Seti@Home optimized science apps and information
Optimized Seti@Home apps => Discussion Forum => Topic started by: areese on 18 May 2007, 02:27:39 pm
-
Since the huge outage last week all of my optimized apps cannot get new work. Any machines I have running the standard app work just fine. I even tried a full reinstall on one machine with the optimized app and still nothing. Anyone have any ideas?
Thanks
-
Please see http://setiathome.berkeley.edu/forum_thread.php?id=39531&nowrap=true#570201 for a workaround.
HTH,
Simon.
-
Simon....I did the disable app info bit and my hosts were then able to download all the work they wanted. 'Til either they reached their daily quota, or got a short WU and went into EDF and stopped looking for more work. Put the app info file back so I can keep crunching at full speed, but I suspect this will muck up comms with Seti again. May have to go through this once a day or so until they get things fixed again.
Have you any insight as to how running the optimized app could be hindering comms with the server? Did they do some code mods coinciding with the crash and new server migration that are taking exceptiion to the requests the optimized apps make vs what the stock app makes?
'Tis a puzzlement.
-
... Did they do some code mods coinciding with the crash and new server migration that are taking exceptiion to the requests the optimized apps make vs what the stock app makes?
If i read the whole tech news and blogs at the s@h board correctly, then i assume it's due to the fact that they updated their boinc server software....
If you ask me ... a "very" good idea while having major issues from clients connecting...
If you need answers on who to blame for this one and some other major issues like "ghost WUs" ( not the ones atm but the previous ones) ... send a pm if it's not allready clear who to blame.
-
I can't do that disable app info thing on my Pentium D because I'm using the 64 bit BOINC. It won't get work units without it. :'( Guess I'll have to switch it back to 32 bit.
I got my Xeon X3220 going with openSuse 10.2 and am using the 32 bit BOINC and it got work right away. :) Switched right to the 32 bit 2.2B, so I'm starting to get back on track.
-
Did they do some code mods coinciding with the crash and new server migration that are taking exceptiion to the requests the optimized apps make vs what the stock app makes?
'Tis a puzzlement.
They wouldn't do that on purpose, would they? This would hurt anybody using anonymous platform, who needs it for 64 bit or any platform not supported, not just optimizers.
-
Hi Pam,
I'd think it wasn't on purpose, no. However, it came at a very inopportune time; from what I read, Matt made some changes to the code, then there were compilation problems, and Matt had to leave before they got fully fixed.
So in any case, it's never a good idea to do any sort of code changes during an outage, which is what Crunch3r wanted to express, I believe. Usually leads to what we have now, total havoc.
Regards,
Simon.
-
Simon....There is some discussion going on in the Seti NC Ghost WU thread about rmoving the app info xml file and finding that the host is still running optimized apps. At what point would the host revert back to the stock app?
If you restart Boinc with the xml file in place, and then remove it to allow comms with the server, will all the WUs in the cache then continue to run optimized?
-
Simon....There is some discussion going on in the Seti NC Ghost WU thread about rmoving the app info xml file and finding that the host is still running optimized apps. At what point would the host revert back to the stock app?
That's certainly unclear. If the core client deletes the optimized app because it's no longer linked to anything, then it has to download the stock app. But as little as having one finished and uploaded WU left to report means the optimized app is still linked.
When I did the workaround changes on SETI Beta, the stock app was not downloaded and my personal build continued to run. There were still unreported results for those cases. But when I did the workaround for Main the stock app was downloaded and put to use immediately, I hadn't done any work on Main recently so there was nothing to report.
It seems likely that a queue of work might prevent ever switching to the stock app, every request to the Scheduler would have some linkage to work on the host. Whether that departure from the design goals is totally in the BOINC core client or part of recent changes server side, I don't know.
If you restart Boinc with the xml file in place, and then remove it to allow comms with the server, will all the WUs in the cache then continue to run optimized?
Yes, because the BOINC core client only looks for app_info.xml files at startup. But that's not what you want.
After any change of app_info.xml you have to restart BOINC for the change to take effect. For minimum crunch time loss, making the change while BOINC is running, then doing an exit followed immediately by restart is a good plan.
Joe
-
Joe.....
This may help explain why, when I followed your advice in Seti Beta, and removed the app info file, it was posted that my rigs were still reporting all the work as being crunched with optimized apps. A few days ago, I set all my rigs to no new work for Beta, and will crunch all as is and delete the optimized app file totally. This may take a while, seems that my quad rig has latched onto crunching Astropules WUs for 60-70 hours at a time. Ouch.
-
Pam, on my 64bit I just renamed App_info to Appp_info and was able to download fine, like 300 units. When I renamed it back the speed came back.
-
Folks,
I'd like to direct your attention to this thread by Carsten (http://setiathome.berkeley.edu/forum_thread.php?id=39636). He outlines a method similar to the Richard's workaround, but his does not require any manual intervention after installation (until a new stock application version comes along).
His post is titled "using optimized apps without an app_info.xml" and seems like it should work for anyone who is not attached to SETI Main and SETI Beta at the same time.
Reason being, to run both on the same host needs a "faked" higher version number to work for Main, which makes Carsten's method unworkable. Of course, that's an assumption, but I'll be able to prove or disprove it in a bit (will try Carsten's method on a number of different hosts to see what happens).
Regards,
Simon.
