🚀 go-pugleaf

In article <48gheb$fke@cluon.com>,
Scott Statton N1GAK <scotts@cluon.com> wrote:
| In article <48g8si$1p1@crash.microserve.net>,
| Dave Ratcliffe <dave@frackit.com> wrote:
| >dane@sonic.net (Dane Jasper) wrote:
| >
| >>32 megs of ram is nothing for news - our news server has 96 megs to handle
| >>about 12 days of news.
| >
| >What does the amount of RAM in a server have to do with how many days
| >worth of news you retain in the spool? 
| >
| 
| from icd.c in the inn1.4 distribution:
| 
|     ICDactpointer = mmap((caddr_t)0, ICDactsize, PROT_READ|PROT_WRITE,
|                         MAP__ARG, ICDactfd, (off_t)0);
| 
| ICDactfd is the 'active' file.

Of course the size of the active file has absolutely nothing to do
with the number of days you keep news, either, the mmap() call is
conditional depending on whether you use MMAP or not, and the size
of an active file on a server with ~16k groups is only 700k.

I don't think the original "96 megs" post understood the
relationship of retention and memory, and this certainly doesn't
clarify it.
| 
| If you're on a machine that gives poor swap performance (i.e. any toy
| computer being pressed into Real Computer tasks) your ratio of
| physical to virtual memory had better be pretty damned near unity, or
| you're going to be hating life.

Since the performance is related to swap rate, which may or may not
be related to the real/virtual ratio. The critical item is locality
of reference. This is (properly) discussed in comp.arch, since it
not really a news issue.