On Thursday 04 September 2003 17:56, Toad wrote: > On Thu, Sep 04, 2003 at 08:26:39AM +0100, Gordan wrote: > > Is it normal for a node without any bandwidth throttling with about 1 Mb > > of symmetric bandwidth to max out at under 16 KB/s (1/8 of the actual > > available bandwidth)? At first, for about an hour, the node will go and > > consume 100% of the available bandwidth, then it will start to overload > > and slow down, and very quickly reach a reasonably steady state at about > > 16 KB/s. This has been observed on a dual 1 GHz P3 server. Quite > > frequently, the node will spike it's CPU consumption to 100% (both CPUs), > > but most of the time it is using up at least all of one CPU. > > Bandwidth throttling is enabled by default at 16kB/second outbound. > Please check that you have uncommented the line: > outputBandwidthLimit=0 > > Otherwise, I have no idea...
Yes, I explicitly set it to 0, which IIRC means unlimited... > > I have deliverately switched off bandwidth throttling to see if this was > > the cause of the problem, and it would appear that it isn't. > > > > The number of threads and the routing time are on overload most of the > > time. > > That's probably something to do with it. Yes, it looks like the node is under too much load to serve any requests - goodput collapse. > > This has been observed with unstable 6162, 6163 and 6167 on Linux, Sun's > > JVM 1.4.2. 6168 seems to be doing slightly better but not by much. > > With 6162? So it's pre-ngrouting. Okay, that's a useful additional > datapoint. Actually, 6168 has been doing quite well for me today, unlike yesterday, so it is possible something strange was going on with the network yesterday. I've also had a really bizzare crash yesterday mentioning pthreads, which it shouldn't have been using... Having said that, 6168 looks rather stable at the moment. The bandwidth seems to be somewhat erratic, but at least it bounces back after it bottoms out. Something that went in between 6163 and 6168 seems to have helped noticeably. Unfortunately, it's difficult draw any conclusions from observations of a live node, because it may well have been external factors causing the poor performance, e.g. a large number of heavy-duty nodes going off line simultaneously, thus causing congestion on other nodes. Gordan _______________________________________________ Devl mailing list [EMAIL PROTECTED] http://dodo.freenetproject.org/cgi-bin/mailman/listinfo/devl
