do it right way. And The Right Way(TM) is going to look really cool.

____________
I'm counting for science,
Points just make me sick.

ID: 78048 |

Volunteer moderator
Project administrator

Send message
Joined: 21 Jan 10
Posts: 14011
ID: 53948
Credit: 433,157,676
RAC: 1,017,804

Proper statistics would provide material to compute, let's name it, avoided frustration period. Right now I'm still unskilled to do it myself and I don't have the statistics anyway.

You would be VERY surprised at the answer.

Most people think that "late" tasks usually miss the deadline by a small amount. That perception exists because you usually only are looking at the workunit when YOU finish the task, and the late task is usually only several days late at that point.

That's not the reality of the situation, however. The surprising truth is that the mean and median "late" times are absolutely huge. There's many tasks that come in quite a few weeks past the deadline.

Because of this, your strategy of "waiting a period of time" to make it less likely for a late task to come in probably won't work unless you're planning on waiting a really, really long time.
____________
My lucky number is 75898⁵²⁴²⁸⁸+1

ID: 78049 |

whynot
Send message
Joined: 9 Mar 11
Posts: 127
ID: 89957
Credit: 46,462,161
RAC: 5,153

That's not the reality of the situation, however. The surprising truth is that the mean and median "late" times are absolutely huge. There's many tasks that come in quite a few weeks past the deadline.

That's a baseless suggestion. "Mean and median" doesn't mean anything; "aboslutely huge" is quality, not quantity; and "many tasks" doesn't say anything about population.

Because of this, your strategy of "waiting a period of time" to make it less likely for a late task to come in probably won't work unless you're planning on waiting a really, really long time.

In contrary, it totaly does -- watch probabilities at work. Couple of times (at least two, IIRC) I spotted a late WU to come when it was too late to do anything -- I already reported it (and I wasn't 3rd wingman anyway). The last time I've considered shall I inform 3rd wingman that he's wasting time. But I've got to conclusion he doesn't care.

Now, I've considered a situation when I would be hit with more than 10 days late WU. Shall I increase timeout? May be. May be not.

It's possible that 10 days is an overkill. I can't say any better -- screening (from here) is technically unfeasable (at least now) and probably anyway harmful, my sample is of questionable value (and I don't collect that data anyway).

Please note, I'm not suggesting *you* standing up to task. Because, statstically speaking, nobody cares. And I'm totally fine with that. It just, you know, wasting time hurts.

p.s. One thing to note -- in course of IWOMTAITII project I've read book on probabilities and I'm kind of high on statistics right now.

____________
I'm counting for science,
Points just make me sick.

ID: 78187 |

composite
Volunteer tester
Send message
Joined: 16 Feb 10
Posts: 1149
ID: 55391
Credit: 1,095,582,095
RAC: 734,071

Let's face it. The BOINC client is bad at significant things: estimating deadlines, progress reporting, panic scheduling, backing off or switching error-prone workloads, awareness of CPU groups and affinity, comprehensive configuration with friendly online preferences.

So fork the BOINC client and customize it for PG. Borrow heavily from PRPNet client. Then let users run it side-by-side with regular BOINC if they want to contribute to other projects. Allow problematic applications to work only with the custom client.

ID: 78188 |

whynot
Send message
Joined: 9 Mar 11
Posts: 127
ID: 89957
Credit: 46,462,161
RAC: 5,153

Let's face it. The BOINC client is bad at significant things: estimating deadlines, progress reporting, panic scheduling, backing off or switching error-prone workloads, awareness of CPU groups and affinity, comprehensive configuration with friendly online preferences.

Remembering that upstream has gone for Credit Diktat(TM) instead, I suggest that they are completely unaware of this comprehencive list of deficiencies.

So fork the BOINC client and customize it for PG. Borrow heavily from PRPNet client. Then let users run it side-by-side with regular BOINC if they want to contribute to other projects.

Please don't consider this as "make your move" call, it's not. However I appreciate it and I think that more talks about fork the more chances the fork actually happens. Besides, one day I'll have to make a choice -- fork or quit. Unless I'll die before.

Allow problematic applications to work only with the custom client.

I totally disagree on that particular point. "Problematic applications" are those of SZTAKI and Primaboinca (I have to deal with them). If you don't know what they do you don't know what problematic application is.

Shortly, it's not a problem. It just hurts.

