WARNING - unified installer may (currently) delete all work units

[Pynda]

Windows 7 64 bit
Intel E8400, Nvidia GF8800 GTX
Boinc 6.10.18

Since a lot of people seem to be having this problem I thought I would start a new thread that has a subject which is a little more clear. The best information I've been able to find so far can be found in this thread, though (the possible?) workaround is still not explained simply enough that I myself would be tempted to try it.

http://lunatics.kwsn.net/windows/help-i-guess.0.html

The problem comes about because the project doesn't have separate applications for the windows_x86_64 platform, so the work downloaded for stock uses an <alt_platform> mechanism to enable use of the windows_intelx86 applications. The <alt_platform> mechanism isn't available in app_info.xml so the work is deleted because you no longer have windows_intelx86 applications.

It is possible to put <platform>windows_intelx86</platform> entries in app_info.xml for each app_version, but that causes other problems under some circumstances. We've thought about doubling the number of app_version sections to have both platforms listed for each, but that's untested.

I think I'd edit BOINC's client_state.xml if I were making that transition. First I'd do the optimized install, then change the platform for each <result> listed in client_state.xml to windows_x86_64 before starting BOINC again. Since you are smart enough to make a backup, you might try this approach.


My own trouble is that I just did a fresh install of Win7 64 bit, installed Boinc and Seti, and got a whole buttload of work units (30+) because my preferences hadn't kicked in yet. And now I'm pretty sure these WUs are gone for good. I also wasn't smart enough to back them up before I installed Seti Optimized - allthough I did read the FAQ first and saw no mention of this particular horror.

- Is there any way to cancel these lost work units? Let alone having all of them go unfinished for the next month, I'm actually afraid I might be kicked from the program.

- If anyone can do a step by step explanation for dummies on how to currently install optimized (64 bit is the problem area?), I'm sure others here might appreciate it.

- Shouldn't the current installer be withdrawn, or at least a very strong warning placed next to the download and in the install FAQ until these issues are resolved?

Here's my app_info.xml in case it will help with anything. It looks to me like the astropulse units are still listed (the WU themselves may also still be around?), but that all the CUDA files have been dumped and are not listed.

--------------------------------------------------------------------------------------------------------------------
E:\BOINC\Program Data\BOINC\projects\setiathome.berkely.edu

- <app_info>
- <app>
  <name>setiathome_enhanced</name>
  </app>
- <file_info>
  <name>AK_v8b_win_x64_SSE41.exe</name>
  <executable />
  </file_info>
- <app_version>
  <app_name>setiathome_enhanced</app_name>
  <version_num>528</version_num>
- <file_ref>
  <file_name>AK_v8b_win_x64_SSE41.exe</file_name>
  <main_program />
  </file_ref>
  </app_version>
- <app_version>
  <app_name>setiathome_enhanced</app_name>
  <version_num>603</version_num>
- <file_ref>
  <file_name>AK_v8b_win_x64_SSE41.exe</file_name>
  <main_program />
  </file_ref>
  </app_version>
- <app>
  <name>astropulse_v5</name>
  </app>
- <file_info>
  <name>ap_5.03r112_SSE3.exe</name>
  <executable />
  </file_info>
- <app_version>
  <app_name>astropulse_v5</app_name>
  <version_num>503</version_num>
- <file_ref>
  <file_name>ap_5.03r112_SSE3.exe</file_name>
  <main_program />
  </file_ref>
  </app_version>
- <app>
  <name>astropulse_v505</name>
  </app>
- <file_info>
  <name>ap_5.05r168_SSE3.exe</name>
  <executable />
  </file_info>
- <app_version>
  <app_name>astropulse_v505</app_name>
  <version_num>505</version_num>
- <file_ref>
  <file_name>ap_5.05r168_SSE3.exe</file_name>
  <main_program />
  </file_ref>
  </app_version>
  </app_info>

   

[Pynda]

If it may be of any help, "Reset" ing the project did nothing to cancel my outstanding work units. Perhaps that command is based off information that needs to be in the app.xml file? And even  more disappointingly, "Detaching" from Seti@home didn't work either. I waited a couple of days and then rejoined, and there were all the same incomplete work units still outstanding on my results page. Guess there's nothing left to do but to let them time out.

