Posts by kashi

log in
1) Message boards : News : linking citizen science grid to SubsetSum@Home account now available (Message 646)
Posted 22 days ago by kashi
After they have linked, contributors to SubsetSum or DNA@Home without Wildlfe@Home credit are unable to post on the Citizen Science Grid forums until you successfully transfer their SubsetSum and/or DNA@Home credit.

This is due to an anti-spam forum setting common to many BOINC projects that will not allow anyone to post unless they have credit in that project.
2) Message boards : News : Another application update (0.11) (Message 283)
Posted 842 days ago by kashi
A greater incidence of tasks running in high priority mode is caused by the changes to work cache settings in BOINC 7.0.xx versions. Whether this is a side effect or intentional I do not know.

It commonly happens on projects with a shorter deadline when the Minimum work buffer is set to 1 day or higher. Tasks with a 2 day deadline will even go into high priority mode with a Minimum work buffer of 0.8 days. The only thing you can do to prevent it is use a smaller Minimum work buffer. Just need to keep reducing Minimum work buffer until high priority mode ceases. You could increase Max additional work buffer to compensate.

A small Minimum work buffer may cause work fetch issues with some projects with small number of tasks allowed in cache at once and/or intermittent work availability but that is another issue.

With earlier BOINC versions it should sort itself out over time as Conan said. However with BOINC 7.0.xx versions when you have a mixture of shorter and longer deadline projects then you sometimes need to fiddle with buffer settings to prevent the shorter deadline project tasks constantly going into high priority mode.
3) Message boards : News : Another application update (0.11) (Message 247)
Posted 845 days ago by kashi
I completed 35 v0.11 tasks, 29 are Validate error for me and wingmen, 6 are pending.

The v0.11 tasks take about 20% longer to complete on my computer than previous version tasks I have processed.
4) Message boards : News : Added more workunits (Message 122)
Posted 854 days ago by kashi
Yes and the validator is also still validating work units on a single task replication. 4 of my 5 Validate errors in this recent batch validated for my wingmen without an additional task being sent out.
5) Message boards : News : Added more workunits (Message 120)
Posted 854 days ago by kashi
Yes, I just had a look in BOINC Manager Event Log. The 5 "Validate error" 0.05 tasks on my computer were all suspended and restarted. When they completed they failed to upload correctly because invalid signature was reported. Here are completion/upload Event log entries for one of these error tasks:

30/04/2012 5:14:20 PM | SubsetSum@Home | Computation for task subset_sum_44_22_652372583280_0 finished
30/04/2012 5:14:20 PM | World Community Grid | Restarting task cfsw_0506_00506590_0 using cfsw version 605 in slot 3
30/04/2012 5:14:22 PM | SubsetSum@Home | Started upload of subset_sum_44_22_652372583280_0_0
30/04/2012 5:14:25 PM | SubsetSum@Home | [error] Error reported by file upload server: invalid signature
30/04/2012 5:14:25 PM | SubsetSum@Home | Giving up on upload of subset_sum_44_22_652372583280_0_0: permanent upload error
30/04/2012 5:14:29 PM | SubsetSum@Home | Sending scheduler request: To report completed tasks.
30/04/2012 5:14:29 PM | SubsetSum@Home | Reporting 1 completed tasks, not requesting new tasks
30/04/2012 5:14:31 PM | SubsetSum@Home | Scheduler request completed

Don't know if valid tasks also give those upload error reports the same as that. Just thought I'd mention it in case it relates to the checkpointing failure.
6) Message boards : News : Added more workunits (Message 115)
Posted 854 days ago by kashi
Completed 9 tasks, 4 validated, 4 Validate error, 1 pending which will probably error. All the error ones were restarted after suspending the project, stopping BOINC and rebooting for a MS update.
7) Message boards : News : Added more workunits (Message 113)
Posted 854 days ago by kashi
I just downloaded a batch of tasks and the 0.05 application downloaded too and is being used. Win 7 64.
8) Message boards : News : Updated applications to 0.05 (Message 109)
Posted 856 days ago by kashi
Yes my BOINC didn't download the new 0.05 application either. I didn't even notice until I read these posts a few minutes ago. I just had a look and 0.05 is in the download area, so don't know why it didn't download.

