Posts by Landroval
|
21)
Message boards :
Number crunching :
What's more important to faster SETI processing times? Floating point, or Integer ops? TIA!
(Message 220645)
Posted 24 Dec 2005 by  Landroval
Post: My bet's with floating point, because I'd guess the nature of the data being processed is best done via that route (floating point calcs vs. integer ones). Benchmarked CPU speed bears only a tangential relationship with time to crunch. In part this is because making a benchmark that accurately predicts performance is much more difficult than it sounds, in part because there are a lot of things (L2 cache being the biggest) that affect crunch times that aren't measured by the benchmarks. The best way to accurately determine which version produces faster crunch times on YOUR hardware is to crunch a few units under each and compare. Cheers, Brian |
|
22)
Message boards :
Number crunching :
correction factor?
(Message 219606)
Posted 22 Dec 2005 by Landroval
Post: with your permission practice is different the 7 workunits i was calculating with the optimised client came back with claimed credits of 16,68 / 16,69 /16,51 / 11,74 /11,83 /12,6 /11,37 that does not look like a coincidence because before the claimend credits where way up in the twenties or thirthies. [edit: the original post was rather rude in places. Edited for tone.] The DCF is part of the boinc client, not the science app. You said you were running an optimized *client* and had a question about the DCF. The DCF doesn't affect your claimed credit. I thought that was your question. My mistake. Claimed credit is based on benchmarked CPU speed * time to complete. An optimized science application completes the work in less time. Therefore the estimated (claimed) credit is off. Solutions: 1) Run a longer work queue, so the units have made quorum (and credit is calculated) before your result is returned. This is what I do on one of my hosts--I run an optimized science application (which lowers the credit) under Linux (which also tends to have lower benchmark scores, even on the same hardware). So the claimed credit from that machine averages about 4 to 5. I run a longer work queue so my low claim isn't usually counted as part of the quorum. 2) Some 3rd party optimizers have produced "optimized" BOINC clients that inflate the claimed credit. Many consider this to be at least borderline cheating, but they're out there. 3) The _enhanced application, due to be released Real Soon Now, uses an actual count of operations rather than estimates based on a flawed benchmark. From what I've heard from the developers, claimed credits are very consistent regardless of hardware or OS. So within a couple of weeks the entire issue becomes moot. There's more on claimed & granted credit in the Wiki. Now if you'll excuse me, I've got to get back to sucking my fingers. |
|
23)
Message boards :
Number crunching :
correction factor?
(Message 219594)
Posted 22 Dec 2005 by Landroval
Post: so you say that if i do twice the work i get half`payed? seems paradise for komunists The correction factor has NO EFFECT on your claimed or granted credits. Its ONLY use is estimating how long a work unit REALLY takes on your system, so the manager knows how many work units to download if you want, say, 3 days worth of work. Claimed credit and granted credit are NOT AFFECTED by the DCF. |
|
24)
Message boards :
Number crunching :
correction factor?
(Message 219574)
Posted 22 Dec 2005 by Landroval
Post: still does not explain why this correctionfactor is applied now and before was not the correction factor was introduced with (I believe) version 5 of BOINC. It wasn't present in earlier versions of the client. Addendum: In earlier versions of the client, some people would have to request 10 days worth of work to get 3 days' worth. With the correction factor, the client adjusts its estimate of how much the work on hand will take to finish, so if you request 5 days worth of work, you'll get 5 days worth of work. |
|
25)
Message boards :
Number crunching :
correction factor?
