Message boards :
Number crunching :
How accurate are predicted completion times for various projects?
Message board moderation
Author | Message |
---|---|
Benher Send message Joined: 25 Jul 99 Posts: 517 Credit: 465,152 RAC: 0 |
How close does each project's WUs estimated completion time come to actual time? For various types of hosts? Kind of an important number as the servers use it to determine how many WUs to send your host. By way of comparison, try checking a simple math equation. On any one of your hosts that is doing multiple projects... 1. Write down the estimated completion time of any project's unstarted WU (from the gui [work] tab, To completion colum). 2. Write down the actual average completion time. (you can look on the web on your "Your Account -> My computers -> [some host] -> Results" link. These are listed in seconds.) 3. Divide the estimated by the actual (probably easier converted to seconds). 4. Now do steps 1 to 3 for any other project. What numbers do you come up with? Project, Estimate, Actual avg, CPU type, Mhz speed, Hyperthread/etc. |
Adrian Send message Joined: 15 May 99 Posts: 8 Credit: 1,000,152 RAC: 0 |
Project: Seti@Home Estimate: 17,267 sec (4.80 hrs) Actual avg: 12,541 sec (3.48 hrs) Est/Actual: 1.38 CPU: AMD Athlon 2100+ 1.73GHz Project: Seti@Home Estimate: 27,052 sec (7.51 hrs) Actual avg: 17,242 sec (4.80 hrs) Est/Actual: 1.57 CPU: Intel P4 2.4C, Hyperthreaded Project: Seti@Home Estimate: 55,980 sec (15.55 hrs) Actual avg: 56,054 sec (15.57 hrs) Est/actual: 1.00 CPU: Intel Celeron 566MHz Use www.google.com for time conversion. Enter this into the searchfield: ## hours ## minutes in seconds |
Paul D. Buck Send message Joined: 19 Jul 00 Posts: 3898 Credit: 1,158,042 RAC: 0 |
> What numbers do you come up with? > Project, Estimate, Actual avg, CPU type, Mhz speed, Hyperthread/etc. As best as I can come up with for my 7 computers of 5 types it looks like this: Computer CPU Information Speed Cache HT? -------- -------------------- ------- ----- --- Xeon Xeon "Nocona" 3.4 GHz 2M Yes EQ-1/P4a P4 "Prescott" 3.2 GHz 1M Yes EQ-2/P4b P4 "Northwood" 3.0 GHz 512K Yes RAID Svr. P4 "Northwood" 2.8 GHz 512K No G5a G5 PowerPC 970 (2.2) 2.0 GHz 512K No Predictor@Home SETI@Home Einstein@Home ------------------ ------------------- ------------------- Computer Time Seconds Ratio Time Seconds Ratio Time Seconds Ratio -------- ---- ------- ----- ---- ------- ----- ---- ------- ----- Xeon 3:43 13,380 2.43 5.11 18,600 1.97 7:26 26,760 0.727 EQ-1 4:06 14,760 2.19 5:43 20,580 1.96 8:12 29,520 0.756 P4a 4:01 14,460 2.14 5:36 20.160 1.92 8:02 28,920 0.741 EQ-2 4:31 16.260 2.13 6:19 22,740 1.65 9:02 32,420 0.834 P4b 4:13 15,180 1.99 5:54 21,240 1.54 8:27 30,420 0.785 RAID Svr. 3:46 13,560 2.91 5:16 18,960 1.66 7:33 27,180 0.912 G5a 5:03 18,180 3.49 7:04 25,440 2.77 10:07 36,420 1.30 all of the predictions are high, which is better than being too low of course ... Worst predictions (overall) are for the PowerMac, which may simply be due to the current emphasis on the Windows versions. Not knowing for sure what the breakdowns are for the processor types I can only guess what the fraction is ... You can, of course, look at my computers through the account page, but the actual times are best seen in my tracking tables (the Xeon numbers are more limited of course as I only have had it running a couple of days now and have not posted the updates needed to the live site yet). Also of interest if you plan to tilt more at this windmill is the old benchmark study which contains benchmarks for my older HT and non-HT machines including BOINC's benchmark and a synthetic benchmark program ... enjoy ... :) |
Divide Overflow Send message Joined: 3 Apr 99 Posts: 365 Credit: 131,684 RAC: 0 |
You should also include the BOINC core client version number since the benchmarking changes from some versions to another and this will change your completion estimate times. BOINC 4.19 Project: Seti@Home Estimate: 17,609 sec (4 hours, 53.5 minutes) Actual avg.: 7,600 sec (2 hours, 11 minutes) Est / Actual: 2.32 CPU: Intel Pentum M 1.8 (2MB L2 cache) |
MikeSW17 Send message Joined: 3 Apr 99 Posts: 1603 Credit: 2,700,523 RAC: 0 |
> > > Use www.google.com for time conversion. Enter this into the searchfield: ## > hours ## minutes in seconds > > LOL! Be lazy.... 'xx hours nn minutes in seconds' = 30 keystrokes 'xx*3600[M+]Cnn*60[M+][MR]' = 17 keystrokes |
Benher Send message Joined: 25 Jul 99 Posts: 517 Credit: 465,152 RAC: 0 |
Several systems so far, but only Paul's has multiple projects. - but really the average should be weighted for Qty of cpu type participating. (example. If 70% of CPUs are Pentium IV then that should be factored into avg ratio) Seti average ratio so far: 1.79 Predictor avg: 2.47 Einstein avg: 0.87 Paul, do your G5s really estimate 17:37 to complete a predictor WU? |
