|
|
| Author |
Message |
andyv
|
|
planning/finances
| 
Dec 21 17:58 UTC 1994 |
I would like to get a discussion going about specifics concerning
plans for Grex's future, especially its financial future. I would
like to here about the gripes people have about the way the services
are rendered and the wants and the fears. Of course, a good deal
of information put up for discussion should come from the people on
the inside of operations here.
Andy V
|
| 171 responses total. |
andyv
|
|
response 1 of 171:
|
Dec 21 18:03 UTC 1994 |
Since this is the first time I have entered an item here on Grex,
I decided to see what I had wrought. One thing, I don't seem to know
the difference between here and hear. Sorry ;-)
Are there any Electronic Engineers on board who would know how far into
the bath tub curve most of the equipment is in? Is anyone keeping track
of how many hours of operation is on each piece of hardware?
Knowing that would help us plan some of our future needs.
|
steve
|
|
response 2 of 171:
|
Dec 21 18:36 UTC 1994 |
Since most of the equipment we have is old, we don't know how many
hours of operation a lot of things have had. I can probably break things
down a litte better:
- modei: 5 dial in that we purchased in July of '91, each with a five
year warrenty. One was a donation in '93 so we don't know how many
extra hours that one has.
- Sun-3/260: unknown how many hours the CPU disk two 330M SCSI disks
have on them. The 32M cards (one of which is Marcus's) are not new
as well as the SCSI-3 card. The ALM-II serial card is also unknown
as far as previous usage goes.
- The 2G disk and shoebox are new and in operation as of July 1994.
The disk is 350,000 (or 500K?) hours MTBF. Don't know what the
MTBF is for the showbox but its certainly the power supply that will
go first.
- The Internet link modei are a pair of Hayes optima 288's; they were
purchased in December '93. They have a five year warranty.
- The 9 track tape drive is an old 1600bpi unit, number of hours
on it unknown. Has occaisonal errors on tapes. The board has
authorized a new 8mm (or 4mm maybe depending on price?) unit, and
one was found December 20th but we don't know if its been sold yet.
So, most of the things we have are old. The most likely thing to die
are the old 330M disks, but Wren IV's have had a good reputation overall,
and there is a good chance that they will last several more years. Given
that there are many Sun-3 systems out there with more than a decade of
continous use, I don't think anyone can really say what the life span of
them is. They're built considerably better than todays units in terms of
physical/machanical things, so there is likely to be considerable life left
in them. The power supplies on the 2400bps modei have started to go (two
of them failed within two months) so a new set was gotten from Supra. The
Hayes link modei have been extremely stable themsevles, with the only problem
I can think of being that the one in the warehouse went a little wild a couple
days that the temperature went above 97 degrees.
I think the state of the hardware can be said to be pretty good. The
state of the software leaves something to be desired, in that we're still
having the disk instability problem. Marcus made an interesting observation
that the disk problems might be lessened becuase of the lower load average
on the system. Since *two* fsck's of /home resulted in no files being
lost at all, I think he is right. So while we haven't solved it, it is
better.
After we've moved into the Dungeon and have gotten the disk problem
solved we can start the debate of what to do next for future versions of
Grex's hardware. There are two distinct paths to take: 1) stay on the
trailing edge of systems and go with a SPARC (which can be had for $500)
and eventually move of the Sun-3 to a SPARCStation (they really are comming
down in price) and keep upgrading SPARC machines as new table scraps become
available.
2) is to abandon the Sun line for the ever popular 486 box, which
means going to different everything (ie, more initial cost), but has
the advantage of hundreds of sources to buy from.
|
andyv
|
|
response 3 of 171:
|
Dec 21 21:31 UTC 1994 |
Given your familiarity with the hardware, how much money would you be
comfortable with in an account specifically for hardware upgrades or
replacements? since we have to pick up on buys while we can, would you like an
account to buy items you might come across that we need or will need?
|
steve
|
|
response 4 of 171:
|
Dec 21 22:24 UTC 1994 |
Speaking as just one staff person who deals with the hardware, I'd like
to see about $1000 for two SPARC CPU cards, $500 for another 32M card,
$1700 for two 2G disks, $600 or so for a terminal server, $800 for eight
14.4Kbps modei, $400 for misc stuff. Thats about $5000 which would give
us a faster Grex (using a Sun-4/260 card), A second Sun system for news/
mail, more disk on the 'main' system and a terminal server to handle fast
modei. Thats using the "sun" model of contiuing to use trailing edge
technology.
