What? Sending a 4th WU after 4 days? So now the deadline (with respect to a 4th WU) is 4 days instead of 3 weeks. WCG has effectively changed the rules of the game but not announced the change in a way that everyone will see it. How can we crunchers be expected to do our part of the job properly if we don't know what the rules are?

This is no reflection on you personally, lawrencehardin, but the new 4 day limit is extremely poor foresight. With management like that the new system of 3 WUs is doomed to fail and we'll soon be back to 4 WU.

Hi Dagorath
I think you may have misunderstood Lawrence's comments
If a WU is sent out and not validated within 96 hours then another one will be sent out to attempt to validate the quorum
This has nothing to do with the expiration period of the WU, as it still has the same 21 day validation period before it expires
I believe that what Lawrence is trying to convey, is it would be in all the members best interests to ensure that they do not cache more than four days of WU's at any given time
Otherwise a 4th WU could end up at the end of another members four day queue and will not help matters in getting a quorum validated
This would have the effect of a member waiting for validation of a WU from a member who will not get to process it for 4 days by which time a 5th would invariably be sent out in an attempt to validate the original 3 WU's
This would effectively waste valuable processing time
I hope this makes sense

There are a few parallel discussions around here, in the BOINC block and other sections.......UNinteresting to see our moderators/community advisors now start throwing ".....wide of the mark" comments even if little. I'm not really interested into polemics, and i think most other contributors crunchers/advanced crunchers are not (how did i become an AC?), but let me indulge....what is (a little) wide of the mark? A clear observation is:

1. WU sets of 3 are send out initially
2. After 4 days, if quorum of 3 is not found, additional 1 or 2 copies are send out (I just received one that has copy 4 and 5 send out within minutes)....i'm honoured and pulled if forward in the BOINC queu.
3. The 3 week rule remains......which i think is a big driver in peoples thinking that caching work for longer periods is a perfectly valid approach.

Dagorath is exactly on the mark with his "WOT?". Personally, i think the 3 weeks rule has lost its use entirely, since WCG or who-ever seems to have decided not to wait for those, which i guess are the slow machines. I'm really interested as to why 4 days and not 1 week. Anyone taking 1 week to crunch on a unit has effectively been devalued to wasting his time since it will pretty much always be copy 4 being returned (and that was fine since always 4 units were send). If WCG is not interested in them, be fair and tell them.....an overshoot for i thought the volunteering thing was intended to be with a spirit of the more the merrier long as a machine has a minimum spec.

I cache 2 days and when i know to be away for a week or more the period of planned absence, for i dont let my machine hooked to the grid when away. You're telling WE/US (not people)....switch off the box as you're wasting your time, electricity and ours, or are you?

So Lawrence, please experiment with a longer period than 4 days .....we'll get used to the longer pending validation period....we do it for curing deceases, not points.....that's just a the game of psychology (wont call it a psycho-game). The door should be open to as many as We can take.

thanks and good night .

I'm also confused as to how the report deadline can still be given as 3 weeks, when clearly under the new system a WU copy will always be useless to the project well before that time.

To me that's the essential definition of the deadline; how late can I turn this WU result in and still have it be worthwhile? This lets volunteers with slower machines or infrequent connections make informed choices about which projects they can effectively contribute to. No volunteer wants to crunch in vain, with their results just being logged and ignored, not actually contributing to the real science at all. Even the most score-obsessed like to feel that they're accomplishing something while racking up the points.

There's an implicit promise that a result turned in 1 hour before deadline is still valued and worthwhile, and not a waste of electricity and server bandwidth. This implicit promise is even more important for those with slow systems. If I just spent 2 weeks slowly and painfully crunching through a WU on my 100MHz museum piece, I sure don't want it to have been wasted effort.

And with BOINC, there's the added issue of CPU share. Many BOINC users like to participate in several different projects at once, and even a non-slow machine can take a long time to process a WU if the project is allocated a low CPU share. Again, we depend on the deadline to tell us how low a CPU share is still worthwhile.

Finally, the deadline is vital to the BOINC manager's scheduling algorithm. With accurate deadlines, the manager can shuffle jobs to prevent WU waste. When deadlines are set beyond the actual useful life of the WU, you lose this valuable optimization.

Under the old system, a 2.9 week old result still had some value as a backup in case one of the other 3 copies had failed (although a slow system that routinely took that long to return results would *only* be filling in for failed copies, thus wasting the great majority of all that slow and painful crunching).

But now, a 2.9 week old result is clearly worthless. The WU will have been reissued to someone else long since, several times if necessary, and the final verified result saved weeks ago.

To me it seems the reissue time is now the natural deadline. If you turn in your result before the reissue time, it's important and valuable. If you haven't turned it in yet by then, your WU will be reissued to a replacement system, and your copy becomes mostly redundant. The only real value for a result turned in after the reissue time would be if the replacement copy fails to complete, and since replacement systems are specially chosen for speed and reliablility, that won't happen much. It might also help to reach a quorum a bit faster and speed up validation for the first 2 returned results, but again, the replacement systems are chosen for speed, so the difference will be small at best.

Also, I think the focus on cache times so far is a bit incomplete. What matters isn't just how much work you keep cached, but how long it takes you to turn in your results. To prevent wasted effort, your WU must be finished and reported before it gets reissued to a replacement system. Once your WU gets reissued, a copy has been wasted (either yours or the replacement, however you want to look at it), and it doesn't matter how long it sat in your queue before you got to it except in as much as that contributed to your not getting it done in time.

So, turn in your WU result before the reissue time, and it's definitely good and useful. Turn it in any later, and it's very likely redundant and pointless. That sounds pretty much like the expected definition of "report deadline" to me. The first 3 copies of a WU should therefor have a stated deadline of 4 days (or whatever duration is finally settled on). And to remain accurate, a reissued copy should have a stated deadline of 1 day or however long the "fast & reliable" replacment system is given before the WU is reissued another time.

This method will still be correct regardless of what final reissue duration is chosen. Longer reissue times will allow slower machines to usefully contribute, but will increase people's average validation queues, and I'm sure we're in for much experimentation to find the proper balance. But however it settles out, the deadline should accurately reflect the reissue time, which will likely never be anything like 3 weeks.

I'm sorry if this upsets those with slow machines/long connect intervals/low WCG CPU shares and makes them feel like their help isn't wanted, but the brutal truth is that they probably weren't really doing as much good under the old system as they thought anyway, being mostly stuck with submitting 4th copies which were mostly unneeded. At least this way people will know the real situation and can reallocate their resources more usefully. From this project's point of view, the new WU scheduling system is a huge benefit and should greatly increase the project's total WU throughput, which is something we should all be happy about.

And now I think that's quite enough of a brand new (but not newb) project participant telling all the old hands how things ought to be run. Thanks for reading this far.

I think RickH has made a valid point relative to the current projects that WCG is running as far as I understand it.

Clearly the recent change in validation was made to improve the efficiency of the workforce, and RickH is pointing to what appears to be an unintended consequence of that change that exacerbates an existing inefficiency.

There could be something I am missing but I think this warrants a through analysis by the development team.

Edit => This post was not intended for this thread. Sorry.