Hi! > From: John Lauro <jlauro@umich.edu> > To: nntpcache-users@suburbia.net > Cc: nntpcache-users@suburbia.net > Subject: NNTPC: NNTPC : Multiple servers for redundancy? > Date: Thursday, July 10, 1997 6:47 AM > > On Thu, 10 Jul 1997 09:56:06 +1000 (EST), proff@suburbia.net wrote: > >The different article id's is unfortunately a problem that can't > >be resolved. Infact nntpcache goes to quite some lengths to avoid > >article ID collisions even in the normal case i.e for crossposts > >to the same group from different servers. > > It would mean that for every group scanned, the server would have to > download the message ids of every article in each parent server and > keep it's own history database (with multiple possible article sources > (which local article #, server1/article#, server2 article #, etc...), > and it's own active file not based on the cached servers. This would > be significantly more complicated programming and probably generate a > little more traffic then it does now, but still far less traffic, and > far less storage, and a smaller history then a full load generates.... In fact, as far as I understand, it already has something like this mapping (message-id <=> article id). It "just" would have to be extended to multiple servers. However, I too feel that NNTPCache should use its own independent article numbering (for single server it would degenerate to direct mapping). Among other pluses, it would solve the Xref problem. Hopefully, it would make the behavior more sensible if the primary server is down (the 'Group not available'/'Article not available' errors seem fairly silly. Maybe some generic NNTP error code would be better for articles, and active/XOVER should work as if nothing happened (i.e. no new articles). What do people think about this? CU Genie