(Message 219570)
Posted 22 Dec 2005 by Landroval
Post: After downloading "optimised client software i suddenly have a new line saying correction factor = 0,54 in my Desctop and 0,32 in my laptop. What exactly does this mean, does this compensate the higher velocity i get with the optimized client and is this ment as reduction of my points to punish better usage of my chipset? The estimate of time needed to complete a workunit can vary wildly. The "correction factor" is applied to the estimate, to produce a more accurate estimate of how much work to keep on hand. It has no effect on claimed or granted credit. Edit: It means your desktop takes about 54% of the estimated time to complete a WU, and the desktop 32%. Cheers, Brian |
|
26)
Message boards :
Number crunching :
Cache'ing work
(Message 219548)
Posted 22 Dec 2005 by Landroval
Post: I figured it was something simple. Thanks for the tips gents. I have mine set at .9 now. Now I need to figure out how to update the setting on my computer. On the "Projects" tab, select the project and hit the "Update" button. Or wait until the next time the BOINC client contacts the server and it'll update on its own. Cheers, Brian |
|
27)
Message boards :
Number crunching :
No work from project
(Message 219342)
Posted 21 Dec 2005 by Landroval
Post: As for what cache size... That is almost a religious question with just as many "interpretations"! If you run a large cache, it probably won't delay credits unduly, as most of the time the other 3 units will be returned & validated first. However, you will be returning the 4th work unit most of the time, which can touch off the "4th work unit is wasted effort" religious war, so there's still room for "interpretation". [snicker] I'd just let BOINC handle it, and if someone else gets some crunch time over the holidays, that's fine. I've got one machine that mostly does SETI, with Einstein at a small resource share as backup. During this last outage Einstein got some extra time, and it'll probably be months before it gets to download again...unless SETI runs dry. |
|
28)
Message boards :
Number crunching :
No work Sent
(Message 218833)
Posted 20 Dec 2005 by Landroval
Post: Done they whole lot to 3 as Sztaki was 1 and someone else was 1 as well. Posted a message on CPDN on how to clear disc as it is now 960 MB another 4MB in the last hour can anyone help me here? CPDN leaves some files on your drive after the work unit is finished. (A summary is uploaded, detailed results are on your drive.) If something from the summary data catches the eyes of the scientists, they'll request those files from your computer. Those files can be moved to a different folder, so they don't count against your BOINC total, or deleted. If they're deleted and the scientists want the detailed data, they'll need to run that simulation again. But it'll free up the space on your drive. Here's a Wiki article about it. Cheers, Brian |
|
29)
Message boards :
Number crunching :
No work after reinstalling Boinc, Boinc thinks I have work, so no d/l
(Message 218830)
Posted 20 Dec 2005 by Landroval
Post: 12/20/2005 11:41:09 AM|SETI@home|No work from project Most of the splitters are turned off. There isn't much being generated at the moment; no one's getting much, lots of us aren't getting any. Once the splitters are back up, the situation should take care of itself. This kind of thing is exactly why BOINC is set up to support multiple projects. If one project doesn't have work, for whatever reason, you can crunch for someone else. Once you can get work from SETI again, the scheduler will see to it that SETI gets some extra CPU time until the long-term resource shares are in line. In short, it's behaving exactly as it's supposed to. Happy crunching, Brian |
|
30)
Message boards :
Number crunching :
Enhanced version 10x longer!
(Message 218656)
Posted 20 Dec 2005 by Landroval
Post: i am currious as to why my Athlon XP 3000+ isn't working with an optimized client...if you have a suggestion much appreciated. Is the optimized client CPU-specific, i.e. different optimized clients for Intel vs AMD chips? |
|
31)
Message boards :
Number crunching :
I haven't gotten a new WU in over 6 hours
(Message 218653)
Posted 20 Dec 2005 by Landroval
Post: If I increase the time in "Connect to network about every X days." will it maintain that cash size or will it not get new WU until the end of the set time? It'll maintain about that cache size. It'll fluctuate up & down a little. I've got one machine I run a 6 day cache on. At any given time the work on hand varies from about 5 1/2 days to 6 1/2 days or so. |
|
32)
Message boards :
Number crunching :
I haven't gotten a new WU in over 6 hours
(Message 218645)
Posted 20 Dec 2005 by Landroval
Post: I haven't gotten a new WU in over 6 hours. Whats going on? How can I have a cash of WU like I had with SETI classic? A couple of the splitters were offline for a while. The current queue of work waiting to be sent is a mighty 256 work units at the moment. They're being downloaded as soon as they're being produced. The splitters are back online and should catch up sometime later today...at least, that's the word from UCB. [edit: Nope, they're still offline at the moment. Check the server status page for up-to-date info.] To set a cache, go to Your Account, General Preferences. There's a setting for "Connect to network about every X days." That can be set for anything up to 10 days. If you want to have several days' worth of work on hand, I'd suggest moving it up gradually--say, setting it to 1 day, then bumping it up to 3 days tomorrow, then 5 days the day after. (Sometimes the client scheduler goes a little nuts on downloading new work after that setting is changed.) If you set the "connect every" to a higher number, then as work becomes available, you'll download a store of units to have on hand. Happy crunching! Brian |
|
33)
Message boards :
Number crunching :
Earliest Deadline First
(Message 218637)
Posted 20 Dec 2005 by Landroval
Post: You can't have a different "connect" setting for the different projects, unless you set up different venues. Actually, I'm not even sure the one computer can have different venues for the different projects?? So, EAH will not be 0.1, it will be whatever Seti is - around 2 days or so at a guess. I don't know about the different-venues-for-different-projects; I've not tried it. In general, though, the BOINC client will use whichever set of preferences has been updated most recently. FWIW, I agree that in the long run, the scheduler will take care of it. A project gets a few hours extra time today and gives up a few hours tomorrow, big deal. And if you have a host that's consistently claiming lower credit, running a longer work queue on that host may be a good idea. |
|
34)
Message boards :
Number crunching :
Enhanced version 10x longer!