-
Simon, so far the simple work-around that has been working for me is to simply rename the app info file and restart Boinc. Period.
It clears up the comms problem with Seti Main (still get http server errors trying to report Beta, but I think that's been seen as a separate problem) and all the WUs continue to crunch using the optimized app. It has continued to work even after some of the rigs have rebooted. It seems that unless the host runs totally out of WUs or Boinc decides to update the stock app, the link to the optimized app is maintained even though the app info file is not present and Boinc does not on start the 'anonymous platform' when starting up.
This has worked on all of my rigs since yesterday, all 32 bit, 2 xp, the rest win2k.
I also changed my cache preferenc from 10 days to 4 days and this did not cause any problems.
-
Simon, I made the last post before reading the Carsten posts and had another thought...
Could the reason this is working for me have something due to the fact that the app info files I am using contain references to 512, 513, 514, 515, and 517?
-
Simon
I was waiting for the Server side issue to be resolved so I removed the app_info.xml on all my devices and all my devices are running 515 workunits. I am not part of the beta SETI so to run the chicken apps DO I remove all the references like below that refer to the following version numbers 512,513,514 and 517 then put the edited file into place till server side issue is resolved ? An example of some of the lines to be removed would be as follows
<app_version>
<app_name>setiathome_enhanced</app_name>
<version_num>512</version_num>
<file_ref>
<file_name>KWSN_2.2B_SSE3-C2_Ben-Joe.exe</file_name>
<main_program/>
</file_ref>
This will be diffrent dependent on which Chicken App you use.
Please clarify
Nutsalot
-
Ni, er, yes, Sir Nutsalot,
that's what I did and it worked fine for me - having just the "515" entry corresponding to the currently used stock app version in app_info.xml.
So the procecure is - IF your app_info.xml file previously had any HIGHER version that 515 - either run your cache dry OR detach/reattach your host (I kept my host ID last time, so that works even better because all Ghost WUs get redistributed).
Then, remove app_info.xml, stop and start BOINC. Let it grab the stock app and a few WUs and start crunching the first WU (to make sure everything's there).
Final step: put in an edited app_info.xml file and corresponding optimized app. Stop and restart BOINC, verify it's crunching with the optimized app. Stop BOINC again, remove app_info.xml.
Lean back, have a coffee. You'll need it after all that ;)
ALTERNATIVELY if you had any WUs left and if you previously had an app_info.xml file with versions only up to 5.15:
simply rename or remove app_info.xml, stop and re-start BOINC. You're done.
HTH,
Simon.
-
Simon
Tried it and all is fine if you follow the intructions and understand the caveats.
Thanks for posting the link to the other thread. This is a better workaround for me noting that i currently don't participate in the SETI Beta programe yet.
Thanks Nutsalot.
-
Ni, er, yes, Sir Nutsalot,
that's what I did and it worked fine for me - having just the "515" entry corresponding to the currently used stock app version in app_info.xml.
...
But there's no need to trim out the other <app_version> entries. Once you remove the app_info.xml file the Scheduler will tell the BOINC core client which version to use, and so long as 515 is available there's no problem. The others will have no effect other than making the client_state.xml file marginally larger.
Edit: The conflict with running Beta is also gone when not using app_info.xml for the same reason; the Scheduler tells the core client which version to run.[/edit]
The real issue with running with no app_info.xml is only that if the server problems or other factors run the queue dry at some point there's a possibility that the stock app will be automatically downloaded. I had a case yesterday where the last few WUs were all noisy, so when that host contacted the Scheduler it was reporting those as well as asking for new work. The contact succeeded so the reporting was complete, but the feeder problem and/or real work shortage caused a "No work" reply. The subsequent request which got new WUs also downloaded the stock app with which to crunch them. Because I'm on dial-up and always watching when I allow BOINC to connect, it was simple to switch back to 2.2B after the downloads completed and then off app_info.xml again. For someone with an always-on connection it could be confusing, I'm not sure what's the best way of keeping track.
Joe
-
Well, I think the last vesitges of Setiability just went into the tank 'till morning. Cricket has taken a dive, no respnse from the servers, and I doubt that anybody is there resetting things at midnight..
Wish I could just go to sleep and have this Seti nighmare over.
-
Hi.
I did my own work-around which is slightly different to others...
I had run out of SETI work, but had about 7 ghost WUs.
I detached from SETI after first backing-up my statistics_setiathome.berkeley.edu file and the contents of the projects\setiathome.berkeley.edu folder.
I then re-attached to SETI which downloaded 4 WUs (DCF had reverted to 1). BOINC Manager also downloaded the stock science app. I suspended crunching before 1 minute had elapsed so no state file had been written.
I then exited BOINC Manager and copied the 3 files from the opp app to SETI folder. Re-started BOINC, set No new tasks, and started crunching optimized.
I also started, and suspended after 1 minute, 3 other SETI tasks, and confirmed that \slots\ folders had the usual 6 files.
Then deleted app_info to the Recycle Bin (I think I stopped crunching first). I also edited client_state while BOINC was closed to make DCF 0.543210 which is a good figure for me. Enabled New Tasks and downloaded 5 more WUs. app_info is still in the Recycle Bin.
I have not tried this yet but suspect that making the optimized science app files read-only would prevent them being deleted if work runs out. If this works it would be a simpler work-around. Any thoughts? Keith