On Mon, Sep 27, 2004 at 06:24:03PM +0200, Someone wrote: > Disclaimer: The following is only based on my experience with running > freenet on Win2k/XP with the 1.4.2_05/1.5 RC JVM and may not apply to > anyone else. > > 1. The memory usage > I noticed that the aggressiv garbage collector is now disable per default > (aggressiveGC is set to 0), in my old config I set it to run every 60 > seconds. > Setting it back to this made my nodes memory usage normal again, I'm > currently > trying it with a setting of 90 seconds and it also looks quite well. But be > warned, the aggressiveGC will cause a high CPU spike when it runs, which > will > clearly be noticable on the node and on the machine. But to me it's better > than > a node that OOMs within a day.
THAT is interesting. We have repeatedly been fed the dogma that the JVM can handle this perfectly well by itself... I'm quite prepared to reinstate it if this is commonly seen... Having said that, reports of OOMs are rather rare - what did you set the memory limit to? > > 2. The "routing" > With the default settings my node didn't perform so well there eighter, > frost > messages appeared slow, fuqid downloads and fiw inserts seemed to take a > good > amount longer. What's the CPU usage? > After I got the memory usage back to a normal level I started > to look at this (at first I thought it was related to the memory usage, but > it > didn't look this way then).After I tried some different settings I may have > found the reason, it is the failure table. On my old config I toned it > heavily > down: > > failureTableSize was at 2000 > failureTableItems was at 1000 > failureTableTime was at 18000 I very much doubt it. The failure table does not DO anything - or very much. Look at the stats... > > Setting them back to these seemed to help my node to get up to it's old > performance again. We will get rid of the failure table soon but this looks like a placebo to me... > > For the memory usage I'm quite shure that the aggressiveGC is the solution, > for the "routing" performance I really can't say if it's really related to > the toned down failure table or if it's just some sort of conincidence. > > I hope this may help some of you and I would be very happy to hear wheter it > did or not. Very interesting. Everyone else on the list buys into the holy doctrine that the JVM should be left alone to deal with GCing. -- Matthew J Toseland - [EMAIL PROTECTED] Freenet Project Official Codemonkey - http://freenetproject.org/ ICTHUS - Nothing is impossible. Our Boss says so.
signature.asc
Description: Digital signature
_______________________________________________ Support mailing list [EMAIL PROTECTED] http://news.gmane.org/gmane.network.freenet.support Unsubscribe at http://dodo.freenetproject.org/cgi-bin/mailman/listinfo/support Or mailto:[EMAIL PROTECTED]
