On Thursday 24 January 2008 21:49, Robert Hailey wrote: > > On Jan 24, 2008, at 3:14 PM, Matthew Toseland wrote: > > >> Unless, I suppose, there was an instant when it was the only not- > >> backed-off peer; then it would average in real quick.
Not that quickly. However, if we start rejecting all requests, we will probably lose our opennet peers. > >> > >> More plausible since (1) these slow peers show up as NOT backed off, > >> and (2) being in this situation makes your peers backoff from you (is > >> a request reject more likely then?). > > > > But the long round trip time will cause it to timeout on Accepted's > > and get > > backed off, surely? The ping time limits (3s max) are lower than the > > timeout > > for Accepted, admittedly (5s). > > Except that the node is not processing any requests, it is rejecting > them all because of the ping time, and I'm not sure how this starts, > maybe it is with requests which are not rejected (like announcements). The question is why does it not stop? The ping time is updated every time a message is acknowledged! Even if neither side is accepting any requests, it should recover quickly, because we are constantly sending swapping related messages, keepalives etc. And the averager in question is time sensitive rather than sensitive to the number of samples. (The problem might very well be there btw). The global averager only has a 30 second half-life so it should recover quickly. The per-peer averagers are much slower, but you claim that there was only one peer which had big problems. > > > Are you sure that this was the problem when you got it? > > No. I'm only sure of the symptoms: (1) sendSync timing out, (2) > pInstantReject=100%, (3) SUB_MAX_PING_TIME and MAX_PING_TIME are the > highest (presumably-only-current) rejection reasons, (4) no requests > being processed, (5) one peer with an outlandish ping time. So replicate it and debug it. For example, NodePinger logs every time it reports an overall median ping to the temporal averager. Another interesting fact is that PacketThrottle works only on the basis of *the last sample*, not the average (to more closely mimic TCP and rapidly respond to network conditions). You might want to add the ping time averages to the advanced mode connections pages. -------------- next part -------------- A non-text attachment was scrubbed... Name: not available Type: application/pgp-signature Size: 189 bytes Desc: not available URL: <https://emu.freenetproject.org/pipermail/devl/attachments/20080124/06a13c56/attachment.pgp>
