Am Donnerstag, 20. M?rz 2008 schrieb Matthew Toseland: > On Thursday 20 March 2008 20:37, Guido wrote: > > Am Donnerstag, 20. M?rz 2008 schrieb Matthew Toseland: > > > On Thursday 20 March 2008 19:51, guido wrote: > > > > For me, this build has brought a significant drop in the load caused > > > > by freenet on my machine. Until now, running freenet has always > > > > caused an average load of at least 4, slowing down everything else on > > > > this box. After this upgrade, the load is a much more bearable 0.2, > > > > and everything else has become much snappier. > > > > > > This is very odd. It's probably back to normal with 1126? Is this load > > > due to CPU or disk access i.e. how high is your typical CPU usage? > > > > I think you did not read my message right. I said the load /dropped/ from > > unreasonably high to reasonably low. This is a good thing, and I really > > hope it won't go "back to normal" again with 1126. > > 1125 had some major problems, it is likely that 1126 will again use lots of > whatever it was using.
The node has upgraded now, and at first glance it behaves the same as with 1125. (Fortunately. The load that was caused by the earlier builds was a real pain...) > > The load used to be mainly IO (disk access) and memory usage (causing the > > machine to swap). It's still swapping, but this time that doesn't seem to > > have that much of an impact on the overall load. > > Java and swapping go together *really* badly. I suppose if you reduce the > memory limit you get OOMs and 100% CPU usage? I have reduced memory usage to 120 MB and have not seen either OOMs or 100% CPU usage. Also, just because there is swapping, it doesn't necessarily follow that it's freenet or Java that's being swapped out.
