LHC@home

Not sure if this is a BOINC problem or LHC related :


From stderr.txt :

2004-10-02 00:30:10 [LHC@home] SCHEDULER_REPLY::parse(): bad first tag Content-type: text/plain
2004-10-02 00:30:10 [LHC@home] Can't parse scheduler reply

And from sched_reply.xml :

Content-type: text/plain


    Server can't open database
    3600
    


LHC@home

Problem seems to be this Content-type: text/plain

Same here, I also get multiple errors from the Website (too many connections, unable to connect to Database).

Seems they've hit their first real bottleneck on the Servers ;)
___________________________________________
<p>Scientific Network : 36200 MHz «» 8204 MB «» 815.0 GB </p>

2004-10-02 00:30:10 [LHC@home] SCHEDULER_REPLY::parse(): bad first tag Content-type: text/plain
2004-10-02 00:30:10 [LHC@home] Can't parse scheduler reply
And from sched_reply.xml :
=========

Well at least now I know I'm not the only one getting that Message ...

Been getting a lot of 2 & 3 Min tunescan WU's also, but just on one Computer, it could be the Computer I don't know, but it was running ok up until a few hours ago & then I had to do a reset & final uninstall BOINC & reinstall it to get it to run.

Naturally I got a fresh load of WU's so maybe it's the WU's also...???

I believe LHC better go back to beta. Seems it can't handle the bandwidth, ????

Seems they underestimated how many wu's we can run and how much bandwidth we can gobble up. ;)

look this maybe helpfull
http://www.google.de/search?hl=de&ie=UTF-8&q=Can%27t+parse+scheduler+reply&btnG=Google-Suche&meta=
feel free to visit www.guidowaldenmeier.de

> Seems they underestimated how many wu's we can run and how much bandwidth we
> can gobble up. ;)


Well, many (or all?) 8-hours WUs (sixtrack 4.46) run only 3 minutes - I doubt this is the normal behaviour.

edit:
This is the complete output btw., it ate the XML tags.
So the scheduler reply was absolutely correct and contained
a valid error message, it just shouldn't have had the
mime type in the header.


From stderr.txt :

2004-10-02 00:30:10 [LHC@home] SCHEDULER_REPLY::parse(): bad first tag Content-type: text/plain
2004-10-02 00:30:10 [LHC@home] Can't parse scheduler reply
2004-10-02 00:30:10 [LHC@home] Deferring communication with project for 1 minutes and 0 seconds

And from sched_reply.xml :

Content-type: text/plain

[scheduler_reply]
    [message priority="low"]Server can't open database[/message]
    [request_delay]3600[/request_delay]
    [project_is_down/]
[/scheduler_reply]
[scheduler_reply]
[project_name]LHC@home[/project_name]
[/scheduler_reply]

got the same problem:

> 2004-10-02 00:30:10 [LHC@home] SCHEDULER_REPLY::parse(): bad first tag
> Content-type: text/plain
> 2004-10-02 00:30:10 [LHC@home] Can't parse scheduler reply

I just re-installed Boinc by re-running Boinc update 4.13. Now Bionc says:

> No work from project
> Deferring communication with project for...

But it did upload my finished results now. and didnt do that before i reinstalled. Maybe the re-instal did help.

A definitive answer (at least, it's as definitive as anything is with BOINC)

When the server is under heavy load it can exceed the number of connections allowed to the database resulting in the message 'Server can't open database'. There is a problem with the server software that returns a malformed XML message in this circumstance that the client can't understand. The result is a 'Cannot parse scheduler reply' message displayed by the client. What it means is 'Server is a bit busy right now'. Reinstalling the BOINC software won't help directly since this is a server problem. Waiting a few minutes (while you re-install BOINC, perhaps) will allow some connections to clear, and things might then work.

The connection limit is configurable by the admins. Raising it can improve connection performance, but raising it too far can result in much reduced throuhput. Currently it's set at 100 connections unless Markku has changed it recently. It had been set at 400 connections prior, but this resulted in only 25% of the throughput!

IMO the whole BOINC database set up needs a good overhaul since these performance problems pop up at what seems to be quite low levels. As always, I suspect that the database performance may only be improved by the expenditure of money - something which most BOINC projects are short of.

I think that's it. Anyone got anything to add?


Giskard - the first telepathic robot.

About the connections to the database:

@Developers:
Have you ever heard of connection pooling? That's the solution to problems like you are getting here. No offense intended, but that's what i've (as a web developer) seen always if a bottleneck like this occurs...

Just a thought.



greetz, Uli

All you need to know about this error is in this thread. Just wait and your work will be reported.


Giskard - the first telepathic robot.

> About the connections to the database:
>
> @Developers:
> Have you ever heard of connection pooling? That's the solution to problems
> like you are getting here. No offense intended, but that's what i've (as a web
> developer) seen always if a bottleneck like this occurs...

Maybe, maybe not. What you actually mean with connection pooling?

Markku Degerholm
LHC@home Admin

MikeW got it right.

System overload is actually pretty natural after longer periods of no work / no service, because most of the 7500 active hosts try to connect within one hour and download/upload dozens of workunits at once. And if connections start to fail, they will try to reconnect soon... And so there are yet more connections attemps. And then there are forums and other web things that also generate more load to the database.




Markku Degerholm
LHC@home Admin