Using the "486" model, we're probably looking at $1040 for 32M,
$700 for a 486 motherboard, $200 for a case/power supply, $200 for
a VGA card/mono monitor, $250 for a SCSI controller, $300 for 600M
disk, $200 for a SCSI CDROM, for the system itself, making a *rough*
estimate of $3000 to recreate our current Sun-3 system minus the 2G
disk. Also, the $3000 does not cover the cost of a terminal server
or intelligent serial cards for modei to attach too.
|
mdw
|
|
response 5 of 171:
|
Dec 22 00:16 UTC 1994 |
What I am in favour of is keeping Sun or equivalent hardware for the
login machine(s) & anything else users can program, while using
386/486/whatever technology for routers, servers, and other components
of the system that users don't have direct access to.
I've long had reservations about the reliability and ruggedness of PC
equipment; with care it is possible to find stable hardware; but not an
automatic given. It is pretty clear the PC market will tolerate
unreliability that would be unacceptable for Grex: the success of
windows proves that, as well as the many relatively flakey modems that
have enjoyed success through the years.
More recently, as I've investigated the 386/486 systems programming
world, I've formed the distinct notion that it would be dangerous to
just assume the pentium & future chips don't have a bug that would allow
a clever programmer to break system security. On the 286 thru 486,
various "oversights" in the chip's protection model have allowed clever
programmers to do things like demand paging in real mode, or access
extended memory in 16-bit mode. On the pentium; intel has started
*hiding* significant details about the architecture of the chip; opening
the possibility that it might be impossible to implement a fully secure
OS on the pentium or future chips without paying big bucks to intel, to
find out all the secret details of the chip. Chip designers have long
been known for hiding artistic flairs, such as their telephone number or
initials, in the chip's mask. Software game authors have been known to
hide trap doors or special tricks in game software. I worry an intel
chip designer might do the same; and whereas with IBM or motorola, I
have some confidence this would be found out, or the chips recalled if
necessary, with intel, I fear they'd pay the bastard off and the first
we'd find out would be when some system cracker runs the latest
underground surprise on the system. I had formed all of these notions
*before* the Pentium floating bug was discovered. The pentium bug has
only reinforced all of my impressions of intel; and that we should be
very careful where and how we use intel equipment.
|
tsty
|
|
response 6 of 171:
|
Dec 22 02:40 UTC 1994 |
Long live the curious!
|
andyv
|
|
response 7 of 171:
|
Dec 22 03:11 UTC 1994 |
I am a junk collector from way back and I have gained lots of respect
for older equipment which keeps plugging along. The security issues
raised by you Marcus are enough to make one shudder. Besides, the system
we have is a good starting point and proven.
Marcus, what is ball park figure in $$$ you would feel comfortable with?
Is Steve's $5000 figure high or low?
Steve, what time schedule would you like to have for your items? Grex really
seems to get bogged down with the load. Is carrying the load more effectively
a priority item? What item is #1 on the list?
|
mdw
|
|
response 8 of 171:
|
Dec 22 05:30 UTC 1994 |
I don't see hardware upgrades as a one-time project, but a continuous
process - so I'm not entirely sure the $5000 figure is meaningful. I'm
sure STeve's figures are accurate, so if an anonymous donor showed up
tomorrow and said "I will pay to upgrade your system to a Sparc" that's
how much we'd ask for, and that's what we'd spend it on. That is also(
about the minimum we'd have to spend to switch to 486 technology -
STeve's $3K didn't include a tape drive or terminal concentrator, or
software--and if we were setting up that kind of system we should be
seriously considering BSDI.) However, one of the advantages of our
current Sun hardware is we have a fairly compatible upgrade path and can
space things out a bit - any one of those things could be bought and
used to advantage in the current hardware environment. Some further
factors to consider are that as time goes on, all of these things are
getting cheaper, and there is a good likelyhood that we'll frequently
discover and want to resolve bottlenecks that we hadn't anticipated.
