Message boards :
Number crunching :
Wish we could chose wingmen we would not accept?
Message board moderation
Author | Message |
---|---|
VQ-2 Ghost Send message Joined: 18 Jul 02 Posts: 55 Credit: 1,165,715 RAC: 0 |
Arrg! I wish we could turn in a list of those computers we would not accept being paired with. Why would some people have computers crunching that are 'always' returning errors, downloading more units, and again return more errors? I've had quite a few of those as wingmen, I'd be willing to bet most of us have. It's not that I don't make mistakes, I do, just not so many and so often and repeatedly as I have seen lately. I stayed away from crunching for about 2 years in the past, as soon as I came back I started getting paired with a whole bunch of such machines. My crunchers are slow enough already as it is! How do you give those wingmen a symbolic and friendly swift kick? What a waste of their time and mine... Anyhow, due to FPL's 16% increase in the in electric rates plus other moneys they are collecting monthly in case there is a hurricane and still more money they are collection to build a new plant starting in 2010, I am being forced to slowly shut down everything. 90 degree plus weather is not helping, either. I might be able to get by with just one eruncher, time will tell... |
kittyman Send message Joined: 9 Jul 00 Posts: 51468 Credit: 1,018,363,574 RAC: 1,004 |
Pending credits may frustrate some users, but you have to get used to the fact that they are part of the way the project functions..... But being paired with poor wingmen does nothing to waste YOUR time or contribution to the project as long as you are returning valid results....you will get the credit when somebody finally turns a valid result in to be paired with yours.... I still take a small hit here and there from a wingy that is running an outdated client that underclaims the credit amount, and I get nicked by his lower claim, but it seems to me that even that is occurring a little less as time goes by.. Hang in there... Now the heat and the power bills are a different story!!! "Freedom is just Chaos, with better lighting." Alan Dean Foster |
JDWhale Send message Joined: 6 Apr 99 Posts: 921 Credit: 21,935,817 RAC: 3 |
Yes, I've noticed a few also, though they do not bother me as much as wingmen running old versions of BOINC and under reporting credit. If your wingman "errors out" quickly then there is no harm as the WU is resent to another host quickly. Just rest assured that your results are saved and the credit will come, eventually. Living in SE Texas, I too feel the pain of summertime heat and running the crunchers. Since they generate heat, they add to the AC costs... And my computer room stays at ~89F... the rest of my home is just a couple degrees cooler. In winter they add just the required amount of heat. BOINC on...on... JDWhale |
1mp0£173 Send message Joined: 3 Apr 99 Posts: 8423 Credit: 356,897 RAC: 0 |
Seems fairly obvious, they don't know there is a problem.
Their time, sure, but yours? You crunch a work unit in 6,000 seconds, and you get full credit for those 6,000 seconds of work when someone returns valid work. If someone with a broken host returns bad work, the WU is reassigned, the new wingman completes it safely, and you get credit. The credit (and your time) isn't lost, it just isn't granted as promptly as we'd all like.
This is another issue altogether, and SETI has never asked anyone to "make" excess clock cycles for the project. I do not crunch 24/7, my machines that I own and use anyway crunch when they're on for other reasons. |
Luke Send message Joined: 31 Dec 06 Posts: 2546 Credit: 817,560 RAC: 0 |
If you run a celery or a P3... It can take 8 odd hours to complete a WU, only running one at a time, And then the wingman will either Not Reply by the report deadline (which can be up to a month away in some cases) or have a Client Error, And then you have to wait for another wingman to complete it, and in rare cases they don't complete it (happened to me twice I think, so far)... ...You can be waiting for up to 2 months to be granted to credit, and all this time while you wait, your RAC of 20 will drop to about 10, and then It takes a week of recovery to gain it back up to the previous levels..... Oh well.... I suppose everyone just keeps on crunching... - Luke. |
1mp0£173 Send message Joined: 3 Apr 99 Posts: 8423 Credit: 356,897 RAC: 0 |
If you run a celery or a P3... ... and how many times have you lost (not delayed, but actually never received credit) because someone didn't report? My "slug" crunches pretty much 24/7 (pretty much because there is another app that hangs the whole system sometimes) and produces a RAC of 14. Yes, fourteen. It will complete the WU it is working on in another 40 hours. ... but my wingmen always get credit, and no actual work is lost. (and yes, it has another purpose for existing, and running 24/7) |
kittyman Send message Joined: 9 Jul 00 Posts: 51468 Credit: 1,018,363,574 RAC: 1,004 |
That's ok Ned, you can be my wingy anytime....... "Freedom is just Chaos, with better lighting." Alan Dean Foster |
1mp0£173 Send message Joined: 3 Apr 99 Posts: 8423 Credit: 356,897 RAC: 0 |
