On Wed, 05 Sep 2001, Oskar Sandberg wrote: > On Tue, Sep 04, 2001 at 10:39:45PM -0500, thelema wrote: > > Most of this is conjecture, but this is what I believe is causing the > > problems: > > > > 1) the Heisenbug > > There's a lot of restarts in the network; nodes aren't routing > > nicely, and this part of the problem I blame on the heisenbug: a > > problem somewhere between the JVM and the code that causes nodes to > > pervert unique IDS changing 123456789a to 1234512345. > > This one is truely weird, and I really can't imagine anything but a JVM > error. I don't know how far Tavin got on it before his current > jail stint (you gotta stop exposing yourself in public Tavin!)... > I wonder if there was a defcon-like event in china that tavin went to... :) (think skylarov)
> Are you seeing Tavin's "PANIC" log entries? > I don't have the string "PANIC" in my log anywhere. <SNIP> > I think the whole algorithm needs to be improved somewhat. The connetion > pruning code needs to run early (as soon as the connection limit is > 3/4ths full or something), and should remove old connections more > agressively. > I keep wanting to setup a system where nodes keep a single connection to any neighbor they're going to talk to and use that to determine whether the other node is up and for routing requests. Seperate connections for pushing data around are fine, but they should be short-lived as opposed to these longer-lived connections which could speed up routing tremendously. The two types of connections would be subject to different connection limits, but pruning would be nearly trivial. > > 3) some sort of actual connection problem > > Nodes shouldn't have any problems authenticating to each other, but > > I see way too many connection problems for it to be just network > > issues. I have to suspect some heisenbug in the authentication > > (possibly related to the real heisenbug due to a defect in > > encryption/decryption code) > > It's very hard to believe that an error in the stream code could cause > the problem to occur in the same field every single time. > I agree that it'd be amazing, but I don't believe in an error in the JVMs causing this problem, and when you have eliminated the impossible, whatever remains... blah blah blah. > Oskar Sandberg Thelema -- E-mail: thelema314 at bigfoot.com If you love something, set it free. GPG 1536g/B9C5D1F7 fpr:075A A3F7 F70B 1397 345D A67E 70AA 820B A806 F95D -------------- next part -------------- A non-text attachment was scrubbed... Name: not available Type: application/pgp-signature Size: 232 bytes Desc: not available URL: <https://emu.freenetproject.org/pipermail/devl/attachments/20010905/7dec8b30/attachment.pgp>