[Jason G]

Having the same issues here.  Something has changed in recent version(s) of Boinc that has made it dump tasks when a new application is installed via the installer... Will be investigating a bit later in the week.

To release tasks to other users would require a full detach followed by reattach.  Unfortunately that resets back to stock applications, for which then boinc  immediately attempts to download tasks ... It appears these are being marked with *something* incompatible with the current installer version, so tasks are dumped (rather unceremoniously, creating 'ghost tasks'), and new ones downloaded.

Whether this probllem is isolated to Windows 7 x64, combined with current Boinc version, or that the issue applies to 32 bit OSes / apllications now also, is unclear.

For now, the best advice I can think of Is to run your cache down prior to installing the optimised applications.

[Jason G]

We beleive we have isolated the mechanism to blame, primarily that the project in fact does not actually have or distribute any stock Windows x64 applications, whch means stock tasks are marked with windows_intelx86 platform.

We will verify and incorporate in the next installer release, platform tags to match stock applications.  This should smooth new installation, but could potentially cause complications for existing optimised installations.

I have added the following warning to the Installer v0.2 release info page & Readme FAQ.

[WARNING - drain cache first, recommend detach/reattach as effective release of tasks prior to installation] Added 30th November 2009

Doing this before installation should minimise the number of  tasks that need to time out, by releasing them to other users, and restore the instalaltion to a 'known state'.

Jason

[k6xt]

Hope I am in the right thread for this. On a laptop Dell M6300, XP 64 bit, FX1600GPU 256MB, Core 2 Duo T9300, RAM 4GB. Running BOINC/SETI without optimised apps everything seems OK. I get CPU and GPU WU.

I flushed my cache, set SETI temporarily to "no new tasks". I installed using the Windows 64 bit unified installer, enabled new tasks. I get no WU downloads. My other PCs, all 32 bit versions of  XP or W7, are not having problems.
Only this msg is returned, and the same without "Update requested by user" if I leave the PC alone for a time:
1/22/2010 9:05:59 AM   SETI@home   update requested by user
1/22/2010 9:06:02 AM   SETI@home   Sending scheduler request: Requested by user.
1/22/2010 9:06:02 AM   SETI@home   Requesting new tasks for GPU
1/22/2010 9:06:07 AM   SETI@home   Scheduler request completed: got 0 new tasks
1/22/2010 9:06:07 AM   SETI@home   Message from server: No work sent
1/22/2010 9:06:07 AM   SETI@home   Message from server: No work is available for SETI@home Enhanced
1/22/2010 9:06:07 AM   SETI@home   Message from server: No work is available for Astropulse v5
1/22/2010 9:06:07 AM   SETI@home   Message from server: (reached daily quota of 10 tasks)
1/22/2010 9:06:07 AM   SETI@home   Message from server: (Project has no jobs available)

I thought to try the fix posted below. But of course I have no SETI <results> in client_state.xml.
Any hints what I should try next?
Thanks Art

Edit: According to SETI Server status there are plenty of WU available.
Edit2: BOINC is installed as a service.

[k6xt]

Its worse than I stated below. Every task this PC has attempted since December has errored out. Here's an example:
http://setiathome.berkeley.edu/result.php?resultid=1491230880

Can someone help me out? I have no idea what might be wrong, except for one little thing: CPU-Z says this PC is capable of SSE4.1. During Unified installer installation I checked that box. The result message above says SSE3.

Thanks
Art

[Raistmer]

stderr is pretty clear saying IMO:
After app init: total GPU memory 268107776    free GPU memory 127549184
Cuda error 'cudaMallocArray( &dev_gauss_dof_lcgf_cache' in file 'd:/BoincSeti_Prog/sinbad_repositories/LunaticsUnited/SETI_CUDA_MB_exp/client/cuda/cudaAcc_gaussfit.cu' in line 417 : out of memory.

Your GPU has not enough memory.

[k6xt]

stderr is pretty clear saying IMO:
After app init: total GPU memory 268107776    free GPU memory 127549184
Cuda error 'cudaMallocArray( &dev_gauss_dof_lcgf_cache' in file 'd:/BoincSeti_Prog/sinbad_repositories/LunaticsUnited/SETI_CUDA_MB_exp/client/cuda/cudaAcc_gaussfit.cu' in line 417 : out of memory.

Your GPU has not enough memory.