But at this speed, probably won't happen very often. :-) P.S. (edit), it's 40 hours in, and half-way. |
zoom3+1=4 Send message Joined: 30 Nov 03 Posts: 65750 Credit: 55,293,173 RAC: 49 |
I've got one of the two licked, The bill though comes every month, But I'm workin on It, I have to get some plumbing work saved up for 1st(33"x19"x6" Stainless Steel Kitchen Sink and 150gpd RO filter w/a 4G tank installed for about $290 or $72.50hr + $30.00 trip charge), Then 2 more dual pane windows ordered and installed($350 each), Always work around here to be done, If I'm lucky It'll only take 11 to 24 months to save up for(allowing for the DMV once a year of course), But that's the priority now or at least starting in August or so. :D Right now I have a cooler to get home and installed, then when the 3 retrofit windows arrive I'll have to have some silver window film added, then 3 of 7 Windows will be done, 2 are on the ends of the house(bathroom and a bedroom used for storage) and the other 2 are on the east side, The the main 3 are on the west side where the house gets the most sun. :) The T1 Trust, PRR T1 Class 4-4-4-4 #5550, 1 of America's First HST's |
Luke Send message Joined: 31 Dec 06 Posts: 2546 Credit: 817,560 RAC: 0 |
If you run a celery or a P3... A few times, not many, I'm just saying it's a pain in the neck when it isn't reported on time. Once you've got a decent RAC, you don't really seem to take notice of individual tasks, you forget all about it ;) - Luke. |
1mp0£173 Send message Joined: 3 Apr 99 Posts: 8423 Credit: 356,897 RAC: 0 |
If you run a celery or a P3... Luke, It's really easy to say "I didn't get credit" but can you state that with any certainty? I mentioned my "slug" -- when I look at what it has accomplished, there are no completed work units. It always reports last, and completed work always transitions quickly -- they just don't show at all. At the moment, under that computer, I see two work units: the one it is crunching, and the one it will start when this one finishes. The only time I catch a "completed" WU is when I know it has just finished (a minor event around here). Yet, it has a RAC of 14, and the credits keep trending up. So, while I understand that you'd like to see everything returned promptly, "reported on time" is relative. ... and you'll always get "paid." -- Ned |
Luke Send message Joined: 31 Dec 06 Posts: 2546 Credit: 817,560 RAC: 0 |
I've been trying to say people always get some credit for it at the end (whether this be the full credit or a portion of it, I have had two times when I get a portion of the credit I was granted)... just not as quickly as some people would like. As I said before If you run a P3 (or around the same) it can be a pain when a wingman doesn't report it or a client error occurs... Hope that clears things up... Best Regards, Luke. - Luke. |
Josef W. Segur Send message Joined: 30 Oct 99 Posts: 4504 Credit: 1,414,761 RAC: 0 |
... and how many times have you lost (not delayed, but actually never received credit) because someone didn't report? A WU discarded because of too many errors? I haven't seen that except in cases where the splitter had produced bad WUs, but you're right that it could happen on a sequence of 4 wingmen who didn't do the work they accepted. The lost credits for a case like that would not concern me at all, the fact that any signals in the WU would not be entered in the master science database would sadden me. Joe |
1mp0£173 Send message Joined: 3 Apr 99 Posts: 8423 Credit: 356,897 RAC: 0 |
I've been trying to say people always get some credit for it at the end (whether this be the full credit or a portion of it, I have had two times when I get a portion of the credit I was granted)... just not as quickly as some people would like. As I said before If you run a P3 (or around the same) it can be a pain when a wingman doesn't report it or a client error occurs... Luke, There is no such thing as "partial credit." Two copies of a work unit are sent out, it is crunched by two different computers, and the work is returned or times out. If one (or both) time out, a third copy is sent. The two work units are compared, and if they are only weakly similar, or don't match, then a third work unit is sent out. This continues until two strongly similar work units are received. Then credit is granted. The rule for credit is that the highest and lowest requests are thrown out, and the average of the remaining requests is filled. If there are only two, then the highest is thrown out. One result never happens on SETI, but there is a rule for that. If you are unlucky, and get someone running an old version of BOINC, they can request less credit than they should. This has everything to do with the BOINC versions being run, and nothing to do with time. Crunchers running old versions of BOINC can report quickly, and you will get less credit. So, your statement that you got "partial credit" because the other guy was late is simply not correct. -- Ned |
Luke Send message Joined: 31 Dec 06 Posts: 2546 Credit: 817,560 RAC: 0 |