Mike Send message Joined: 17 Feb 01 Posts: 34263 Credit: 79,922,639 RAC: 80 |
Hi AMD Athlon XP 2800+ @2,1GHZ WinXP SP1. Seti estimate 4.05 actual 3.4 Einstein estimate 5.46 actual 6.12 greetz Mike With each crime and every kindness we birth our future. |
Astro Send message Joined: 16 Apr 02 Posts: 8026 Credit: 600,015 RAC: 0 |
P4 1.8ghz, Winxpsp2, Boinc 4.27 Einstein estimated 13:00:01 actual 13:11:53 actual 13:08:18 PPAH Estimated 06:30:00 actual 05:05:10 actual 05:05:42 actual 04:54:53 actual 00:24:12 actual 05:01:53 Seti Estimated 09:04:33 actual 00:03:29 actual 05:15:31 Athlon XP2200, Winxpsp1, Boinc 4.27 PPAH Estimated 06:11:30 actual 02:28:57 Sorry, I just automatically sent in this morning work before checking the time. more later tony |
Saenger Send message Joined: 3 Apr 99 Posts: 2452 Credit: 33,281 RAC: 0 |
AMD Athlon XP 2200+, WIN XP SP1, Boinc4.25: (Benchmarks: 1664/2785) (Last results first) Seti estimate: 4:30 Real: 3:48 3:25 4:13 3:14 3:42 3:38 4:15 0:02 4:10 3:38 3:40 4:16 4:07 4:04 Predictor estimate: 3:20 Real: 1:27 1:14 1:36 1:15 1:41 1:14 1:14 1:30 1:16 1:11 1:34 1:13 3:01 Einstein: estimate: 6:40 Real: 7:32 10:01 7:52 10:02 7:50 7:47 9:53 9:44 7:52 9:51 7:49 7:50 9:60 7:44 |
Paul D. Buck Send message Joined: 19 Jul 00 Posts: 3898 Credit: 1,158,042 RAC: 0 |
> Paul, do your G5s really estimate 17:37 to complete a predictor WU? Where do you see 17:37? In the table I see it is 05:03 ... I just got some work in LHC@Home, though I am not sure that these will be usable as a metric in that the models are known unstable ... And I can see where my dat might be confusing, the times in the table are all the estimates, estimated time in clock form, the estimated time in seconds, then the ratio ... the average time used to get the ratio is from my tables on the web page I quoted ... |
W-K 666 Send message Joined: 18 May 99 Posts: 19109 Credit: 40,757,560 RAC: 67 |
Can I stick my oar in here, I'm new to Boinc but have processed Classic Seti since 18 May 1999. For those of you with HT P4's are you processing one or two units, My machine is slow a Dual P3 866MHz, (I'm on Hand me downs from my sons machines and it's fast enough for what I do). The predicted time is reasonable if I only process one unit but if I do two then the time is about 20% off. i.e. so far with mine prediction is 9:47, one unit processing varies from 9:20 to 10:40 and with two units going they take about 12hrs. Operating systems used win2000, win2003 server ent, winXP pro. |
Benher Send message Joined: 25 Jul 99 Posts: 517 Credit: 465,152 RAC: 0 |
> In the table I see it is 05:03 ... Sorry Paul - Was reversing the ratios. The two columns of times you list are predicted (identical hh:mm and seconds values), and the ratio is used to determine actual. Updated ratio Averages (unweighted): Seti@Home: 1.72 Einstein : 0.89 Predictor: 2.31 Without the Macintosh G5: Seti@Home: 1.63 Einstein : 0.86 Predictor: 2.16 |
Divide Overflow Send message Joined: 3 Apr 99 Posts: 365 Credit: 131,684 RAC: 0 |
I can add my home system to this little experiment: BOINC 4.19 Project: Einstein@Home Estimate: 20,528 sec (5 hours, 42 minutes) Actual avg.: 22,500 sec (6 hours, 15 minutes) Est / Actual: 0.91 CPU: AMD XP 2600+ |
Adrian Send message Joined: 15 May 99 Posts: 8 Credit: 1,000,152 RAC: 0 |
> You should also include the BOINC core client version number since the > benchmarking changes from some versions to another and this will change your > completion estimate times. I'm using BOINC 4.25 on all 3 machines I've listed in my previous post. |
Paul D. Buck Send message Joined: 19 Jul 00 Posts: 3898 Credit: 1,158,042 RAC: 0 |
> > In the table I see it is 05:03 ... > > Sorry Paul - Was reversing the ratios. The two columns of times you list are > predicted (identical hh:mm and seconds values), and the ratio is used to > determine actual. Yes. I thought I made that clear ... but probably not ... I have been doing REAL bad the last couple days so my thinking is none too clear. Actually the ratio was cvalculated FROM the actuals ... do you want the paper I scribbled my notes on? :) The other reason I referred to the page for the actuals is that page is updated on a weekly basis to my current actuals (if you want to track that). You can look at my computers and see the benchmark scores too ... And since I have 2 pairs of nearly identical machines you can see that there are differences between the two pairs ... The EQ-1/P4a are using A-Bit IC7-G MB with P4 3.2 GHz processors and the EQ-2/P4b are using Intel D875PBZ MB with P4 3.0 GHz processors. |
©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.