____________
I'm counting for science,
Points just make me sick.

ID: 78322 |

corwin
Send message
Joined: 19 Jun 07
Posts: 5
ID: 9232
Credit: 108,635,054
RAC: 0

Not exactly a big poster here, but I thought I would throw in my two cents, considering at the moment I have more pending credits than current credits for Seventeen or Bust. I don't know that these suggestions are workable, or would solve every problem if they were, they're just posted for discussion purposes.

1. Have a trickle up message from each computer running this project, or maybe extend to any long running project (over 2 weeks for reporting after delivery to computer). If checked maybe once every few days or once per week, then we could be sure that the task was making good progress, rather than waiting for it to time out, and then send out again.

2. Set up tiers for users. New users could start with simpler and shorter projects, that would take under 6 hours. Once they successfully complete a set number (maybe 20 such tasks with 90% acceptable rates), then move them into running longer projects, say up to 48 hours average compute time, and only after finishing 20 of these projects, move them into running the longest tasks, like SoB.

ID: 78326 |

gazzyk1ns
Send message
Joined: 2 Aug 11
Posts: 1329
ID: 107047
Credit: 367,379,246
RAC: 0

How about some sort of minimum hardware requirement for SoB? I know that doesn't sound like it's in the spirit of PG, i.e. "Anyone can run anything", but it wouldn't be a first - for example, GFN-WR WUs are still disallowed on CPUs when some people (possibly me) could genuinely churn out a couple at a time, without the unreliability of cutting-edge GPUs overclocked for gaming. I think that's very similar to stopping someone with a Pentium 4 running an SoB; the only obvious problem I can see is implementing it. Maybe a more definite/specific "warning" message in the BOINC preferences panel would be a good compromise?

On this Haswell @ 4.3, with 2133MHz RAM, HT off, and only 3 cores running, a new SoB WU takes about 95 hours. It's clearly not suitable for some users who select SoB, considering that they might only run PG in the daytime only, or at night. I think to meet the deadline, it takes at least a fast C2Q running almost 24/7, doesn't it?

ID: 78329 |

Honza

Volunteer moderator
Volunteer tester
Project scientist
Send message
Joined: 15 Aug 05
Posts: 1957
ID: 352
Credit: 6,137,993,476
RAC: 2,260,339

...in the spirit of PG, i.e. "Anyone can run anything", but it wouldn't be a first - for example, GFN-WR WUs are still disallowed on CPUs when some people (possibly me) could genuinely churn out a couple at a time, without the unreliability of cutting-edge GPUs overclocked for gaming.

I would encourage those participants hungry for VERY large primes to prepare for PRPNet GFN524288.
We are now ~826k, we were ~812k just 3 weeks ago and we can hit GPU b limit in just 3 weeks (~840k).
A prime found there would be a nice one and powerfull CPU will be the way to go soon.

On a side note, I can imagine we might need such a step (disallowing some older CPUs from a subproject) in a future. I.e. disabling non-SSE3 or 32-bit OSes. Even BOINC might be ready for this now. But I don't see PG need to do that now (or least for SoB), may be couple years later, may be sooner.

Personally, I wouldn't disallow someone from running what he wants.
Perhaps opt-in option - this would mean excluding SoB from "Send work from any subproject if selected projects have no work" and user would need to actively select SoB in order to run it (hence be noted those are very long tasks).
Basicaly a soft rule, not hard-coded.
____________
My stats

ID: 78332 |

gazzyk1ns
Send message
Joined: 2 Aug 11
Posts: 1329
ID: 107047
Credit: 367,379,246
RAC: 0

Yes - and possibly a different warning on SoB on the configuration page? i.e. not just "Very long tasks...", but possibly "A Pentium 4 will NOT finish this", or similar? I'm not suggesting that anybody be prevented from running anything which they can complete, it's just that someone could very easily and innocently come along and think they could run SoB, but then discover after a few days that either it's going to run for a lot longer than they thought, or that their hardware isn't up to it.

Having tickboxes for almost everything (aside from GFN-WR, annoyingly for me), might be shooting yourselves in the foot.

Just some thoughts.

ID: 78333 |

Dave

Send message
Joined: 13 Feb 12
Posts: 3207
ID: 130544
Credit: 2,285,331,764
RAC: 787,864

1. Have a trickle up message from each computer running this project, or maybe extend to any long running project (over 2 weeks for reporting after delivery to computer). If checked maybe once every few days or once per week, then we could be sure that the task was making good progress, rather than waiting for it to time out, and then send out again.