What I think makes the most sense is to plan on a fairly constant income
stream, and then plan on making periodic purchases of new equipement
from this stream, based on what the most pressing bottlenecks are, and
to a lesser extent, on any good buys we happen to run across. Where
planning can help is, in terms of making sure we stay on a hardware path
that fits in with this continuous improvement process (ie, we probably
don't want to price out a pa-risc machine unless we get a *very* good
price), and identify if we are running low on memberships and ought to
be more aggressive in terms of letting people know that we need
memberships.
|
steve
|
|
response 9 of 171:
|
Dec 22 05:41 UTC 1994 |
While I'm not nearly as concerned as Marcus is about software weirdness
in Intel CPU's, such things could happen I suppose. Actually, that happened
once with a controller chip made in the 70's... What *does* bother me
about the PC world is the simple fact that the hardware is designed under
conditions that Sun never had with the older stuff: 1) Because of the
large expense of a Sun-3/260 (more than $30,000 originally), adding things
like a thicker case made much less of a difference to the bottom line in
the PL statement with that type of equipment. So, the Sun stuff frequently
has saftey interlocks, mechanical stabilization and overall rugedness
features that the PC class systems just don't have. 2) The technology is
changing so bloddy fast in the PC world that manufacturers of motherboards
are now routinely planning on *six month* useful selling life! Things are
changing that radidly. So, corners get cut. Here and there pennies get
shaved, sometimes with bad results. I now know of a motherboard manufacturer
who took out a 3-cent (0.03) comparitor chip off the cache control circuitry,
such that the cache was refreshed when it should not have been, and was thus
10% to 20% slower than it should have been. 3) Because more than 1000
companies are creating competing equipment, not all of them get it completely
right. Add to the fact that the 1000+ motherboards can mate with 10 different
SCSI cards and 80+ VGA cards, and you can produce combinations that Just
Don't Work Right. If you care to see an example of this, read Sun's manual
for Solaris 2.4 X86, for the PC. The manual lists all these different little
"problems" of what not to use with which. Sun has done a *beautiful* job of
this, but it kind of doc just doesn't exist with their SPARC line. 4)
Quality control, untill lately has been done by the customer as much as at
the factory. I don't know how many times I've helped someone with a problem
only to find out that the was a hardware patch for the card I was working
on, such that we had to get a new one to fix the "problem".
Boy, I'm getting good at rambling these days. ...Uh, just how big is
a "keats" of text?
|
steve
|
|
response 10 of 171:
|
Dec 22 06:02 UTC 1994 |
Heh. Marcus and I attacked this item at the same time. ;-)
What would I like to see first, Andy? Well, we need to get a Exabyte
tape drive. This uses 8mm "video" cartridges that can store 2.1G of data.
The board has authorized us to get one, and we're looking right now. In
fact, I just found a person in Ann Arbor with one they wanted to sell, so
we'll see if they respond quickly. I didn't mention this in one of the
above statements because its something we're already going to get, and
if we get to future Sun stuff or 486 stuff we'll still be able to use it
since its a SCSI device. Marcus is right that there are a few other things
I left off the list--I plead hardware fixation while I wrote that. If we
were to go to BSDI for example, we'd pay about $1000 for the license with
source (which we do need).
As for a "time line", I'm going to stick my neck out a little, and
assume that Marcus is willing to be in the thick of things. ;-) For
$500, we could get a Sun-4/260 card, which is a SPARC CPU card that is
completly compatible with our current hardware. Putting that CPU card
in Marcus's system would let us bootstrap ourselves, in that we could
compile the half-million little problems that we'd need, from party to
PicoSpan to write, and all the other little things we use, but would have
to recompile becuase we're changing CPUs. Once we got things running on
the SPARC Sun-3 (well, actually its a Sun-3 body with s Sun-4 "head"),
we could then move everything over to Grex in a long painful day (save
the world, tear it apart, rebuild with SPARC binaries). Doing this would
give us a system that is 2 to 2.5 times faster on average. When (not if)
we outgrow that SPARC system, we have a fork in the road: 1) a Sun-4/300
style card or 2) a SPARCStation or similiar. With 1), we could get
the CPU card for about $1000 and compatible memory for $2000 or so--
*But These Prices Are Changing* and the cost for them 1+ years from now
is likely to a *A Lot* different than now. 486's are pushing the prices
of older Sun stuff everdownwards. Option 2) is to get a used SPARCStation
of some sort, which would be faster than our first SPARC CPU card, by
a factor of two, if not more. ...It's all a function of prices. $20,000
SPARC systems are now becomming available now, just like the Sun-3 stuff
did in 1991: down, down, down.
