Clients can't get work units!??

[spectra]

Several of my clients are no longer able to connect to the server and get a work unit. They return this statement:

"Java.Net.ConnectionException: Connection Refused: Connection
No worries, server is probably being restarted."

8 of 14 cannot connect. Another seems to be working but is not getting any credit on daily work.

Anybody? Class? :)

[adream]

i think miguel must be away or busy cos as far as i can tell the server has been down for 48 hrs now...

ho hum

regards

adream

:elephant:

[safemode]

apparently not everyone is effected, just almost everyone, since some teams still have stats.

[adream]

yes i think your right, i have 4 clients still ticking over fine, its just the ones that have started up over the last couple of days that havent been getting through

oh well ...

:elephant:

[safemode]

this might be explained by the client continuing to work on the genetic code it already had at the time of the server distribution failure. Because this distribution is done by separate sub-programs, one sends out genetic codes to clients, the return is handled by a different program likely, or at least module. Something has happened to sending the units, but recieving is fine, so all clients that have been working on units continue to (this leads to highly different genetic codes since no cross breeding is occuring) and can continue to send back to the main server.


Basically this is the behavior you'd get if you were running the client with the net connection off after you first start it up and download a genetic code from the server.

We're all screwed because we started the client up after the malfunction.

[spectra]

Well phooey. Ran quite well for about 24-36 hours. Now it's kaput again.

Safemode:

When the stats aren't being updated, is it your impression that we are still doing useful work or are we running in place so to speak? In otherwords are clients still accumulating points or is the power going to waste?

[safemode]

hard to say. The scripts used to generate the html pages might have borked, in which case we are still doing useful work. The program responsible for handling info from our clients may also be hosed, in which case we should see no ability to connect to the server, in which case we're basically running like we're offline. In this case when we are able to reconnect to the server, our client will upload all data acquired since the last time we connected. If the server is up, but the logging (perhaps database died) is borked , then i dont know....we could be losing data.


The only way we'd actually not be generating stats would be if the database died but the rest of the program didn't and it had no way of saving the data it's still recieving. Every other situation should still save our stats.

[prokaryote]

we're still doing useful work, we're just like isolated islands evolving our circuits with out any "cross-pollination" with nearby islands. That means that each client will hit a plateau of improvement and just depend upon random mutation to improve the circuit. This will take quite some time before a random mutation is better than the plateau'd circuits. Not losing any data either since the only data that counts are the current "individuals" in a generation. The "parents" essentially die off. It's the last surviving generation that contains the valuable data. The other part is the generation count. The number of circuits should = the #generations times the number of individuals (circuits) in each generation.

At least that's my take on how this particular genetic algorithm works.

[HitchHiker]

It would be nice to know if one can at least submit results, otherwise your hard work is lost with the next restart/power down. This may take just 5 minutes or another week but sooner or later it is going to happen.
This is something that has been bothering me for some time now as upon each restart I've done now my Mac had to to several hours of work once more to get where it was before the reboot. Connect interval is set to the standard 1 hour, so I would expect not to lose more than an hours work.
In these cases, it would be nice if you would be able to force a connection to submit results.

[HitchHiker]

It turns out there's a simple way to find out if you can submit results. At least it seems so.
Daily stats for today (18 jan 2004) are 0 for everybody. My best individual (in the client) at the moment is much worse than I'm am generating for many hours now. This suggest that this is updated (or fetched?) only on contact.
Both observations suggest to me that no circuits can be submitted, so everybody is completely on his/her own. It's lonely at the top, but also lonely on my island. :cry:

[michaelgarvie]

Server went down due to a general power cut in the lab on friday night I think. Repairs I guess.

However Prokaryote's take on the evolutionary algorithm is correct. If you haven't switched off your clients they've been evolving on their own and whenever the server's back they submit their best results. In fact the current best circuit at this moment was submitted as soon as the server started so it must have been found by lowlands_3 over the weekend. So no useful work has been lost.