Not sure this would work - the unit could be in the cache & the trickle-up could still say not-yet-started & cancel the unit just before it starts on its own accord & indeed could finish.

2. Set up tiers for users. New users could start with simpler and shorter projects, that would take under 6 hours. Once they successfully complete a set number (maybe 20 such tasks with 90% acceptable rates), then move them into running longer projects, say up to 48 hours average compute time, and only after finishing 20 of these projects, move them into running the longest tasks, like SoB.

Sounds like an unlock system which gives an added level of achievement. But I agree it's not as 'open' to new users.

ID: 78335 |

Volunteer moderator
Project administrator

Send message
Joined: 21 Jan 10
Posts: 14011
ID: 53948
Credit: 433,157,676
RAC: 1,017,804

How about some sort of minimum hardware requirement for SoB? I know that doesn't sound like it's in the spirit of PG, i.e. "Anyone can run anything", but it wouldn't be a first - for example, GFN-WR WUs are still disallowed on CPUs when some people (possibly me) could genuinely churn out a couple at a time, without the unreliability of cutting-edge GPUs overclocked for gaming. I think that's very similar to stopping someone with a Pentium 4 running an SoB; the only obvious problem I can see is implementing it. Maybe a more definite/specific "warning" message in the BOINC preferences panel would be a good compromise?

On this Haswell @ 4.3, with 2133MHz RAM, HT off, and only 3 cores running, a new SoB WU takes about 95 hours. It's clearly not suitable for some users who select SoB, considering that they might only run PG in the daytime only, or at night. I think to meet the deadline, it takes at least a fast C2Q running almost 24/7, doesn't it?

I hate it when people read my mind. (Just kidding)

I was actually talking with Jim earlier today about the GFN-WR on Haswell idea. More on that later.

As for SoB, we don't really say, "anybody can run any project." Clearly that's not true. What we do like to say is, "Anybody can run something on PrimeGrid." Clearly, very, very slow computers can't run the big tasks in a reasonable amount of time. We make no attempt, for example, to allow non-GPU netbooks with Atom CPUs to run GFN-WR tasks. That's pretty close to the worst example of something that's technically possible but would take an extraordinarily long time to run.

Along those lines, if you're running a really old, really slow computer, we want to have something for you to run. PPSE and PPS, or the TRP Sieve and ESP Sieve tasks should be possible. Even some of the slightly larger tasks will work. I run TRP-LLR tasks on a decade-old 32 bit Celeron and have time to spare without worrying about deadlines, for example. But the big (or huge) tasks aren't intended for such slow computers. So, via the deadlines, we're already limiting some of the projects to "more worthy" computers. As you said, the bar for that isn't very high and a desktop Core 2 CPU (or AMD equivalent) should have no trouble with any of the deadlines if it's running 24/7. My main computer, until February of this year was a rather mediocre Core 2 Quad Q6600 (not the fastest Core 2 CPU by a long shot), and it never had trouble with any deadlines.

Consider this the "later" I referred to earlier...

As for GFN-WR on Haswell. If I can force BOINC to behave, maybe it will happen. If we just turn it on (something that would only take a few minutes to do), we'll have a disaster from missed deadlines and late tasks, because most CPUs can't even come close to the current deadline. We would really have to limit it to only Haswell CPUs, or, with a somewhat longer deadline, Sandy Bridge CPUs. Even then, if you have HT on you'll never make the deadline, and I can't detect that through BOINC.

While there are tools in BOINC to allow let you selectively send tasks only to appropriate computers, past experience (i.e., GeneferAVX) has shown that BOINC's implementation of this ability is less than perfect. As a result, I'm not even sure if it can be done.
____________
My lucky number is 75898⁵²⁴²⁸⁸+1

ID: 78337 |

Willy
Send message
Joined: 24 Sep 13
Posts: 19
ID: 257296
Credit: 48,339,334
RAC: 0

I have a 3630QM which takes around 200 hours (I have 8 running at once) Is that considered too long? I do have HT on though. I use my laptop while these are running and having HT on makes my computer perfectly usable while under full load. (while doing VM work and such)

ID: 78339 |

gazzyk1ns
Send message
Joined: 2 Aug 11
Posts: 1329
ID: 107047
Credit: 367,379,246
RAC: 0

Consider me in for any Haswell testing, at any stage, Michael.

