[Bamboo-Users] high cpu load
Sean Rhea
sean.c.rhea at gmail.com
Thu Dec 20 16:36:46 UTC 2007
Olaf,
What do you mean by constant? One per second? Ten per second? One
per millisecond?
Each Bamboo node uses epidemic protocols that periodically check in
with its neighbors for routing updates, missing key-value pairs, etc.
So they're going to be doing stuff all of the time. But if all of the
time is 10 times per second, that's no big deal.
Sean
On Dec 13, 2007 5:45 AM, Olaf Görlitz <goerlitz at uni-koblenz.de> wrote:
> Hi Sean,
>
> I realized that the high cpu load was caused by a different part of my
> code - not at all by bamboo. So the trace led me into the wrong direction.
>
> But still, as the trace shows, there seems to be a lot of communication
> going on in the core bamboo classes. With all logging turned on there is
> a constant flow of notifications about timer updates etc. - even if just
> one node is running and no other operation is executed.
>
> Is there a way to tune the frequency/amount of these internal updates?
>
> My config looks like this:
>
> <sandstorm>
>
> <global>
> <initargs>
> node_id localhost:3630
> </initargs>
> </global>
>
> <stages>
>
> <Network>
> class bamboo.network.Network
> <initargs>
> </initargs>
> </Network>
>
> <Rpc>
> class bamboo.lss.Rpc
> <initargs>
> </initargs>
> </Rpc>
>
> <Router>
> class bamboo.router.Router
> <initargs>
> gateway_count 1
> gateway_0 localhost:3630
> </initargs>
> </Router>
>
> <StorageManager>
> class bamboo.db.StorageManager
> <initargs>
> homedir /home/goerlitz/dht
> </initargs>
> </StorageManager>
>
> <DataManager>
> class bamboo.dmgr.DataManager
> <initargs>
> merkle_tree_expansion 2
> desired_replicas 4
> </initargs>
> </DataManager>
>
> <Dht>
> class bamboo.dht.Dht
> <initargs>
> storage_manager_stage StorageManager
> min_replica_count 1
> </initargs>
> </Dht>
>
> <Gateway>
> class bamboo.dht.Gateway
> <initargs>
> port 3632
> </initargs>
> </Gateway>
>
> <Vivaldi>
> class bamboo.vivaldi.Vivaldi
> <initargs>
> vc_type 2.5d
> generate_pings true
> eavesdrop_pings false
> use_reverse_ping true
> ping_period 10000
> version 1
> </initargs>
> </Vivaldi>
>
> <WebInterface>
> class bamboo.www.WebInterface
> <initargs>
> storage_manager_stage StorageManager
> </initargs>
> </WebInterface>
>
> </stages>
> </sandstorm>
>
>
> Cheers,
> Olaf
>
>
>
> On 12.12.2007 19:15, Sean Rhea wrote:
> > What OS are you using? Can you attach you .cfg file?
> >
> > Bamboo uses VERY little CPU on PlanetLab.
> >
> > Sean
> >
> >
> > On Dec 12, 2007 3:23 AM, Olaf Görlitz <goerlitz at uni-koblenz.de> wrote:
> >> Hi all,
> >>
> >> I'm experiencing a constantly high cpu load of 100% when running bamboo.
> >> I tried to pinpoint the cause with Java's build in profiler (-Xprof) and
> >> got following output for Thread-1 which is the ASyncCoreImpl instance.
> >>
> >> Flat profile of 198.82 secs (9964 total ticks): Thread-1
> >>
> >> Interpreted + native Method
> >> 99.3% 0 + 9878 sun.nio.ch.PollArrayWrapper.poll0
> >> 0.3% 0 + 31 com.sleepycat.db.db_javaJNI.DbEnv_open0
> >> 0.2% 0 + 15 com.sleepycat.db.db_javaJNI.initialize
> >> 0.0% 0 + 3 com.sleepycat.db.db_javaJNI.DbTxn_commit0
> >> 0.0% 0 + 3 bamboo.dht.ReturnToClient.<init>
> >> 0.0% 0 + 2 java.lang.Thread.start0
> >> 0.0% 0 + 2 java.lang.Class.forName0
> >> ...
> >> 99.9% 5 + 9936 Total interpreted
> >>
> >> Does anyone know why sun.nio.ch.PollArrayWrapper.poll0 is called so
> >> often? How can I reduce the cpu load?
> >>
> >> Regards,
> >> Olaf
> >> _______________________________________________
> >> Bamboo-Users mailing list
> >> Bamboo-Users at bamboo-dht.org
> >> http://bamboo-dht.org/cgi-bin/mailman/listinfo/bamboo-users
> >>
> >
> >
> >
>
>
--
"Humanity has advanced, when it has advanced, not because it has been
sober, responsible, and cautious, but because it has been playful,
rebellious, and immature." -- Tom Robbins
More information about the Bamboo-Users
mailing list