response 1 of 127:

Oct 9 01:46 UTC 1994

Disclaimer 2 looks like a hell of a problem to deal with.

Other than that, I like it.  The only other problem I can
think of is funding it, and that's much less of a problem
these days.

Of course, the staff may know something I don't...

response 3 of 127:

Oct 9 05:53 UTC 1994

If we were running on the same type of computer as M-Net could we take
their fixes to problems and end up a year ahead of where we otherwise
would be?

response 5 of 127:

Oct 9 13:06 UTC 1994

If we were on a 486 would that increase the pool of available
talent to help on hardware problems?  Sometimes I get concerned
that just a very few people here end up fielding all of our
hardware and software problems.  At some point they're going
to need a well-deserved break.  If moving to another platform
would help out in this concern that would be a very good reason
to consider the change.

response 7 of 127:

Oct 9 23:37 UTC 1994

The main problem is software. We would have to rebuild most of what we have
on the system. Did I say most? Change that to "all". I put 200 hours
into setting up the Sun-3, installing the OS, configuring it, douwnloading
PD software, building and installing it. I think Mju put in another 100.
I certainly don't have the time to do that again, and I don't think we
have other people who have the time either.

But before we start throwing out the baby with the bath water, what exactly
is wrong with the Sun-3?
The Sun-3 has actually been very reliable and stable. Much more stable than
the SUn-2 was with 5 times the number of users. Our current problems are:
1.) The new disk has turned up a probelm that is causing lost files and
    system crashes. This problem *is* fixable. It just requires time. If
    staff members can't find time to fix the disk, when are they going to find
    time to build and configure a whole new system?
2.) The password system has major problems. This is entirely a software
    problem and also one that is not related to the native OS. There is no
    gaurentee that moving to different hardware/OS would solve this one.
3.) We had 2 back-to-back breakins by hackers who mucked with the system.
    Again, this is no fault of the Sun-3 and could (and will) happen with
    any system.
4.) The Internet link is slow and has problems. The is not the Sun-3's fault,
    it is a PC-route box problem. This problem would still persist with
    different hardware. And no, we would not want to connect the slip line
    directly to the main box, the interupts would *kill* a 386/486 PC system.
5.) The Sun-3 slows down alot when you put 40 users on it. Yeah, well that's
the only complait I can see about the Sun-3 that makes sense. A 486 would
be faster than a Sun-3, but not by much. The main thing that slows a system
down is swapping. Once you start to swap, the system slows down radically.
We can put 32 meg of memory in the Sun-3 for around $400-500, the same
amount of memory for a 486 box would cost arounf $1100-1200.

In summary, No, I don't think this is the time for new hardware.

BTW, I would like to ask why the original poster of this item thinks we
need new hardware. I seem to remember no specifics, just a vague reference
to "problems".

response 9 of 127:

Oct 10 06:24 UTC 1994

Moving to a different hardare system, in toto, is worth the thought process.
Perhaps nothing more than the thought process at this moment, but
a worthy thought nonetheless.
  
Whilst cnsidering different hardware, what about new(er) Sun stuff? 
 
Anyway, the Route Box can be fixed with an itty-bitty 386 (ofthe
proper type), and the cracker problem is going to exist no matter
what hardware is used. 
  
For ne, I'm pretty pleased with the increased performance that this
Sun3 has permitted - and thankxx! to all who made the transition
so smooth. 
 
Whatever, though, considering future hardware is necessary.

response 11 of 127:

Oct 10 06:27 UTC 1994

Response #10 was in response to #8, TS slipped in at #9.

response 13 of 127:

Oct 10 12:14 UTC 1994

I'm going to ask Jep to take a look at this discussion as he
was part of the hands-on group that took M-net from their Altos
platform to the 486.  He may have some insights as to what kind
of problems we can expect.

Regarding software man hours - I don't know if PicoSpan is as
yet available for the 486.  M-net found this to be a problem
in their transition.  So they moved to a PicoSpan clone, Yapp,
which is available for $100.  Now, I'm not suggesting we run
right out and change our software - not at all.  But what
I am suggesting is that we explore the options and then make
decisions based on what we know.

Having a hardware base where we could afford to move to new
equipment would be rather nice and would again, in the long
haul, make us less dependent on our volunteer staff.

I'm glad Valerie entered this item.  

response 15 of 127:

Oct 10 14:06 UTC 1994

Mary asked if I might drop in on this discussion.  I apologize for not
being around grex much lately, but my time has been decreasing for fun
activities.  At any rate, here goes...

M-net came up on BSDI/386 in a hurry.  Why?  Because the old altos died
and we had no choice.  So, we came up without a lot of the supporting
software, and fixes that we should have.  Given a little less than the
part time week that I put into bringing the system up, we would have
been much better off.  Still BSDI has been a reliable operating system
for us.

