On Sun, 16 Aug 1998, Evan Champion wrote: > > Except basically all article handling will have to be done by > >message-IDs. This will probably slows things down, specially stuff like > >XOVER merging. > > Remember, however, that nntpcache's history file is much smaller than INN's > would be to host a spool covering the same number of days, so you shouldn't > have the same penalties for article retrieval by Message-ID. That isn't the biggest problem. Converting article numbering via message IDs is. For example, one server may have 20000 article in a group, another server may have 15000 article in a group, both with different numbering. nntpcache would have to find the unique article from the union of both servers, and creates its own numbering. > Perhaps a more significant change would be storage by Message-ID. I'd > personally really like to see nntpcache do cyclic article storage because it > does not seem to be very good at all at expiring... Really? I run between 14.5 to 12GB of cached material, and I don't even notice expiration. I don't think you can really store by message-ID. nntpcache needs to remember that article 123 in alt.test is actually article 456 on server 1, and article 124 in alt.test is actually article 460 on server 2, etc. This index needs to be held on a per-group basis. When the GROUP info times out, nntpcache will have to download new XOVER info up to the new highwater mark on each server, and toss out duplicates, and add each article to its group index. Tom
