>My ISP wants to upgrade DNews and move it to a new machine; the new >one is a Compaq PIII 833 with around 512k RAM. I think they will >continue with a Redhat Linux platform. The upstream server has about >54k groups. > >I've suggested they configure the new server for header sucking; >the upstream server is running DNews version 5.4f4 with dmulti, >and it has about 54k groups. The ISP server has no local groups. > >Based on this page http://www.newsadmin.com/top100bytes.htm >I'm proposing this setting for the groups to header suck: > >header_groups *binaries*,*warez*,alt.fan.*,alt.mag.p*,alt.als > >Are the old headers in the header sucked groups automatically removed >by the replicate mode of header only sucking, or are the headers >stored in the piles where 'expire.conf' expires the old headers? Yes and yes. There are two things going on, 1) the sucking mechanism expires the index entries for items that have disappeared upstream as best it can, 2) The actual headers are inside pile buckets and are only deleted when that bucket is deleted >Can you suggest an 'expire.conf' for a server with 54k groups that has >header sucking enabled? How about one main pile, and another rule >just in case the news admin decides header sucking isn't reliable and >disables it? remember 3 pile 1 use 94% groups * pile 2 use 1% groups none unread >Does this look OK? I would recommend another pile for groups which match your header sucking flag, e.g. remember 3 pile 1 use 74% groups * pile 2 use 20% groups *binaries*,*warez*,alt.fan.*,alt.mag.p*,alt.als pile 3 use 1% groups none unread ChrisP.