If you were to choose something like BSDI, I think porting software 
would be a non-issue.  Almost everything we needed came right off the 
distribution CD.  We have added a few things, but primarily we are 
running off the shelf software.  

BSDI comes with it's own shadow password database.  The system itself
has been very secure.  Take for instance the fact that we have not had
to shutdown the system for a security breach yet in the year and a half
that we have been running BSDI.  

I would also disagree with Steve that the slip bandwidth would eat up
a 386/486.  We are running on a 486/33 with two ppp lines, one at 9600,
the other locked at 57600 with v.32/v.42bis 14.4 modems attached.  We
are not seeing anything of the sort.  With 30+ users on the system we
rarely see load averages that approach 1.

Actually being able to buy off the shelf hardware is wonderful.  Want
more serial ports?  Order any of the myriad of supported cards.  Want a
higher speed ethernet or scsi card?  Get one.  All easy options on PC
platforms.

I will admit to being one of the biggest naysayers for the 486 platform
for M-Net.  I thought that there was no way in Hell that we would survive
on a pc platform.  I have pleasantly been proven wrong.

Feel free to ask direct comments, I'll try to answer them.

response 17 of 127:

Oct 11 01:00 UTC 1994

   Its interesting that Valerie brought this up now, because I've
been thinking about this lately.  This is long, but hopefully not
too boring.

   Overall, we're better off with the Sun-3 over AT class hardware,
for at least a while.  With the Sun-3 we're on a cheap platform.  Many
of the components of the Sun-3 are *cheaper* than AT class hardware:
8M of memory for $100, 32M for $450 - $500, SCSI controllers for $125,
tape controllers for $100 - $125, etc.

   So even though the Sun-3 isn't as fast as a 486DX2 at 66Mhz,
in terms of throughput the Sun beats out *most* AT class hardware.
This sounds rash, but consider the fact that the Sun's I/O system
was designed for a 32-bit environment from the group up, and can
do 64-bit I/O from the CPU to cache.

   And we haven't talked about upgrading CPU's yet, either.  We
can get for less than $500 (I've been seeing $450 a couple of
places now) a Sun-4/260 CPU card, which would be just a shade
under a factor of three times faster than what we have now.  All
of our hardware investment remains intact.  If that isn't good
enough, we could later on get a Sun-4/3xx series CPU card, which
is again about 2.5 - 3 times faster than a 4/260.  The drawback
with this is that we'd have to buy matching style memory cards
too.  So a Sun-4/3xx isn't cost effective for us now.  But in two
years, it could be.  Prices are only going to drop.

   On the other side of the coin, there are 90Mhz Pentium class
motherboards out there are are VLB 32-bit creatures that are
pretty fast.  Certainly there is a *VAST* array of choices, some
of which are technically good (remember, a *whole* *lot* of the AT
class hardware isn't very good, and wasn't designed for long term
use--completely unlike the Sun's which were often used in 'harsh'
environments).

   The CPU is actually just about the least important factor
for us.  I/O throughput is.  Certainly anything less than a VLB
would be inadaquite for us.  But that doesn't mean that we shouldn't
investigate this more in the future.

   The *main* reason to think about getting off the Sun platform
is to make sure that future versions of Grex staff can deal with
things in a useful fashion, and make it such that no one staff
person carries completely arcane information with little hope of
passing it on.  With todays AT class hardware, if a SCSI controller
goes south, fine.  You can always find at least one dealer in nearly
any size town that has something you might be able to use, and at
the worst, FedEx can always get something to you in 24 hours.  Because
there are *so* *many* AT class machines about, there will be replacement
hwardware for them well into the next century (Think I'm wrong?  Look
at the number of sources for *pc* hardware still out there today).

   On the other side of the coin, Sun stuff is always getting
cheaper.  Sparcstations of the earlier flavors are now getting
into something near our price range.  Eventually, the book value
of them will drop beneath the notice of accounts and we'll get a
donation of one of one of them.  Assuming we get a terminal server
in the future to handle our dial in lines, we'd be able to move
to one of them without much trouble (Assuming we've already gone
throug the troubles that Greg talked about in initially getting
to a SPARC system).  So riding the trailing edge of technology is
fun, and makes a lot of economic sense.  Castoffs are always in
the offing; we just have to get better at driving cars under
loading docks...

   Someone asked about power.  It is always true that older
computer equipment is more power slurpacious than the new stuff.
Forunately, we've gotten rid of nearly all the *really* power
hungrey stuff, like the SMD disks.  Right now we are still using
one Eagle SMD disk, but once we're off that, we'll be rid of our
major space heaters.  A 486 system of roughly comparable complexity
to Grex's Sun-3/260 would draw about 250 watts of power.  The Sun-3/260
eats about 750 - 800 watts, so thats about 500 watts an hour more.
At 10 cents per KWh, that makes for about $36 a month extra that the
Sun uses.  But since we'd have to spend more than $4000 for a {case,
power supply, motherboard, 2 16M simms, mono monitor/vga card, SCSI
controller, 600M disk, floppy, etc} we're looking at an interesting
payback.  Does spending an extra $36 a month make sense, rather than
spending $4000+ to save that $36/month?  I'm thinking that it does,
since each newer thing we add to the Sun-3 will save money (like,
if we got rid of the 4 8M memory cards with another 32M card, we'd
likely save $6 - $10/month in electricty!).