ID: 78341 |

Volunteer moderator
Project administrator

Send message
Joined: 21 Jan 10
Posts: 14011
ID: 53948
Credit: 433,157,676
RAC: 1,017,804

I have a 3630QM which takes around 200 hours (I have 8 running at once) Is that considered too long? I do have HT on though. I use my laptop while these are running and having HT on makes my computer perfectly usable while under full load. (while doing VM work and such)

200 hours for SoB with HT on? SoB's deadline is 60 days (1440 hours), so that's certainly fast enough! In fact, I'm rather astonished you can run 8 SoB tasks in 200 hours on a laptop. The clock speed's a lot slower than a desktop, that's an Ivy Bridge CPU (vs.s the latest Haswell), and yet that's a very competitive speed. Frankly, I don't know how you can run them that fast unless the CPU is significantly overclocked, which is not something laptops are known for. Looking at other 3630QMs, assuming that the faster times are actually with HT off (4 tasks rather than 8), your computer is significantly faster.

Far from being "too long", it's amazing.
____________
My lucky number is 75898⁵²⁴²⁸⁸+1

ID: 78342 |

Willy
Send message
Joined: 24 Sep 13
Posts: 19
ID: 257296
Credit: 48,339,334
RAC: 0

I run a lot of other tasks and pause the SoB WU's sometimes which is why I have only completed 2 so far. Looking at my computers under my account, I swapped it for a 3840QM (didn't seem to update yet) and that is what I am running the WU's under ( I said 3630QM and that was a lie I swapped it out recently). The TDP is set to 72watts so I turbo boost to 3.8 while all 4 cores are under load. So yes, large overclock.

I get around 95% of the performance of a 3770k (stock)

I will have to see why my computer is not listing the 3840QM, it swaps between my old CPU and my new one sometimes (on my account page I mean). Strange.

Anyways, I am proud of my cooling mods and overclock.. so I'll post this. lol

http://imgur.com/a/4flfX

ID: 78344 |

Volunteer moderator
Project administrator

Send message
Joined: 21 Jan 10
Posts: 14011
ID: 53948
Credit: 433,157,676
RAC: 1,017,804

Consider this the "later" I referred to earlier...

As for GFN-WR on Haswell. If I can force BOINC to behave, maybe it will happen. If we just turn it on (something that would only take a few minutes to do), we'll have a disaster from missed deadlines and late tasks, because most CPUs can't even come close to the current deadline. We would really have to limit it to only Haswell CPUs, or, with a somewhat longer deadline, Sandy Bridge CPUs. Even then, if you have HT on you'll never make the deadline, and I can't detect that through BOINC.

While there are tools in BOINC to allow let you selectively send tasks only to appropriate computers, past experience (i.e., GeneferAVX) has shown that BOINC's implementation of this ability is less than perfect. As a result, I'm not even sure if it can be done.

Forget about this for the time being. While the newest Genefer runs one copy of a WR task reasonably quickly (17 days on this computer), it suffers tremendously if multiple instances ar running. Running on all 4 cores the time doubles. And it's off topic here, so back to SoB...
____________
My lucky number is 75898⁵²⁴²⁸⁸+1

ID: 78346 |

NeoMetal*
Volunteer tester
Avatar

Send message
Joined: 25 Mar 11
Posts: 418
ID: 92179
Credit: 1,747,428,303
RAC: 0

A lot of good ideas (and some not so good) but a simple and relatively fast and a long needed idea is an FAQ with a link on main page above the primes found column in bold such as "Please read our FAQ before deciding which sub projects to run" with as much pertinent info about minimum requirements and such as well as any other useful info about Primegrid. I think this has been long overdue.
____________
Largest Primes to Date:
As Double Checker: SR5 109208*5^1816285+1 Dgts-1,269,534
As Initial Finder: SR5 243944*5^1258576-1 Dgts-879,713

ID: 78391 |

Willy
Send message
Joined: 24 Sep 13
Posts: 19
ID: 257296
Credit: 48,339,334
RAC: 0

so without HT on @ 3.7ghz, I'm sitting at 90% done with a couple SoB tasks at 169hours. Another task is at 161 hours at 89.9% (must have been on a core that wasn't used a lot). Much better with HT off... as everyone advised me to do.

