Why does it take so Long to Get into Grex?

Feb 12 17:43 UTC 1995

Wow, i entered an item yesterday and it has not seemed to materialize
on the list as of yet. My only concern is that grex sometimes takes an
exorbitant amount of time to get from the point when you are first 
connected to the time you actaully get into a conference. What is causing
this problem. Is it overload? Is grex running to slow? Yesterday at six 
o'clock it took me fifteen minutes to get to a conferece after i telnetted
in. Is there any way to cure this problem?

response 2 of 42:

Feb 12 19:59 UTC 1995

   Yesterday was bizarre. We went from a load average of 7 to 85 within
one minute.  If someone was doing something on purpose to hurt the system
I haven't seen it yet.  I still don't know what caused that.

   In general, then Grex gets more than 35 people the system will periodically
get slower and faster.  Despite the added memory of late, when we get so
many people on at once, the fact that so many different processes are running
bogs the CPU down.  If everyone is in party for example, the system load
will stay low.  But once we get lots of different things going on here, the
system does start to slow.

   Added to that, we have mail nearly constantly coming into Grex.  There
are times when 15+ sendmail connections going at once, and that helps bog
things too.  Forunately, sendmail knows enough to throttle itself back
and not deliver mail to people when the load average hits 8, and to not
accept incoming mail at all when the load average hits 12.  This has *really*
helped Grex run faster.  If you think Grex is slow now, think back to the
days of smail, when smail would merrily fork Grex into oblivion. ;-)

   The bottom line is that we're outgrowing this processor.

response 4 of 42:

Feb 13 02:31 UTC 1995

I think that an upgraded processer would be a strategic move. I know 
you are probably going to say, "You are not a member, so who 
should you be talking" I am sorry. In this age of superconductors and
fiberoptics.......Just wanted to voice a concern. Thanx

response 6 of 42:

Feb 13 05:43 UTC 1995

Orwell, I'm pretty sure nobody on Grex would object to a faster processor!
It's just a matter of priorities.  Many people want other things more than
a faster CPU (like reliability, usenet, or generally better connectivity).

response 8 of 42:

Feb 13 18:53 UTC 1995

What Steve said.
The big problem with upgrading Grex to a faster processor is not money,
it's manpower. If there were a direct, compatible CPU upgrade, that allowed
us to pop out our CPU card and pop in a faster one that would run all our
current software, believe me, we would have already bought it!

Unfortuneately, what we have is call a Sun 3/260 computer, it uses a
3/200 CPU board which is a 25Mhz 68020 processor with 4meg of cache ram
and 128bit memory bus. Sun didn't make a faster 68020 CPU.

The logical upgrade path for our machine is to install a 4/200 CPU board.
This board is compatible with all of our memory boards, but it uses a
Sparc processor instead of a 68020. We can get one of these boards for $500.
The problem is, 68020 code doesn't run on Sparc processors.

We would have to install a new version of Sunos from scratch and rebuild
*all* of our user programs. Elm, pine, sendmail, tcsh, bash, procmail,
write, party, newuser, etc, etc, etc, etc, etc.
Go take a look at /usr/local/bin to get a *small* idea of what's involved.

Now this *is* do-able. It's just not the kind of overnight-slap-in-a-new-
processor-board that some people think it is. It will require us to set
up a new system with one of our other Sun 3/260's(we have several) and
then slowly build up the new system over a period of many weeks(months).
We would also have to buy at least another 1gb of disk for the new machine
because we couldn't touch our current disks, they would be needed to keep
the current system running.

response 10 of 42:

Feb 14 01:00 UTC 1995

When the time comes to upgrade to the Sparc card, what will happen
with Picospan, since we don't have the sources?

response 12 of 42:

Feb 14 20:27 UTC 1995

Kent: What about Picospan?
Steve: We'll get a new copy.

Uh, yeah, you sound so confident of that Steve. The bottom line here is
that Picospan *is* going to be a problem. It was a problem with Mnet.
Even if Marcus is *willing* to build a Sparc version of Picospan, he still
has to find the time to do it. I think we'd want to make sure we had the
new version of Picospan before we did anything else. Without it, we really
can't consider moving off our current platform.

response 14 of 42:

Feb 14 21:39 UTC 1995

I feel so good, being the instigator (or at least one of them) of a 
great grex revolution..Carry on......This is going to be cool

response 16 of 42:

Feb 15 07:03 UTC 1995

PicoSpan for the sparc won't be a problem.  M-net was a problem for a
variety of reasons mostly irrelevant here, except that one was they were
in a big hurry.  Remember, back then, m-net had *no* hardware
redundancy; that's never been true of Grex, and would not be true with
the sparc.

response 18 of 42:

Feb 15 08:17 UTC 1995

I believe the Well runs on Sparc processors. 

response 20 of 42:

Feb 15 13:21 UTC 1995

(???)

response 22 of 42:

Feb 15 19:31 UTC 1995

Re #15, I *think* YAPP's $100 price tag includes source.

response 24 of 42:

Feb 21 04:03 UTC 1995

Re #8&#9:  I am also glad that this (hopefully) future transition is
not only , as sidhe says, accomplishable, but has been thought out,
and, to a certain extent, planned.  :-)