Re: High disk IO

On Saturday 14 May 2011 02:53:25 Dennis Nezic wrote:
> On Wed, 11 May 2011 17:48:36 +0100, Matthew Toseland wrote:
> > On Wednesday 11 May 2011 02:31:51 Dennis Nezic wrote:
> > > On Fri, 6 May 2011 15:55:59 -0400, Dennis Nezic wrote:
> > > > On Fri, 6 May 2011 18:57:13 +0100, Matthew Toseland wrote:
> > > > > On Thursday 05 May 2011 03:26:45 Dennis Nezic wrote:
> > > > > > For the past several builds, I'm not sure exactly when the
> > > > > > issue started, before build01364 at least, my node
> > > > > > experiences quite high disk IO, which bogs down my computer.
> > > > > > I think once I left it alone, and it resolved itself after a
> > > > > > few hours, but usually I have to restart Freenet to make my
> > > > > > server more useable.
> > > > > > 
> > > > > > My "vmstat 5" reports bi (bytes-in, that is: reading) between
> > > > > > 4000-10000. (Under 100 after I stop Freenet.)
> > > > > > 
> > > > > > My graphs[1] show it to be suspiciously periodic, although
> > > > > > that might just be a coincidence.
> > > > > > 
> > > > > > Is there some kind of background (periodic) database/datastore
> > > > > > task that might be causing this? My wrapper.log shows no
> > > > > > output during these events, nor does fproxy say that anything
> > > > > > is going on.
> > > > > 
> > > > > Active downloads/uploads?
> > > > 
> > > > Yep ... a few active downloads -- about 30MBs each. (One got
> > > > completed much earlier in the day -- but nothing unusual was
> > > > happening during the problem peaks.) It happened again last night
> > > > [2][3], suspiciously almost exactly 24 hours after the last time.
> > > > I didn't stop my node this time, and it "fixed" itself (at least
> > > > the plateau) after over 2 hours, although disk IO was still quite
> > > > high throughout the night [2] and early morning.
> > > > > > 
> > > > > > [1]
> > > > > > http://dennisn.dyndns.org/guest/pubstuff/freenet-periodic-high-io.jpg
> > > > [2] That's what she said.
> > > > [3]
> > > > http://dennisn.dyndns.org/guest/pubstuff/freenet-periodic-high-io-2.jpg
> > > 
> > > It definitely seems daily/periodic... it seems to occur pretty much
> > > every single day at 8pm. It just happened an hour ago, for example,
> > > and resulted in "load average: 7.09, 6.12, 5.38" which seriously
> > > bogged down my system, so I had to kill/restart my node :\. Ideas?
> > 
> > Maybe you could get a stack dump during that period?
> 
> My stack dumps didn't show anything glaringly wrong. (No (or just once
> a) BLOCKED threads.) However, I think the problem has to do with FMS --
> shutting down FMS stopped the high-IO. (The reason why I originally
> suspected Freenet was involved, however, was because stopping my node
> also seemed to stop the IO a few seconds afterwards.) I'm still not
> sure whether it's purely an FMS bug (perhaps having to do with my
> 150MB fms.db3 file), or some kind of FMS-Freenet interaction. I will
> try to investigate further.

You're looking for threads that are persistently RUNNABLE across multiple dumps. An interaction with FMS is quite possible...

_______________________________________________
Support mailing list
[email protected]
http://news.gmane.org/gmane.network.freenet.support
Unsubscribe at http://emu.freenetproject.org/cgi-bin/mailman/listinfo/support
Or mailto:support-request-RdDMkVZAZeuJnvDnx1genB2eb7JE58TQ@public.gmane.org?subject=unsubscribe

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.10 (GNU/Linux)

iEYEABEIAAYFAk3OeYMACgkQYUNbc3WUHYjXLwCdHRRSC2r76WCSTuSOx4TKlz71
d0IAn1zel7afRdX6Qu586TI0JghqYTJx
=CoKs
-----END PGP SIGNATURE-----