Paralell to all the above, is a second machine to run mail (and maybe
news). Since mail is turning into an incredible resource nuncher here,
we're going to have to do something soon. Now, weather using another
Sun-3 is better than a 486 box is a good question, and bears more thinking
talking, squabbling and arguing. ;-)
The Sun offers us a way of more cheaply getting CPU horsepower than
anything else, given a limited amount of money. Now, if we had a Grex
Sugar-{Daddy,Mommy} who gave us several grand to play with, things might
be different.
But it is important to remember that whatever we do, it will take
some amount of doing to switch over to the next generation of Grex
hardware. Greg spent *quite* a lot of time getting everything together
on this system, and since he did it all alone at his place, its all to
easy to forget the amount of work he did there. Bearing that in mind,
the effort to move over to the SPARC would be a one-time effort that would
let us jump from SPARC to SPARC, limited only by the hardware Grex can
get donated to it, or however else it might magically appear. The same
can be said of 486/Pentium/P6 technology, but those systems simply aren't
as fast. I'm not saying that we shouldn't go to a 486 system just because
they aren't as fast, but when ya compare the two, the SPARC line definately
gives more room for expansion.
|
cicero
|
|
response 11 of 171:
|
Dec 22 06:25 UTC 1994 |
Don't we already have a couple more sun 3s? (I thought I read that somewhere
) Can one of those be used as a mail machine? What about the old sun 2? Or
is that too slow for any purpose?
|
steve
|
|
response 12 of 171:
|
Dec 22 07:31 UTC 1994 |
I'm afraid the Sun-2 is best used as a door stop. Yes, we have a Sun-3/260
that can get turned into a mail machine. I didn't mean to imply that we'd
have to get to a SPARC for a second machine. For another Sun-3 to get
pressed into service, we'll have to get a disk, and another SCSI-3 controller.
Last SCSI-3 I got was $125, and the $950 2G disk is now something like $830.
So its definately possible.
|
tsty
|
|
response 13 of 171:
|
Dec 22 12:54 UTC 1994 |
A keats of text is two screens' full, approximately 46 lines. The
term was coined by mythago, and quantified by tsty.
|
popcorn
|
|
response 14 of 171:
|
Dec 22 13:49 UTC 1994 |
When I read #0, the one big financial concern that came immediately to mind
is: TAXES. I'm steadly more concerned that we need to find out what, if
anything, we should be paying, and, if it's more than $0, pay it.
|
rcurl
|
|
response 15 of 171:
|
Dec 22 15:47 UTC 1994 |
I think we should determine that, too. Whatever the requirements are
will drive other poplicy decisions and actions.
|
andyv
|
|
response 16 of 171:
|
Dec 22 17:18 UTC 1994 |
The object of a non-profit is for the gov't to get a need service for the
public at the cost of forgoing some revenue. It works and the state is
not hard to get along with unless you deviate from this. Money held to
replace equipment or to upgrade isn't a problem. Has anyone out there
been hassled by the gov't for putting some money in the bank? I'm going
to call the coordinator of United Way here for my own info.
Back to the financial planning. I wanted some estimates of money
needed for an imagined static situation because it is easier to get
input that way at first. A begining. Of course we shold take that
and fit it into a picture of dynamic situation. At least three
things have to be happening at the same time over time.
What hardware is needed before this time next year? Then we put that need
in stages which should be attained over the next year. With that info
we can make preparations for the change i.e. find ways to meet those
needs.
What is the best time of the year for all concerned for a major change?
