On Sun, 16 Aug 1998, Evan Champion wrote: > > What is you cache efficiency, btw? You might want to reset > >the stats when the cache is full, to see what your efficieny is like with > >a full cache. I suspect that you are getting less than 20%. Adding more > >drives to cache can often pay for itself. I found that there is > >cache size threshold (depending on the size of your readership, and the > >amount of available articles on the home nntp server), after which > >efficiency starts to really improve. I'm currently at 32%. > > Hum, my stats appreas to be broken: > > article <stats@nntpcache> No, this interface is depreciated. Doesn't anyone read the release notes? :) Use the web interface. > What we found was that an empty nntpcache provides better performance than > no nntpcache at all for low-speed readers. The theory at the time was that > because the bandwidth from server-nntpcache was far greater than the > bandwidth from nntpcache-readers, the nntpcache would get the article and > then feed it in a constant datastream to the user, whereas in the direct > user-server connection there may be stop/starts due to net burps. This is Hmmm, but caching GROUP output and LIST ACTIVE output is very nice too. Caching GROUP doesn't save very much bandwidth either, but readers can issue a lot of group commands, and some servers are quite slow to respond to them. > the main reason we run nntpcache, not for any bandwidth reduction. However, > you're absolutely right, I do need more disk for the cache. I just don't > have any more disk around at the moment :-) Then you might just want to disable article caching (set article timeout to zero in nntpcache.servers). You still will cache LIST, GROUP and XOVER stuff, which generally gets requested a lot more. > Evan Tom