   So, all in all, I think staying with the Sun platform makes
the most amount of sense for the current time.  If a pile of
money came along and staff were able to buy the latest of
whatever we wanted, things might be different.  But since we're
still a small operation, I think we're likely better off staying.

   But what about "the disk problem" some of you are thinking?
Good question.  That is *the main* reason I could come up with
during a debate on this subject that would hurt the Sun camp.
There are solutions for this, crufty at first, and then real ones.
The first (crufty) one is to set the limit of the last partition
on the disk to 1G such that the mythical/mysterious 1G boundary
isn't crossed.  Once we've done that, we'll know much better whats
going on, either way.  If the problem stops, then we've a addressing
problem of some sort in the kernel such that > 1G calculations
somehow get munged.  If the problem stays with us, then we have
a much different (hardware?) problem than we have thought we
had, and we'll go that route.

   So unforunately we've hit just about the "worst case" scenerio
for us; an interesting intermittane problem, on old hardware,
with no support from the vendor.  I still think that someone else
has hit this problem however, so I think that we will be able to
find a solution for SunOS 4.1.1.

   And if we can't?  That might make for a really good case to
switch to a Sun-4 card as soon as we can compile things.  But
why go to another old card?  Simply because even the old SPARC
equipment is based on the still available (and available for a
long time to come) standards like SCSI devices, simm memory, etc.
And, since the Sun-4's are SPARCs, software ceases to be a problem,
in that SunOS 4.1.3 was a rather stable system, and while it had
problems, there were patches for most problems.  And if we (shudder)
decided to go with Solaris, we could.

response 19 of 127:

Oct 11 20:16 UTC 1994

PCI is of course an emerging standard.  We have an EISA Bus-mastering
card over on M-net.  Again, without some of the problems Steve describes.

Buying a 486/66 dx2 might still be overkill.  the 486/33 we are on just
doesn't see cpu bottlenecks.  Grex is very lucky to have Steve and Greg
to handle the Sun technical problems as they come up.  I wonder how many more
people they might have if they were on a newer hardware platform.

The price of PC hardware is dropping almost as fast as the SUN equipment Steve
describes. New disks are int he $.050 a meg range.  486 systems are dropping
quickly because of the Pentium systems that are shipping.

response 21 of 127:

Oct 12 02:13 UTC 1994

   Thanks Mary, but there are others who've done more than me on
this platform, like Marcus.

   As far as universal modern hardware goes, I think I could argue
the case that Suns are *more* standard than AT class hardware is.
Yes, there are hundreds of companies out there that make AT class
hardware, but not all of it is compatible with UNIX.  Sun stuff we
know is, by default.  Among UNIX folk, Sun hardware is very much
respected and liked.  Finding future staff that can deal with Sun-4
systems or a SPARCstation will not be hard.  THats the easy part.

   THe cost to move to an AT platform, right now, using JCC prices
as a *rough* guide are:

486DX2 66MHz motherboard $ 800
2 16M simms              $1200
case & power supply      $ 150
mono VGA card & monitor  $ 200
keyboard                 $  50
SCSI controller          $ 250
660M SCSI disk           $ 350
BSDI op system           $1000
CD ROM drive             $ 250
16 port Annex (used)     $ 700
2M floppy & controller   $  80
4mm tape backup          $ 700
               total     $5730

I need to explain a couple of things.  1) the 660M SCSI disk is
a stupid estimate of what we'd need for the infrastructure of
Grex as we're setting things up.  Unless we want to bring the current
Grex down for a couple of weeks we'll need this disk as we get
things going.  2) BSDI binaries are $600, but we need the source,
hence the extra money.  3) BSDI comes on CD ROM, hance the need for it.
4) An Annex terminal server is more expensive than an AT class
multiple I/O serial card, but well worth it.  5) the 4mm (or maybe
8mm) tape drive is more than a "DC600" style tape cartridge, but
again more worth it.  6) the 2G disk would migrate over to the AT
when time time came to switch things.

So, going to AT class hardware isn't cheap, requires a lot of
work (not that going to a SPARC wont, either), and in terms of
speed, buys us less than we'd hope for.  It does have the
undenyably wonderful feature of making it easier to get parts,
is that 1000 dealers in the US might have a particular part...

response 23 of 127:

Oct 12 02:19 UTC 1994

Steve slipped in.