response 1 of 34:

Jan 27 15:41 UTC 1995

for the moment I agree with sidhe, as I currently use a secure Merit
terminal which I use to connect to MSUGopher/Mlink in order to then
get to Grex. however, I remember one of staff (mju? maybe?) sent
mail to The Powers That Be in an attempt to find out why Merit is
being so fuzzy with Grexing. I think the advantages of the new router
are things that technologically illiterate users like myself *won't*
see, so I guess if the access thingy would clear up, I'd be a 
happier camper. :) :) :)

response 3 of 34:

Jan 27 18:40 UTC 1995

The old router was PC-Route on a 286.  It worked ok for light
interactive loads, but it didn't work so well for heavy loads, and
didn't work at all for sustained transfers >30K.  The router has to
convert from ethernet speeds to serial link speeds, which is only about
1/100 the speed.  In the long run, that's not a real problem because TCP
will eventually figure out how fast the link is, and pace itself.  But
in the short run, making that work depends on the router being able to
buffer the data - and the PC route box didn't have much buffer room, so
it discarded traffic whenever things got too much for it.  Eventually,
of course, the traffic would get resent, but that's where the network
lag came in, and why sustained transfers choked up.

While mju was home over christmas, we had him work on upgrading the
router to a 386 w/ netbsd.  In theory, this will work a lot better.
netbsd supports PPP (instead of SLIP) which should lead to better
efficiency.  Netbsd also supports a lot more buffer space, which should
cure the sustained transfer problem.

Unfortunately, in practice, netbsd isn't fully debugged, and the version
we installed has serious bugs in its serial driver, which manifest
themselves under heavy load.  We fixed a few problems, but by the time
we realized just how serious the problems were, Marc had to go back to
school.  We then were faced with the usual warehouse access issue, plus
also the move.

The current plan is to try freebsd, which (we hope) will have all of the
advantages of netbsd, but which is also supposed to have less flakey
serial drivers.  We already have at least 3 people who are willing to do
this, so the major constraints are scheduling, and figuring out how to
do it with the least amount of disruption and risk.

One question people here might like to think about is: would they mind a
definite 8 hour outtage, if it would make it possible to try freebsd
that much sooner?  This would basically make the difference between
doing it monday, and doing it somewhere towards the end of the week.

response 5 of 34:

Jan 27 22:56 UTC 1995

As a lay-person answer, here's my understanding: the new router should
theoretically double Grex's internet throughput.  This would allow everyone
to telnet faster, or more people to telnet in.  But it's taking a while to
iron out its "kinks" (which mdw & steve elaborate on :-).  I'm sure there's
no conspiracy to degrade your access; staff was and is *trying* to improve
things for you and everyone.  (I'm appreciative - thanks staffolk!)

response 7 of 34:

Jan 28 04:22 UTC 1995

We've been off the Internet for a few hours now, so I guess it might be
being worked on right now.  But, if it isn't, it might make sense to take
gryps out of service until we can get it working.  The old PCRoute box,
for all its faults, worked consistently.  As for the new router, it's nice
to have it fast when it's up, but when it's going down for several minutes
at a time every half hour (which was happening last night) it gets to the
point where any speed increase we are getting from it is not worth the
trouble it's causing.

response 9 of 34:

Jan 28 08:35 UTC 1995

Replacing the PPP link between gryps and ICnet with a SL/IP link,
which is what the PC route uses, requires a coordinated change with ICnet.
This complicates the problem of switching back to PC route in the interim.

response 11 of 34:

Jan 28 20:43 UTC 1995

Linux is SysVish in nature, it's networking code is considered sub-standard
and nobody on staff wants to deal with it. 

response 13 of 34:

Jan 29 04:15 UTC 1995

        Yes, I was <and am> annoyed at the router problem, but, no, I wasn't
pointing to a conspiracy against me in particular. I am grateful that
it is being worked on, and I am also grateful for the explanations as to what
was going on, as that is why I entered this item. Again, thank you all.

response 15 of 34:

Jan 30 03:59 UTC 1995

        As in Grexnews? I'm going to have to go look at that new cf!

response 17 of 34:

Jan 30 22:35 UTC 1995

   We'll definately take the 286 Steve, because if nothing else, it
can turn into JCCfodder and make us a little money.
   Yes, the router is now a 40MHz 386 system with 340M of disk,
running NetBSD.  It even mostly works.  Mostly.  But that ain't
good enough.

response 19 of 34:

Jan 31 02:01 UTC 1995

   We need to enter an item for gryps.  But: we got a 386 motherboard
case & power supply for $95, and a 340M IDE disk and controller for
$230.  These prices are approximate but pretty close.  So we have something
that we already know is fast enough, and has enough disk on it so we can
build new kernels and keep healthy sized log files.

response 21 of 34:

Jan 31 21:32 UTC 1995

   Oh, sorry--from a tech that Marc knew at Domanint Systems, who
was upgrading his system to a 486 or pentium.

response 23 of 34:

Feb 25 08:12 UTC 1995

Umm, I'm still having the problems as described in the header.  Ther are
not happening as often, but it'd be nice if they didn't happen at all.  
Does anyone have any leads concerning this?