> > > Retieval by message ID is ok for one or two articles, but it > > doesn't scale well. Anyone using it as a method of news sucking > > deserves to be LARTed for wantonly wasting CPU on the news server. > > Agreed. > > > Apart from the CPU issues, a machine which can't cope with having > > the entire history database loaded up will basically be unusable > > for all the other readers on it, so message-ID retrieval turns into > > a denial of service attack. > > An idea I had was to have all message-id retrieval handled by an external > process/thread which was memory/bandwidth limited, and all access queued and > serialized (being careful no to trash the disk buffers in the process). Actually, this is more or less how nntpcache does it. There is only one process that read/writes the history database, and all nntpcache-serverlets communicate with it via ipc. Cheers, Julian.