This is all a misunderstanding: (I'm not arguing for or against the case of choosing wingmen, I'm just trying to give my two cents...) I've been trying to say people always get some credit for it at the end (whether this be the full credit or a portion of it, I have had two times when I get a portion of the credit I was granted)... just not as quickly as some people would like. As I said before If you run a P3 (or around the same) it can be a pain when a wingman doesn't report it or a client error occurs... Should have been:
Luke. - Luke. |
Azzitude Send message Joined: 10 Mar 08 Posts: 33 Credit: 1,832,287 RAC: 0 |
those are GREAT explanations, but it still leaves a question of refusing to work with certain people or at the least match the machine speeds and not have 1 wicked fast and one fossil slow working the same WU's. just match em up relatively same speeds is all I would ask |
JDWhale Send message Joined: 6 Apr 99 Posts: 921 Credit: 21,935,817 RAC: 3 |
those are GREAT explanations, but it still leaves a question of refusing to work with certain people or at the least match the machine speeds and not have 1 wicked fast and one fossil slow working the same WU's. I think what you are asking for is too restrictive... What happens when 95% of the computers are very much slower than yours? You might not be able to DL work because all the WUs currently being assigned are already issued to slower computers. Would you rather not be issued any work than being issued work where your wingman is slower than you? Similarly, if 99% of the computers are faster than you, then you might not get work because you are too slow and no "compatible" wingmen are currently available. The schedulers are already complex enough and IMO restrict DLs too much when my hosts request work.... I have a host that continuously requests 7 days of work and are only granted 5 or less WUs per request. This host crunches 50 WUs per day and can only fetch 5 WUs per request! What you are asking might only allow a single WU to be DL per request... at that rate it might take 2 hours to DL the 100-200 WUs required to raise my cache to an acceptable level. That places too much burden on schedulers that are already overloaded. No thank you... Regards, JDWhale |
kittyman Send message Joined: 9 Jul 00 Posts: 51468 Credit: 1,018,363,574 RAC: 1,004 |
Agreed.......With the immense variation in hosts participating in this project, trying to implement an algorithm to try to match computer speed when issuing work would just place an incredible load on the servers...already pushed awfully hard as it is...... And what kind of luck would I have trying to keep the Frozen Penny supplied with work if it could only be matched to the handful of rigs on the project that can crunch at it's speed? I think I'll take my chances with the slow wingmen.....my pending is fairly stable now at around 46,000.....so what goes in is coming back out at about the same pace...... As far as the project itself is concerned, it makes no difference....if 10,000 WUs are issued to a given number of hosts, it will take the same aggregate time for those hosts to process them and return the results....regardless of how those hosts are paired. "Freedom is just Chaos, with better lighting." Alan Dean Foster |
JDWhale Send message Joined: 6 Apr 99 Posts: 921 Credit: 21,935,817 RAC: 3 |
Ehll would freeze over before Penny could load up with enough work. Another perspective on the whole issue that is taught at my former employer is that... "Diversity in the workplace is a positive." I think this mantra can be carried into the scheduling world. The last thing that the project needs is to always have the same sets of hosts paired together. From the validation perspective it is better to have the hosts "randomly" paired to limit introduction of possible patterns based on wingman selection. That sort of behavior could undermine the validity of results.... much like politics. |
Josef W. Segur Send message Joined: 30 Oct 99 Posts: 4504 Credit: 1,414,761 RAC: 0 |
A relatively simple approach (at least while we only have setiathome_enhanced) would be to classify hosts by their average turnaround time. For two groups, the dividing line could be the overall average turnaround shown on the server stats page. That would work out that half the work would be going to hosts with longer turnarounds and half to quicker hosts. However, any such scheme will increase the load on the servers and I don't know if the S@H servers could handle it. I do know that there's some discussion going on in the background about related things. For instance there's a Java/Android port of BOINC and SETI which may allow users to crunch on their cell phones at least while the phones are being recharged, but smaller WUs would be more appropriate. It could be that those smaller WUs could be used for the slowest x86 hardware too. I still don't know any logical reason that one of JDWhale's hosts is getting less work than requested, nor do I know how widespread that extra load on the servers is. My host which had the same effect for awhile is doing AstroPulse beta testing now, and with the length of those WUs I don't think we can check whether the SETI Beta project servers show the same illogic. Joe |
©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.