On Wed, 14 Jan 2004 11:24:17 +0000 Kevin Steen <[EMAIL PROTECTED]> wrote:
> After months of struggling, complaints, frayed tempers and dead-ends, I > think the current Freenet release is a major step forward. I concur. As far as I'm concerned 5058 is, without a doubt, the best stable build in months and one of the best builds I've ever seen. With an uptime of 7 hours: a) my routingTime is only 62ms. b) I have ~50MB of allocated memory unused, so the machine isn't paging. c) CPU is down from constant 100% to averaging about 50% (lots of peaks and valleys in the Task Manager). d) there are only ~2MB of data queued to send. e) so far I haven't seen any more than 200 or so pooled threads running. f) load is hovering at 100% due to outbound bandwidth limit. g) I have 332 distinct nodes connected - holy shit! On my node, all of this was _totally_ unheard of as recently as 5052. RAM and paging in particular had been serious issues for me, as the machine I dedicate to Freenet only has 192MB physical RAM. routingTime would typically stay above 5 seconds, permanently. There would be tens of megs worth of data queued. There were frequently 1,000+ threads going (up until Edward mentioned tfAbsoluteMaxThreads). Load was usually 100% due to either routingTime or messageSendTimeRequest, and my outbound bandwidth was sorely underused. 5088 is superb. Whether it scales well or not, I agree that this is a huge step in the right direction. For once, my node is limited by bandwidth - as it should be - and not by CPU, threads, or RAM. For once, my lousy P3 600 actually seems like an active, healthy participant in the network. Congratulations all around :) -s _______________________________________________ Support mailing list [EMAIL PROTECTED] http://news.gmane.org/gmane.network.freenet.support
