log in

Posts by Kiska

1) Message boards : NFS Discussion : BOINC pentathlon (Message 2183)
Posted 17 May 2021 by Kiska
Post:
Ok Kiska, no more sticks!

It's always fun to wake up and discover Apache PHP crashed shortly after I went to bed. We're back and I'll keep an eye on it.


Sorry not sorry :p



PS: My thing polls the page once per 5 minutes :D
2) Message boards : NFS Discussion : BOINC pentathlon (Message 2179)
Posted 16 May 2021 by Kiska
Post:
PS: Are you able to reduce the time between server status page updates? xD

The server is stable right now. I'm not touching it! :-)


*pokes server with long stick*
3) Message boards : NFS Discussion : BOINC pentathlon (Message 2177)
Posted 16 May 2021 by Kiska
Post:
Work has been going out, but it gets immediately gobbled up. It seems nearly everyone wants work from the lasieve_5f queue, so I've temporarily stopped feeding work from the other queues until those clients are full.


It looks like you're efforts are rewarded:


PS: Are you able to reduce the time between server status page updates? xD

PPS: https://munin.kiska.pw/munin/static/dynazoom.html?cgiurl_graph=/munin-cgi/munin-cgi-graph&plugin_name=Munin-Node/Munin-Node/results_nfs&size_x=800&size_y=400&start_epoch=1620441730&stop_epoch=1621132930
4) Message boards : NFS Discussion : BOINC pentathlon (Message 2168)
Posted 15 May 2021 by Kiska
Post:
To summarize the observations during the first hours after May 12 18:00 UTC (contest announcement), and the first hours after May 15 00:00 UTC (contest start):

  • Work generator speed was a bottleneck during the first hour, but not later anymore.
  • There was so much work generated subsequently, that work generator speed is not going to be a bottleneck anytime soon.
  • However, with more than 250,000 results returned during the first five hours of May 15, first the validator, then the assimilator was maxed out.
  • For several hours now, the server is not assigning new tasks to hosts.

Is the assimilator backlog the reason for the server's refusal to send new work?



Feeders inability to keep tasks in memory for when requests come in.

From memory the scheduler only looks in memory to see if there is any tasks to be issued when its looking for work to send. It is the feeders job to keep said memory full(well cached), this way it doesn't issue lots of tiny database queries for work.
5) Message boards : NFS Discussion : BOINC pentathlon (Message 2163)
Posted 13 May 2021 by Kiska
Post:
yep this is starting to look like someone put 2^x into the work generator limit
6) Message boards : NFS Discussion : BOINC pentathlon (Message 2161)
Posted 13 May 2021 by Kiska
Post:
Well... that is very exponential looking graph




Home | My Account | Message Boards