Message boards :
Number crunching :
The Plan
Message board moderation
Previous · 1 · 2
Author | Message |
---|---|
Sergey Broudkov Send message Joined: 24 May 04 Posts: 221 Credit: 561,897 RAC: 0 |
It's not really a matter of credit. It's a matter of making a useful contribution to the project. In my opinion if my computer spends time processing a unit for which a quorum has already been reached the time spent on that unit is wasted. Don't misunderstand me. You see only one side of the matter. The redundacy is the essential part of the project, so even those who don't provide canonical result or don't make a quorum, they are nevertheless are taking part, and their input is as valuable as other's. Consider it as a sport team. There are attackers who score goals, but there are also defenders, coaches, doctors etc, and they ALL TOGETHER win a cup. EDIT: typos Kitty@SETI team (Russia). Our cats also want to know if there is ETI out there |
W-K 666 Send message Joined: 18 May 99 Posts: 19062 Credit: 40,757,560 RAC: 67 |
As for the three month deadline, it will increase the length of time that information is stored on the drives at berkeley, further burdening an already overloaded file system. Files pertaining to a particular work unit will be held at least three months, possibly much longer if a quorum isn't reached at the end of the three months. This could mean a delay of three months, six months, or longer before credit is granted for some units. Me thinks you answered you own concerns there about the need for the fourth unit. From observation and from what other people have said the number of units not returned before deadline or are not validated has decreased for the 1 in 4 of six moths ago but it is still significant. With a three month deadline, if they don't send the fourth unit then, are the majority prepared to wait up to six months for the credits to be granted on 10% or more of their units. My hunch is that because of the longer crunch times and the need for the BOINC manager to calculate the correction factor, there is going to be significant rise in units aborted or not returned. Especially those who use an optimised app, because their friends said it was a good idea, and are used to crunch times measured in 5 hrs or less, and then see it has suddendly risen to a predicted 100's of hours, and don't see, don't realise that it now says Seti@home ver 5.n.x and not 4.18. [editted for typo's] |
tekwyzrd Send message Joined: 21 Nov 01 Posts: 767 Credit: 30,009 RAC: 0 |
I see a strong possibility that the changeover to the enhanced app as planned could lead to a decision to set minimum processor requirements and eliminate many computers from the project. How do I explain this? As things stand, there's little difference in returning units that complete in 1 hour vs 4 hours. This is due at least in part to work unit caching habits, which will have less of an effect with the enhanced version. When the changeover is made you're going to have differences in run times of days (if optimized apps are available) to a week or more (if no optimized apps are available). As a result, files will reside on the berkeley file system longer. It's likely that some remedy to the buildup will be necessary. A simple solution would be to set minimum processor requirements. Assuming the use of unoptimized seti enhanced, if a work unit is issued to a mix of machines including a p2 400MHz, a p3 800MHz, and a p4 3.0 GHz, the p4 will likely return results the same day. The p3 will likely return the result a week later. The p2 will likely return results two weeks later. In this case, assume a quorum is reached by these three computers. This means deletion after about two weeks. In cases where a quorum is not reached by the three returned results, and the fourth isn't returned, the files will be in the file system for over three months. As files like this build up, file system performance will be further degraded (in a manner similar to that caused by of the buildup of orphaned units). An easy solution would be to eliminate computers below a certain performance level, thus reducing the difference in processing times, interval between result returns, and the amount of time a set of files reside on the file system. Nothing travels faster than the speed of light with the possible exception of bad news, which obeys its own special laws. Douglas Adams (1952 - 2001) |