DCF continued to reduce and is now down to 1.4097 after completing 42 tasks so flops is about right I think.

Had 2 tasks validate with my wingman being invalid. Perhaps this is part of validation issues you were trying to fix with 0.05.

wuid=1978
wuid=1979
9) Message boards : News : Updated applications to 0.04 (Message 94)
Posted 859 days ago by kashi
Yeah, we probably won't have decent information about the credit until the creditnew system really kicks in. But are the new estimates for flops more realistic? I think before they were off by at least a factor of 15.

You mean for the "Remaining (estimated)" time shown in BOINC Manager? I don't know, when they first downloaded the last batch were showing 12 hours to complete and after I had finished them my DCF was 11.38. It was 13.72 after the first 0.02 batch I did that had the low estimated runtime. DCF drops slowly, so I'd probably need to start on a fresh attachment or do a fair few more than 12 tasks to get a proper idea of Remaining time with the new flops estimate. DCF should continue to decrease until it is close to 1 if it is correct.

The only reason I used WCG as a comparison was it is the only CreditNew project I currently contribute to other than this one. If some DNA@Home tasks get released sometime soon I could also compare to that.

A fixed credit project such as Docking@Home currently grants about 55 cr hr on this computer. I don't know about others, that's the only fixed credit CPU project I have contributed to recently.
10) Message boards : News : Updated applications to 0.04 (Message 90)
Posted 859 days ago by kashi
Travis requested feedback on credit so I gave it. "so let me know if this is a better estimate, and how their awarded credit is."

I don't chase high credit all the time or I would not have contributed most of my crunching time for the last 6 months to WCG, except for short periods when there was a team assault on a non WCG project.

The credit I received here can be clearly seen. The problem with commenting on credit rate of small batches of tasks is that it takes a while before the CreditNew rate stabilises and even then it may vary from time to time on some projects, even when the size/runtime of the tasks remains the same.

This variance has been previously experienced on DNA@Home. I have noticed it recently on WCG CFSW tasks also where the claimed/granted credit rate on my computer has dropped by 10% this evening to about 30 cr hr. These are single replication tasks so granted is not affected directly by low or high wingman claims but follows the daily variations of CreditNew calculations.
11) Message boards : News : Updated applications to 0.04 (Message 87)
Posted 859 days ago by kashi
One of my invalid tasks was subsequently validated on only a single replacement replication with no matching wingman. By that I mean four tasks were sent out and only one was validated. Seems a bit wonky.

wuid=958
12) Message boards : News : Updated applications to 0.04 (Message 86)
Posted 859 days ago by kashi
Yes they remain invalid after their replacement resends have validated.

Credit rate of valid tasks didn't alter on my computer, the same or perhaps a bit lower than before on average. If my Windows 7 tasks are paired with fast Linux wingmen then credit is very low, paired with a similar runtime wingmen then moderately low, paired with long runtime wingmen then slightly lower than WCG.

The 2 Linux tasks I did were both invalid but judging from the Linux wingmen of my Windows 7 tasks, credit on Linux remains low also.

Doesn't CreditNew scale credit according to a formula that includes the host benchmark and average performance? If so changing the tasks' GFLOPS value would only have a temporary effect if any. I could have easily misunderstood how it works though, I know on DNA@Home credit rate often used to vary daily. I think if the proportion of faster Linux computers is higher then average credit will get scaled down. This used to work to my advantage on DNA@Home where my Linux tasks finished much quicker than those on Windows hosts and there were relatively few faster Linux hosts attached to the project.
13) Message boards : News : Updated applications to 0.04 (Message 82)
Posted 860 days ago by kashi
All tasks so far "Validate error" here on both Windows and Linux.
14) Message boards : News : Updated applications to 0.04 (Message 79)
Posted 860 days ago by kashi
Now works on Dotsch/UX Linux thank you.