ID: 78443 |

Send message
Joined: 10 May 09
Posts: 804
ID: 39821
Credit: 1,935,039,845
RAC: 1,392,602

...a long needed idea is an FAQ with a link on main page above the primes found column in bold such as "Please read our FAQ before deciding which sub projects to run"...

Think about the basic BOINC user. They install and add projects right from the software without logging into the website itself. Since there is no need to log into the website, this idea, which I agree is good, will still not reach the masses that you may think it would.

Since attaching to the project turns nearly everything on automatically, people set and forget and never look at what they may have just joined. I am sure something could be built into the BOINC software to educate people, but then think of how many manuals people read. Again, it will reach a few, but not necessarily the desired audience.

ID: 78445 |

Volunteer moderator
Project administrator

Send message
Joined: 21 Jan 10
Posts: 14011
ID: 53948
Credit: 433,157,676
RAC: 1,017,804

...a long needed idea is an FAQ with a link on main page above the primes found column in bold such as "Please read our FAQ before deciding which sub projects to run"...

Think about the basic BOINC user. They install and add projects right from the software without logging into the website itself. Since there is no need to log into the website, this idea, which I agree is good, will still not reach the masses that you may think it would.

Since attaching to the project turns nearly everything on automatically, people set and forget and never look at what they may have just joined. I am sure something could be built into the BOINC software to educate people, but then think of how many manuals people read. Again, it will reach a few, but not necessarily the desired audience.

FYI, several weeks ago we changed some of the server code so that at least some new users get directed only to small projects. If the user doesn't select anything manually, they should only get ESP-Sieve (CPU) and PPS-Sieve (GPU) apps, at least in many circumstances. I suspect most of the problem with people running things they don't want to run occurs when they're not even aware there's more than one "PrimeGrid App". Hopefully this will help somewhat.
____________
My lucky number is 75898⁵²⁴²⁸⁸+1

ID: 78447 |

Send message
Joined: 10 May 09
Posts: 804
ID: 39821
Credit: 1,935,039,845
RAC: 1,392,602

FYI, several weeks ago we changed some of the server code so that at least some new users get directed only to small projects. ... Hopefully this will help somewhat.

Some? How is it determined? If they sign up via the BOINC software vs signing up via the webpages?

I would imagine that those that sign up via the webpages probably are more interested in knowing there are more apps. Those signing up via just the BOINC software are probably looking for the quick points. Of course it's not always the case.

What would be cool is if the BOINC user end software would give you the listing of Apps and make you choose when you join a project new from it. So you'd more or less get the listing that you do on the website where it says this project takes approximately this many minutes and warns of long run times.

ID: 78450 |

whynot
Send message
Joined: 9 Mar 11
Posts: 127
ID: 89957
Credit: 46,462,161
RAC: 5,153

It didn't took long this time. Frustration level ~6.38%, ~26.30day secured (it's not correct, that host didn't do any SOB yet). And check this out: *0sec* wasted! Literary! For a couple of days I was considering to give up timeout this time, but nine days late WU. I was surprised, seriously. Good thing I'm not hurrying anywhere.

____________
I'm counting for science,
Points just make me sick.

ID: 78642 |

KEP

Send message
Joined: 10 Aug 05
Posts: 301
ID: 110
Credit: 12,352,853
RAC: 139

It didn't took long this time. Frustration level ~6.38%, ~26.30day secured (it's not correct, that host didn't do any SOB yet). And check this out: *0sec* wasted! Literary! For a couple of days I was considering to give up timeout this time, but nine days late WU. I was surprised, seriously. Good thing I'm not hurrying anywhere.

I'm not sure how easily this can be implemented, however I do think that a solution to this frustration, could be that Primegrid started using "probational credit". What is "probational credit", besides beeing the wording I choose too describe what I think can solve this frustrating issue it is more precisely as follows:

"probational credit" is a suggestion that credit is granted emidiately as the result is reported, for big WUs like SoB, PSP, Woodall, Cullen, 321. The probation/condition is that the credit will not be final untill verification is complete. If you produced a valid result, you'll keep your score from the time you complete and reports the result, however if your result ultimately is deemed wrong (not just inconclusive) then credit will be removed/withdrawn. So "probational credit" is basically the same as we know it from the challenges, only it will now be cobblestones that is granted emidiately and not just challengepoints :)

Any thoughts?

ID: 78647 |