1mp0£173 Send message Joined: 3 Apr 99 Posts: 8423 Credit: 356,897 RAC: 0 |
I see a strong possibility that the changeover to the enhanced app as planned could lead to a decision to set minimum processor requirements and eliminate many computers from the project. I suspect that the slower pace will mostly solve that problem, but a simple solution is to assign work to machines that work at a similar pace. |
tekwyzrd Send message Joined: 21 Nov 01 Posts: 767 Credit: 30,009 RAC: 0 |
I suspect that the slower pace will mostly solve that problem, but a simple solution is to assign work to machines that work at a similar pace. Maybe I should have said easiest solution. I may be mistaken but I seem to remember seeing the idea of assigning work to similar computers mentioned in another thread some time ago but it was rejected due to an associated need to revise the server software and possible impact on performance. Nothing travels faster than the speed of light with the possible exception of bad news, which obeys its own special laws. Douglas Adams (1952 - 2001) |
rsisto Send message Joined: 30 Jul 03 Posts: 135 Credit: 729,936 RAC: 0 |
I suspect that the slower pace will mostly solve that problem, but a simple solution is to assign work to machines that work at a similar pace. I think this is already implemented, but it is only used when they need to send a fith (or higher) unit. In this cases they send them to host with low Average Turnaround Time. |
W-K 666 Send message Joined: 18 May 99 Posts: 19062 Credit: 40,757,560 RAC: 67 |
I suspect that the slower pace will mostly solve that problem, but a simple solution is to assign work to machines that work at a similar pace. It's not similar computers, it's similar turnaround times, that needs to be the deciding factor. i.e. if host 'A' has 0.5 days connect time and crunch time is between nine hours and 12 hrs, then average turnarround would be slightly lower that 0.5 days. While host 'B' has connect time of 10 days and crunch of between one and two hours, the average turnaround would be about 10 days. I'm assuming correct sized caches and connection is always available. |
Ingleside Send message Joined: 4 Feb 03 Posts: 1546 Credit: 15,832,022 RAC: 13 |
It's not similar computers, it's similar turnaround times, that needs to be the deciding factor. Since we're talking about Seti_Enhanced, host A will not have between 9h and 12h crunch-times, but more likely between 9h and 135h, while B will have 1h-15h. Meaning, host A's average turnaround-time on "normal" angle-range should be around 4 days. If host B has decreased cache-size to 4 days they'll both have similar turnaround-times and can be paired together. If it's another "normal" angle-range, no problem, both will report on roughly the same time. But, if it's a "fast" result, host A will report 3 days before host B, while if "slow" result, host A will report 2 days after B. Now, let's keep host A, but change to a host C, with cache-setting 0.1 days and cpu-times from 6 days to 90 days. If host A has just crunched-through a bunch of "slow" results and host C a bunch of "fast" results, they can be paired-together. If they gets assigned a "fast" result, host A will report 5 days before C. If instead they get assigned a "slow" result, host A will report 80+ days before host C... Anyway, don't remember if the idea to send copies to hosts with similar average turnaround-times has ever been finalized and it's just like some other things that not all projects is using it, if it's still on the "to do"-list, or if it's been scrapped due to various weaknesses... |
W-K 666 Send message Joined: 18 May 99 Posts: 19062 Credit: 40,757,560 RAC: 67 |
What I didn't make clear was send the four copies of a unit to computers with similar turn round times, but I agree there are too many factors in the equation for any sort of pairing to be practical, even if you factor out the unexpected. |
zoom3+1=4 Send message Joined: 30 Nov 03 Posts: 65747 Credit: 55,293,173 RAC: 49 |
