Results Processed with Optimized Apps Are Not Validating

[CElliott]

A significant fraction (I estimate at least one-half) of the results that have been processed by the KWSN_2.4V_***_MB.exe client and that I am returning to S@H are not validating.  The S@H server decides "No consensus" and sends out another WU for someone else to process.   In the vast majority of cases, the third WU does produce a consensus, and the WU is retired, but it can take a week or more.  About one WU per week is declared invalid.  I return about 110-150 WUs per day, so this is a lot of extra work for the server and society.  S@H and society are not helped in terms of energy consumption and global warming if use of the optimized clients causes one-third extra work.  Does anyone know why the WUs processed with the optimized clients are not validating initially?  Could the cause be that the optimization effort started with the 5.15 client rather than with the 5.28 cliient?   I predict that S@H is going to forbid use of the optimized clients if they continue to engender extra work and much larger WU database table sizes.  That would be a terrible waste of resources.  Could someone please look into this problem? 

[Jason G]

Hi CElliot, I don't represent either project in any way, though I may be able to add some insights.  I am sure Someone can add to my statements,  or correct me where necessary.

A significant fraction (I estimate at least one-half) of the results that have been processed by the KWSN_2.4V_***_MB.exe client and that I am returning to S@H are not validating.  The S@H server decides "No consensus" and sends out another WU for someone else to process.   In the vast majority of cases, the third WU does produce a consensus, and the WU is retired, but it can take a week or more.  About one WU per week is declared invalid. 

It is not normal for tasks to be routinely behaving in this manner. Your setup needs to be inspected in more detail. If a full one half of the results are going into "Checked but no Consensus", and many end up invalid, then I would check:
           - That you have the right 2.4V application for the CPU/system [Note: This site has no 2.4V builds, I guess you are talking about ones from Crunch3r's site? I use those, they validate fine for me, though going back to stock is certainly an option for someone having issues.  In rare cases some machines just will not run an optimised application correctly.]
           - That the 'Connected Every' setting in boinc preferences is set to something larger than zero. (this is a known issue whereby the boinc servers generate invalid results)
          - The regular Hardware/Drivers/Networking related issues
         -  Possible system overheating / Cooling Issues?
          - High Overclock?
         - Additional Vista Issues with Boinc (not the science app), On Windows Vista Boinc needs to be installed to somewhere other than the default 'Program Files (x86)' directory, if UAC is enabled.  These are Boinc compatibility issues as Boinc is not entirely Vista compatible, If I understand correctly these issues are hoped to be addressed with the release of Boinc 6.  Until then the abovementioned workaround (Installing to C:\BOINC instead) fixes it for most people.

I return about 110-150 WUs per day, so this is a lot of extra work for the server and society.  S@H and society are not helped in terms of energy consumption and global warming if use of the optimized clients causes one-third extra work.  Does anyone know why the WUs processed with the optimized clients are not validating initially?

Many possible issues as described above, including as you suggest possibly some incompatibilities between your system and the optimised application.  You don't mention what OS you are using, whether this is one machine or several, and whether they're identical, on the same network etc... Again, this validation behaviour you are observing is not normal.

Could the cause be that the optimization effort started with the 5.15 client rather than with the 5.28 cliient?

  Unlikely, since 5.28 'borrows' many of the optimisations made in the optimised applications - in effect descending from it...

I predict that S@H is going to forbid use of the optimized clients if they continue to engender extra work and much larger WU database table sizes.

Based on you having validation problems?  No, that you are having validation problems means only that...
From time to time I have seen the project take action on certain 'rogue machines' that have accumulated many results.  These have tended to be 'set and forget' machines far more often  running the 'stock' application than running an optimised one.  In my observation, people running an optimised application tend to keep a better eye on their machine.  This is probably not an observation of the opt. app, but an observation of the kind of person that prefers to run it.

That would be a terrible waste of resources.  Could someone please look into this problem?

Apart from the suggestions I've noted above, without more information  I can't really help further.  If you downloaded your 2.4V application from Crunch3r's site did you post to the forum there for help? you did realise the 2.4 applications here are not Vista compatible?

Jason

[CElliott]

Yes, these are Crunch3R's apps.  No, I did not post on his site.
I am fairly sure I have the correct apps installed on each machine.  They fault so fast when the S/W requires H/W that does not exist, that I think that is  not a problem.
Only two WUs in total have been declared totally invalid, but a huge number have required a third WU to achieve consensus.
I run WinXP, Win2K and Server 2003.  Vista is not an issue.
All the systems are overclocked, but many of them are the same systems I ran the S@H Beta project on (1/2 stock and 1/2 optimized apps) without problems.
Boinc is set on every machine so that the network is always available.

[Jason G]

Ahhh thanks for the extra detail. So my guess is that the higher efficiency of the optimised apps is pushing the OC's/Temps just a little harder than the stock/beta apps.  If you stress test with Prime95, for say 24 hours,  I suspect that similar stability issues may show [indicating a borderline OC].  If so, then simply backing off a little on the OC may improve matters. Alternatively, If the machine(s) pass(es) the Prime95 'torture test' but still generates the borderline results, maybe the rare compatibility issues and/or settings need to be looked at.

[Note: I'm not an OC'ing type guy, There are some very experienced in that willing to help/answer questions over in the number crunching forum at seti@home message boards]

Jason

[Gecko_R7]

Ahhh thanks for the extra detail. So my guess is that the higher efficiency of the optimised apps is pushing the OC's/Temps just a little harder than the stock/beta apps.  If you stress test with Prime95, for say 24 hours,  I suspect that similar stability issues may show [indicating a borderline OC].  If so, then simply backing off a little on the OC may improve matters. Alternatively, If the machine(s) pass(es) the Prime95 'torture test' but still generates the borderline results, maybe the rare compatibility issues and/or settings need to be looked at.

[Note: I'm not an OC'ing type guy, There are some very experienced in that willing to help/answer questions over in the number crunching forum at seti@home message boards]

Jason

Jason is spot-on.  It has been pretty clearly shown previously that Seti is VERY sensitive to OC stability.  While your system may appear "stable", there may be "just enough" underlying instability to generate sufficient computational errors when crunching that change the result accuracy threshold to "weakly similar" or "not similar".  This would lead to what your seeing.

There could be several possible tweaks that could fix this.  Increased cooling, lower clock, voltage adjustments, memory timing etc.
Like OCiing, you may have to play w/ it a little to find the sweet spot.

Msattler is our resident OC expert and may wish to chime in, or you can PM him on this board or Main.

Good luck!

[popandbob]

I ran the S@H Beta project on (1/2 stock and 1/2 optimized apps) without problems.

The way I read this is that you used optimized apps over on beta? The Beta project is for testing stock apps only plz!!