🚀 go-pugleaf

ntmtv!harthc@ames.arc.nasa.gov (Howard Hart) writes:

>>Apart from efficiency the main argument that has been used here
>>against NFS is that news is currently on our gateway machine, and
>>if we ran NFS on that machine it would be a major security hole.
>>I don't know how serious a worry this is.

>Don't forget that wonderful feature of NFS when partitions are mounted
>read/write (such as the case where you have to mount the news batch
>directory read/write for users to post articles). Don't know about
>other machines, but on Sun's, if you loose the server, the client
>hangs indefinitely (I was told at one time that NFS writes have
>higher interrupt priority than mouse clicks, so I do mean complete
>screen/window lockup). That alone makes NNTP worthwhile at our site.
>It was very interesting when our old NFS-moounted newsserver went down
>and 300+ machines lab-wide suddenly quit responding--all for a marginally
>(management-wise) supported USENET feed. I'd strongly recommend NNTP-only.

My experience is the opposite. We read via NFS (exported/mounted
read-only) but post via NNTP (to prevent multiple writers).
Prior to that, we did everything via nntp. Building multiple nn
databases wasn't a good idea, of course, but even without that NNTP
performed worse than NFS for reading.

Mounting the filesystems interruptable prevents hangs with no way out.
If the news server goes down, noone, except those reading news, suffers.

This might require SunOS 4.1 or higher, though, as interrupting NFS
wasn't properly implemented and getwd(3) did complicate things in
an NFS environment as well.

Casper
--
						|	Casper H.S. Dik
						|	casper@fwi.uva.nl