Hi all, I'm about to layout the conversion plan the best I can piece together. Nothing I say here should be interpreted as coming from Berkeley. I'm seated comfortably in my office chair at my home in South Carolina. There the disclaimer is out of the way. Neat, Of course I never received an email at My verizon.net address(I'm on Boinc 5.2.13 now and crunching optimized for sse2&A64), Was this email sent by an SMTP client or PHP? As Verizon DSL requires Server Authentication for all Incoming & Outgoing email and Comcast may not only be experimenting with It, But going in that direction, Heard It on www.dslreports.com of course(about comcast), I've found 21 forums that are verizon.net email compatible so far: Verizon.net email compatible forums list. The T1 Trust, PRR T1 Class 4-4-4-4 #5550, 1 of America's First HST's |
Astro Send message Joined: 16 Apr 02 Posts: 8026 Credit: 600,015 RAC: 0 |
I was glad to see Matt mention the new app, but I'd feel better if he'd have inserted a timeframe....oh well Here is a link to what developer Matt Lebofsky said earlier today |
kevin & seth Send message Joined: 30 Nov 05 Posts: 128 Credit: 258 RAC: 0 |
mmciastro said in Message 214340 : fpmmpfmf mfpmffp pfmrfpf pfmp ffmm p ppfmmf pfpmpp ppfmfpmmpmm pmfmfmp ffmpmmpm mpfmppmfpmfpm. mmfpmfpmfmp mpfmfpmfpmfpppfmp 424 mmfpmmpfmfmp mfmfmfp mfpmfp mfmp. mmpfmpfpmf 4443 pmfmfmm fpfmpmfpmpfpm ppmppmpppfm fpmfpmpp pmfpmpfmpfpmp pfmmppfpffm fpmpfmpfpmp. mfpmpfpmfmpppmf mmfpf fffmp mpfmpmppppmffp pfmpfmpfpfpp pmfpmmmfpp pmfpmffmpmmpm mppmpmmfp pmmmf. What plan? |
Astro Send message Joined: 16 Apr 02 Posts: 8026 Credit: 600,015 RAC: 0 |
Matt, gave an update, so I guess they're pushing off the release of the "enhanced" application a couple weeks. See Matts' Post which states: Matt Lebofsky Forum moderator Project developer Project scientist Joined: Mar 1, 1999 Posts: 287 ID: 122079 Credit: 35,864 RAC: 25 Message 214503 - Posted 15 Dec 2005 5:09:00 UTC Enhanced is still a couple weeks off. These are all regular ol' splitters. - Matt ____________ -- BOINC/SETI@home network/web/science/development -- Get the inside scoop: Matt's Unofficial SETI Log |
kevin & seth Send message Joined: 30 Nov 05 Posts: 128 Credit: 258 RAC: 0 |
mmciastro said in Message 214340 : So, presently, SETI is still broken? :-( |
TPR_Mojo Send message Joined: 18 Apr 00 Posts: 323 Credit: 7,001,052 RAC: 0 |
There is no validation queue, upload queue, server issues, database issues, nada (apart from the usual top performance from overworked hardware, software and personnel). The only slight issue at the moment we have is a lower number of units to send out - but the large number of splitters online will sort that. So HOW precisely is it broken? But of course you already know all this, troll. Join your buddy in my ignore list. |
Paul D. Buck Send message Joined: 19 Jul 00 Posts: 3898 Credit: 1,158,042 RAC: 0 |
Here is a coincidence ... we agree again ... How have you been? I had not seen you much until recently ... |
TPR_Mojo Send message Joined: 18 Apr 00 Posts: 323 Credit: 7,001,052 RAC: 0 |
That must be about twice we have agreed then! Off-topic (sorry Tony) but hello Paul :) I have been busy busy busy I run an English Pub (bar/restaurant) and have the dubious honour of being chief cook. So this month is hmm... passably frantic for me 8-0 Still here, still Boincing, still unhappy about benchmarks, so nothing has changed :) |
Dr. Bob Send message Joined: 1 Apr 03 Posts: 78 Credit: 623,977 RAC: 0 |
[quote]Hi all, I'm about to layout the conversion plan the best I can piece together. Nothing I say here should be interpreted as coming from Berkeley. I'm seated comfortably in my office chair at my home in South Carolina. There the disclaimer is out of the way. Thanks Tony, Many of us had no idea of the plans and the ongoing application of them. Maybe I will look to add other BOINC projects to use my client computer time that may be excess now. Since as of this date, there may be reasons of enhanced efficiency that mean I or other client users won't get new data from SETI? Thanks again...keep on keeping us updated. Best, Dr. Bob Robert L. Hanson, Ed.D. |
©2024 University of California
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.