OK, it has 256MB. I thought that was enough? PC details here: http://setiathome.berkeley.edu/show_host_detail.php?hostid=5242895
This PC was running standard SETI apps until I installed the optimized apps using the KWSN unified installer.
And, now the PC has returned to not downloading any SETI work. No GPU, no CPU.

[Raistmer]

apparently it's not enough.
Try to revert to CUDA 2.2 instead of CUDA 2.3
New CUFFT requires more memory.
Try to disable some visual effects in Windows.
[And get rid from 195.xx driver]
P.S. see no valid results for that host at all so can't comment why it could work with stock CUDA version.

[k6xt]

Working my way through your advice. Now I have some completed WU using stock SETI app. For some reason all Iget is CUDA work. Rescheduler 1.9 doesn't seem to work, maybe because this PC uses XP 64 bit? Here's a (abysmally slow!) CUDA WU:
http://setiathome.berkeley.edu/result.php?resultid=1494165924

And this response from Reschedule:
---------------------------
Reschedule version 1.9
Time: 28-01-2010 10:37:47
User testing for a reschedule
CPU tasks: 0 (0 VLAR, 0 VHAR)
GPU tasks: 39 (6 VLAR, 16 VHAR)
Reschedule needed, found a VLAR
---------------------------
Reschedule version 1.9
Time: 28-01-2010 10:38:15
User forced a reschedule
Stopping BOINC service
BOINC service is stopped
Boinc applications
setiathome_enhanced 608 windows_intelx86 cuda
No SETI main application found
Starting BOINC service
BOINC service started
---------------------------

Maybe there is supposed to be another app? This install was done via BOINC, I did nothing except attach to project. I have no other non-optimized PC's to compare to. Seems best to make sure the stock app situation is correct before installing the KWSN optimized app again.

So the questions are two that I can think of:
Should Rescheduler be working with stock apps?
Should there be another app that BOINC should have installed?

BTW, re graphics usage on this PC, there is none other than wallpaper that run in background etc. Its a laptop that, most of the time, is only running BOINC. XP 64, no fancy effects.
Thanks
Art

[Claggy]

I don't think you're done any stock CPU work, so you haven't got the CPU app,
and so haven't got it mentioned in your client_state.xml,
get some CPU work from the project and try again, try suspending any other CPU projects first,
then check your Setiathome project preferences to see if you're allowing CPU work.

Claggy

[k6xt]

RIght-o. All the work I see from this PC in "valid" is GPU completions. I run a number of PCs all of which, except for this one, have CPU work. I don't know how to tell the project  to "send me some CPU work" as opposed to "send me some work". Is there a way to do that?

And, I wonder why the CPU app wasn't installed by BOINC to start with. This PC is a XP 64 bit which is the first 64 bit machine I've used, could that have something to do with it?

Thanks!

[Claggy]

As i said above, try suspending any other projects on that PC, Boinc should immediately start asking for CPU work,
you can tell by looking in the messages tab what type of work Boinc is asking for,
If it doesn't ask for CPU work, Check your Setiathome project preferences to see if you're allowing CPU work,
Make sure that your PC is set to the same venue as your setiathome preferences.

Claggy

[k6xt]

I did and it didn't on that PC. It immediately downloaded a bunch of only CUDA work. And I repeat I have several PCs this is the only one not getting lots of CPU work. It is a new PC with a hardly used XP x64 install. There's nothing on seti or boinc prefs on the PC that would prevent receiving CPU work that I know of but just in case I used the "Clear" button. Well I dont know, just let er go I guess. Probably turn SETI off on that PC because the GPU WU take forever using the stock app.
Thanks
Art
Edit: I don't understand "...same venue..." what is that? I just logged the PC onto my account at BOINC (BAM) no different than the others, BOINC did (or in this case didn't do) the rest.

[Claggy]

You can set your PC's to 4 different venues, 'Default', 'Home', 'School', or 'Work',
meaning you can have 4 different preferences for 4 different sets of PC's, eg, 1 day cache on default, 3 day on Home, 7 day on school, 10 day on work,
and you can set 4 different sets of project preferences too, e.g, no CPU at default, no GPU at home, neither at school, both at work,
and you could have different apps set to each venue too,
Note: you've got to have Boinc 6.10.19 or later for the CPU/GPU preferences to work properly.

Claggy