(Message 218486)
Posted 20 Dec 2005 by Landroval
Post: does running an optimized BOINC client takes (longer, shorter, same) ammount of time compared to 5.2.xx There shouldn't be any noticeable difference in crunch time running an optimized BOINC client. The main difference is claimed credit. The optimized BOINC client may handle some of the overhead tasks more efficiently, but that's a small part of the total load. Careful measurement may pick up a nonzero difference, but I wouldn't expect anything dramatically noticeable. (Obviously, crunch time for optimized science applications will be different, but should be about same for optimized BOINC.) Cheers, Brian |
|
35)
Message boards :
Number crunching :
Totals are out of date
(Message 215669)
Posted 16 Dec 2005 by Landroval
Post: My SETI credits under BOINC only reflect my SETI credits as of March 15, 2005. What about all the WU's processed from March 15 to Dec 15? Once everything in Classic is closed down, there's going to be a final update to show your totals. Cheers, Brian |
|
36)
Message boards :
Number crunching :
Any chance of having classic #'s updated on Boinc account?
(Message 214483)
Posted 15 Dec 2005 by Landroval
Post: Since switching to Boinc I still had computers running SETI classic. Is there any way (or plans)to have one last update from the old SETI account to the new BOINC account to show additional units crunched since switching? Yes. Once Classic is closed there's one last update planned, to show your total number of Classic workunits processed. |
|
37)
Message boards :
Number crunching :
recieving NO CREDIT
(Message 214261)
Posted 14 Dec 2005 by Landroval
Post: There's one problem with "answers".......they always lead to more questions. Why do you think that is? [grin, duck, RUN] |
|
38)
Message boards :
Number crunching :
Difference in computing
(Message 213646)
Posted 14 Dec 2005 by Landroval
Post: I have noticed that some PC's with pretty much same hardware takes half of the computing time as mine, A couple of things: 1) The benchmark speeds measure CPU performance ONLY, and there are other things (such as L2 cache) that can affect crunch times but aren't reflected in the benchmarks. 2) Some people run optimized clients. The science app is open source, and people have gone through the code and tweaked it for speed. More info is available here, in the Wiki. Given that the enhanced application is going to be coming out Real Soon Now (tm) and it will make the current optimized applications obsolete, I'd suggest running the standard client for the time being. Once the new application is out and the optimizers have had a crack at it, it may be worth moving to an optimized application for your OS and hardware. Happy crunching, Brian |
|
39)
Message boards :
Number crunching :
recieving NO CREDIT
(Message 213480)
Posted 14 Dec 2005 by Landroval
Post: edit: Bad info, deleted. Move along, nothing to see here.... |
|
40)
Message boards :
Number crunching :
Result was reported too late to validate
(Message 213152)
Posted 13 Dec 2005 by Landroval
Post: If, because you have a large cache, you consistently return your results after the first 3 are validated, then you are wasting your valuable computing power. Please spare me. I have no interest in getting into the "the 4th unit is wasted effort" debate. |
Previous 20 · Next 20
©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.