So far doesn't seem to be much faster than Windows, only a little, I must have misinterpreted the statistics I looked at from the first batch. Still useful for me to have the Linux option when I'm also contributing to other projects in my Linux VM. And good for others who prefer to use Linux all the time and have an older Linux version.
15) Message boards : Number Crunching : stats update frequency? (Message 78)
Posted 860 days ago by kashi
I think it is policy for BOINCstats to not include a new project's stats until this has been requested/approved by the project's administrator. This is probably the meaning of Crystal Pellet's question.

"Project seems to be in early stages of development, no stats exports, no permission given." [BOINCstats] Willy 14 Apr 2012.

Unless I have misinterpreted the meaning of your smiley and you're already aware of that, in which case ignore my explanation.:)

Thanks for updating the stats more frequently.
16) Message boards : News : validator up and running (Message 66)
Posted 860 days ago by kashi
Thanks for the explanation. I understand now, the tasks with the updated values are not available yet. Have to finish off the others first. Be good to get some adjusted ones, last tasks I just reported all have very low credit rate, one was 17.2 cr hr.
17) Message boards : News : validator up and running (Message 64)
Posted 860 days ago by kashi
I think I should be able to fix this, I just need to take the code that I used to statically compile dna@home for linux and use it with this. Might need to wait until tomorrow when i get back to my office though.


That would be excellent if you could do that, thank you. It will possibly affect some others too like at DNA@Home, so they would appreciate it also.

Edit: The 4 tasks that I had pending from my first batch have now been marked inconclusive and another replication sent out. I just downloaded another 12, most are replication _2, but newly released ones still have the same Estimated computation size of 1000 GFLOPS as the first batch I did. Not sure if that relates to the estimated FLOPS value you said you were updating.
18) Message boards : News : validator up and running (Message 61)
Posted 860 days ago by kashi
OK, I'll try a few more, got my first Computing for Sustainable Water badge now so my WCG cache can wait a bit.

Will do them on Windows again because Dotch/UX not working here. Errors immediately with "/usr/lib/libstdc++.so.6: version `GLIBCXX_3.4.11' not found" message. I'll have to start looking for a compact, fast Linux crunching distro. Dotsch/UX GLIBCXX version is often too old now to work with recent Linux project applications unless they are compiled to have greater compatibility with different Linux versions.

As I suspected, a quick look at the statistics page shows that it seems Linux is currently a fair bit faster than Windows.
19) Message boards : News : validator up and running (Message 59)
Posted 860 days ago by kashi
I did a dozen on Win 7 64. 8 have validated. 4 are pending validation although wingman has returned result. Wingman only reported recently on 3 of these so not sure if it is just validator delay.
wuid=133
wuid=134
wuid=135
wuid=138

Estimated time of this first batch was much too low. My duration correction factor is now 13.72.

Credit not so hot, 5 tasks with wingmen at similar runtime granted 30.8 cr hr. Currently World Community Grid is granting 33 cr hr on this computer. 3 tasks with Linux wingman at lower CPU time granted 21.7 cr hr, which is poor.

Hard to tell for sure though with early tasks under CreditNew, I think you have to complete 10 tasks successfully before host credit scaling is applied. You'll get more useful feedback from someone who has completed a larger number of tasks.

I've currently got a large cache of work for another project, as soon as that is completed in a day or so I'll do some more if there is still work here. I'll give Linux 64 a go now on a VM core to see if Dotsch/UX Linux works and if I can get a speedier runtime than Windows 64.




Main page · Your account · Message boards


Copyright © 2014 Travis Desell, Tom O'Neil and the University of North Dakota