🚀 go-pugleaf

In article <47vvve$v2i@usenetw1.news.prodigy.com>,
bill davidsen <davidsen@tmr.com> wrote:
>In article <47sg5v$pci@yama.mcc.ac.uk>,
>Geoff Lane <zzassgl@cs6400.mcc.ac.uk> wrote:
>
>| Broadcast TV channels have huge data capacity.  The entire news feed for a
>| day could be broadcast in a few minutes.  You do need special h/w at the
>| receiver though.

PageSat has a new USENET Satellite receiver capable of up to like 115K
baud.  Initially the feed will be at 38.4K.

>Based on the rate at which RS/6000 (R20) dedicated servers will
>accept news, and what I've seen on similar Suns and HPs, the
>"special equipment" might be a Cray. You can accept and store
>batches at high rates, but the unpack time is still going to be
>long.

Doesn't have to be.  A lot of the delays are with the news software,
which does a whole lot of stuff to insure no duplication among other
things.  With adequate memory, and if you can be relatively sure you
aren't downlinking any duplicates you can speed up memory unbatching by
a LARGE factor.  like 4-8 times as fast.  I'm working on a custom news
system for the satellite feed.  It makes a LOT of assumptions, but is
totally designed for SPEED, SPEED, SPEED.  It is going to be a real
memory PIG, and will probably require at least 32MB of RAM to run
effectively.

>I see 15-20 art/sec unpacking batches from UUCP feeds, so 300k/20
>would be ~15000 sec, or over 4 hours. Assuming that the system can
>crank at maximum rate, which it certainly does not under reader
>load.
>
>At some point in the not too distant future the idea of a "full
>feed" is going to change. Fortunately the technology to handle higher
>volumes is understood, it just needs to be deployed. That would
>appear to be a readily scalable solution.

The biggest problem with the current usenet system is the way articles
are stored, ie. as individual files.   Filesystem code is not very
efficient at handling file open/close/create/destroy calls due to the
dynamic nature of modern filesystems.  A better model would be something
like the old S/34 S/36 linear model with static linear filespace, or
simply using an article database.  Unfortunately the former would be
difficult to implement directory tree's in, and the latter would
dis-allow the use of common newsreading software.