response 32 of 181:

Oct 13 23:41 UTC 2001

  Several people have already suggested this, but let me just state 
  again as plainly as possible that it makes no sense to talk about
  what kind of upgrade is desirable until there's a general consensus
  about what's deficient with the current setup.  Make a clear list
  of what it is you want to be able to do and *then* start talking
  about what kind of upgrade is needed to do it.

response 34 of 181:

Oct 14 04:24 UTC 2001

``weblogs'' and ``wikis'' doesn't strike me as being all that compelling
for grex.  One can find that functionality elsewhere.

Grex currently does some unique things that only grex does.  grex has a
unique user community running picospan (as well as the web interface).
grex also provides some unique services to ann arbor residents, and provides
some amount of unix shell accounts to anyone who wants one.  Perhaps grex
does some other unique things that I'm forgetting at the moment.

Keeping as many of the unique services that grex offers that people use
as possible is vastly more important than adding services that are available
at numerous other sites, I think.

(Which is not to say that grex shouldn't experiment with things that are
available elsewhere, but if grex causes itself to lose its uniqueness, I
think that would be a significant loss.)

response 36 of 181:

Oct 14 05:19 UTC 2001

I don't think anyone is talking about abandoning any service we now 
provide.  Exploring some new directions would be interesting.

I don't see any reason to upgrade to a new operating system on the 
current hardware.  The main options would be Solaris or netBSD.
I don't think the newer versions of Solaris support this machine, so 
we'd have to upgrade to an old version of Solaris (could be wrong about 
that).  However, I believe we could quite easily get significantly 
better machines for free that would run Solaris or netBSD (such machines
have been offered to us in the past) and it would actually be easier to
get it working on a newer machine because the newer operating systems do 
a better job of supporting newer machines than old obsolete machines. So 
what would be the point of doing the work to install a new operating 
system on an old machine, when we could do less work to install it on a 
newer machine?

I really think we need to be talking about a new OS on a new machine.

I also think we can afford a new machine.  Just to get a sense of a 
possible price, I randomly went to the Dell small business website, 
picked a rack-mounted PE 1550 server, with a 1GHz Pentium III processor, 
and space for a second one, 512MB memory, 36 GB Ultra3 SCSI disk and 
controller, no OS, keyboard or monitor.  That costs $2060.  We have 
$1024 in the hardware fund already, donate a bit from the general fund, 
do a fund raiser, and we'd pretty easily get to say $2500.  I think we 
can get a respectable new machine for that.  It'd take a lot of research 
to figure out exactly what we really want (the machine above definately 
isn't it, for many reasons).  We might end up wanting to build our own 
from parts.  But I think it's doable.

The other advantage of moving to an newer x86 system is that there are 
just *so* many more people who know the hardware than there is for Sun 
hardware.  I could probably entirely stop pretending that I do 
hardware.

response 38 of 181:

Oct 14 05:37 UTC 2001

Good explanations Valerie and Jan; I support your recommendations (as long
as we move from a poxy old OS to one which supports Java, too).

response 40 of 181:

Oct 14 17:26 UTC 2001

Why new and not 2-3 years old?  What can you get for $1024 instead of $2500?
Do we need 1G processor speed? (What do we have now, by the way?).

response 42 of 181:

Oct 14 19:01 UTC 2001

From a pure price/performance viewpoint, I don't think buying used
computers is advantageous.  A five-year-old car will perform pretty much
just like a brand new car, and will cost less (though with a shorter
expected lifetime).  An two year old computer, on the other hand, is
going to be substantially less capable than a new one.  You pay a lot
less, but you also get a lot less.

If Grex couldn't afford a current generation machine, then looking at an
older machine might make sense.  If we can afford a current machine, then
what's the advantage?  Well, there are some, but I'll get to those later.

The Dell quote above was just a quick way to get an idea of what a current
generation machine might cost.  If we actually decided to build a x86
server, then we would almost certainly put it together ourselves from
parts acquired separately.  For some of the parts, second-hand makes
sense.  Power supplies, chassis (so long as they support the form factor
for a modern motherboard), ethernet card, video card, etc.  I'd want to
go with a fairly modern processor, SCSI controller and drives, memory.
I think a lot of the choice would be driven by what kind of memory we can
get most cheaply, because I'd guess that a machine loaded with lots and
lots of memory would be advantageous for us.  I think if we are going
to invest the work to build and set up a new machine, it's stupid not
to invest the money to make it a high quality machine.

