"Empty work server assignment" - Linux GPU's

[billford]

It would seem we're beta testing new AS code…

https://folding.stanford.edu/home/issue ... rver-code/

[bruce]

It would seem we're beta testing new AS code…

https://folding.stanford.edu/home/issue ... rver-code/

Yes, we are.

The Assignment Servers are Real-Time systems so it's virtually impossible to beta test them -- they can't be exposed only to a limited group of volunteers. After the pre-beta testing has been completed successfully, the code must be exposed to EVERYBODY.

There's one redeeming feature, however. Under normal circumstances, there are two or more servers running the relased code, providing a backup. The new code is never installed on both the primary and backup server at the same time. If a problem is found, a message is issued (such as "Empty work server assignment") and your client notes that a failure has occurred and reverts to the other AS which is still running the released code. It successfully provides you an assignment and all that has been lost is a few elapsed seconds which is meaningless since the default value of next-unit-percentage=99. Development can proceed to update the new code to fix the empty assignment bug and beta testing can continue until the observed problems are fixed. Depending on the nature of the observed bugs, the new code may be tested continuously or on-and-off so as to minimize lost time, but as I said earlier, exposing it to the general public cannot be avoided.

Since the "Empty work server assignment" bug is a soft failure (a valid assignment is promptly issued) it's a lower priority bug that one which, say, assigns a WU which cannot be processed by your system.

[Sn1ken]

Thank you so much for a detailed explanation

[billford]

Yes, we are.

Thank you for the explanation, I appreciate it.

However, when version 0.9.2 is rolled out, could Dr Pande perhaps be persuaded to let us know in advance?

I realise this would be a departure from established norms, but if nothing else it might result in a more helpful and structured reporting by donors of problems rather than everyone randomly scrabbling around trying to find out what the hell has gone wrong this time.

[bruce]

The AS code has been rolled back to the released version so there should be no more problems until a new version is ready to test. It's difficult to predict when another version will be ready to test ... or how long that version might be tested before it's rolled back (assuming the worst). I don't think anybody is going to be updating the blog in real time, but I'll pass on your request.

The PG does attempt to minimize problems that you encounter and they will continue to exercise their best efforts in that regard. Perhaps the next version to be tested will be bug-free; perhaps not. (Some people are always optimistic and others are always pessimistic.) Personally, I would like to have seen the blog post precede the testing ... but that didn't happen. Nevertheless, since there's no blog post saying that testing has been temporarily suspended, it doesn't seem logical to repeat the latest post at some time in the future.

[VijayPande]

This is a good question. We're constantly improving the FAH infrastructure and the AS system (it's now multiple machines, so not a "single" AS) has been updated gradually over the last few months in small steps. My original thinking was that announcing each step wouldn't be helpful since we'd be saying there's changes pretty frequently. However, I think the posts here have changed my mind. I'll ask Joe to at least post updates to system infrastructure in the forum so donors have a better idea of changes on the backend.

[billford]

Thank you Dr Pande, I'm happy with that.