There is, however, something you can gain by going with old machines.
Our current Sun 4/670 probably cost something like $120,000 when it
was new.  It was meant as an enterprise server for large companies with
critical needs for reliability.  When you are building a machine that you
expect to sell for $120,000 you don't cut corners.  The chassis alone is
to die for - seems like a hundred pounds of stainless steel in multiple
layers.  When we first got it, I had to remove 25 screws just to be able
to get deep in enough to remove a disk drive.  It's cooled by an array
of nine box fans blowing a steady stream of air up through the chassis.
The power supply is similarly overbuilt, as is every other part of the
machine.  By modern standards the Sun 4/670 is a pathetic performer,
but it is totally reliable and solid.  If you want to drive a computer
into a demolition derby, I'd bet on the Sun 4/670 over typical x86 boxes
any day of the week.  This level of quality not a inconsiderable advantage.

So though I see very little reason to get any mass-market x86 boxes
second hand, there is are certainly good arguments for getting old
enterprise server machines second hand - things like Suns and Alphas.
Though you do get higher levels of reliability that way, you do pay
some price.  Typically we'd be talking substantially lower performance
and higher power consumption.  They are big and bulky (it'd take two
strong men to lift the Sun 4/670, but anyone could lift the Dell machine
I quoted above one hand).  There are fewer well-established operating
system alternatives for them.

My personal inclination is to go with a mostly new x86 system instead
of some fine old salvaged lady, but tastes vary and I'm OK with either
approach.

response 44 of 181:

Oct 14 23:08 UTC 2001

What would four faster modems cost? (33K or even 28K).

response 46 of 181:

Oct 15 10:26 UTC 2001

Assuming a constant rate of improvement in computers, why is it that 
much less advantageous to buy a 2-year-old machine than a new one, if 
the price is significantly better?  Even if you buy a new machine, in 
two years it'll be as out-of-date as the 2-year-old machine is now (if 
the initial assumption holds).

Why is grex's user base dwindling, I wonder?  Is it because the primary 
mode of access is telnet, and that's too retro for today's users?  I 
use the web interface myself: it's significantly less clumsy by my 
standards than most interactive web interfaces.  (But I'm someone who 
has never found a Usenet news reader that I like at all, so my 
standards may be strange.)

response 48 of 181:

Oct 15 13:27 UTC 2001

<knee jerk mathematical reaction>
(d/dx)(e^x) = e^x, which is not constant.
</knee jerk mathematical reaction>

response 50 of 181:

Oct 15 14:13 UTC 2001

Re #41: I would worry about buying an Alpha machine.  Why saddle 
ourselves with a platform that's already been orphaned?  We'll be left 
in the same situation we are now, having to dig and scrounge for parts 
when something breaks.  Support for hardware in open source operating 
systems also tends to be related to how new and interesting it is.  
People eventually get bored with older stuff and stop maintaining code 
for it.

I think we should seriously consider an x86 based system.  Yes, it's an 
ugly architecture, but it *works*, it's mature, and there's no sign of 
support for it drying up any time soon.  (I actually maintain two Dell 
PowerEdge servers like those mentioned in #36, that are running Linux.  
They've been very reliable, however we've only had them about four 
months so far and they aren't under the same load Grex is, so you can 
decide how much weight to attach to that.  RedHat Linux is fully 
supported on them by Dell.)

However, since we rely on volunteer staff, and I get the impression 
some staff dislike Intel enough that they'd refuse to work on an x86 
machine, that may not be an option.  I don't actually mean that in a 
judgemental way; I think that since we ask people to maintain Grex for 
free, the *most* important criterion is going to be that we pick an 
architecture and operating system that they enjoy working on.  Anything 
can be made to work if you like it enough to devote the time to it, and 
anything will fail if you dislike it enough to neglect it.

Re #44: Seeing as use of the dialin lines continues to steadily 
decline, I don't think we should invest much of anything in the modem 
bank.  It's a dead end.

response 52 of 181:

Oct 15 15:05 UTC 2001

At first my gut feeling was to suggest a RAID array, but then I thought 
about it a little.  What would Grex gain from a RAID array?  The main 
reason to use one is high availability.  In the rare event of a disk 
failure, the system can keep running until you replace it.  But being 
down while a disk is replaced and restored from backup doesn't really 
cost Grex any money.

Another factor to consider:  Will we want to co-locate in the future?  
If so, we should look for a system that can be easily rack mounted.

response 54 of 181:

Oct 15 17:01 UTC 2001

A constant rate is a constant.  Acceleration can be a constant, as a 
result of which speed keeps getting faster.  (Think of car or rocket 
speed, to make this easy.)  Not a strange form of constancy at all.  
Moore's Law is a constant.

The really strange form of constancy is the platitude "change is the 
only constant".  What makes people think change is a constant?  Often 
the rate of change increases (because acceleration is a constant).  
Sometimes change speeds up or slows down (as in the S-curve).  Some 
things don't change at all (think of the French proverb "plus ca 
change ...")