Fund raising is usually slower in the summer. Has that been the case
here too? Is summer a time when there is more spare time for implemetation?
|
cel
|
|
response 17 of 171:
|
Dec 22 20:08 UTC 1994 |
here's about $0.02 worth wrt costing out hardware alternatives.
a few years ago, the U-M computing center collapsed its 2 MTS systems
into a single MTS system. a big reason for this was that maintaining
several systems with slightly out-of-sync OS versions was a pain. i
see a parallel between this situation and grex staff choosing to run
more than one type of hardware (Sun3 + 486 for news/mail). main-
taining such a situation might be confusing to staff and more trouble
than it is worth, given limitations of staff availablility.
an advantage to having all the systems (terminal servers, news, and
mail machines) use the same hardware is the usefulness of spare parts.
you could keep one CPU board, say, as a spare for all of your machines.
if you used several different hardware configurations, spare parts for
one machine might not be compatible with any of the others.
finally, if you are planning to use 8mm carts for backup, you should
carefully design your backup scheme. although 8mm carts can store
gigabytes of data, they are not as reliable as some other forms of
storage. WORM is certainly an option for convenient and reliable backup.
it might be reasonable to consider off-site storage for some
of your backups. if the dungeon is a basement anything like others
i've seen here in the OWS, you will not have an optimal ambient
environment for data storage, especially in terms of moisture.
|
tsty
|
|
response 18 of 171:
|
Dec 22 22:39 UTC 1994 |
From resp #10:
What would I like to see first, Andy? Well, we need to get a Exabyte
tape drive. This uses 8mm "video" cartridges that can store 2.1G of data.
The board has authorized us to get one, and we're looking right now. In
fact, I just found a person in Ann Arbor with one they wanted to sell, so
we'll see if they respond quickly.
Well, if they act that fast, what would the cost be, approx?
Can they beat this SCSI Exabyte 8mm Model EXB-8200-B03 that i have
in my hands right this minute? Wasn't willing/able to mention this
until I had it in *my* fingers. I do. Want it?
|
steve
|
|
response 19 of 171:
|
Dec 22 23:25 UTC 1994 |
We're interested in looking at it, thats for sure! Is it in a case
or what?
|
tsty
|
|
response 20 of 171:
|
Dec 23 02:43 UTC 1994 |
It's original intent was to be an internally mounted device.
It was removed, working, from service a couple months ago and
replaced with something bigger/better/larger/whatever.
More than that I can't state. There is a list of "things" on
a lable on the 8mm drive. Here are the buzz-acronyms, lemme
know which ones (maybe all) you want the numbers for:
DR, MX, MXPROM, DF, CD, RW, SV, SVPROM and DECK.
What was the cost for this other one, btw?
|
steve
|
|
response 21 of 171:
|
Dec 23 03:51 UTC 1994 |
About $400 approximately. Numbers for all those things would
be good, espically the prom numbers.
|
danr
|
|
response 22 of 171:
|
Dec 23 04:52 UTC 1994 |
I'm going to turn my attention to taxes right after the first of the
year. I don't think we have a lot of liability, but I could be wrong.
Andy, we've handled the financing of most major hardware upgrades
by holding a fundraiser. That is, we've asked members and users to
donate money over and above any membership dues. This is the way we
paid for our 2.1 Gbyte hard disk. Only recently have we been able to
pay for these items from the general fund.
I expect we will be financing major purchases (anything more than $1000)
this way in the future, or at least in the near future. We are
currently bringing in about $450/month in membership dues. When we
move to the dungeon, our expenses are going to be around $300/month.
At the max, that leaves only $150/month for hardware purchases.
We seem to have hit a plateau as far as membership goes, too. We've
been hovering around 90 members for a while now. So, I think for the
near future, we are going to have to rely on the largesse of donors for
the money we will need to upgrade our hardware.
|
mdw
|
|
response 23 of 171:
|
Dec 23 05:20 UTC 1994 |
Cel does have a point - we certainly don't want to end up with a mix of
386, 486, pentium, sun-2, sun-3, and sun-4 equipment, and try to keep up
with all of it. But I think it's reasonable to try to support 2
architectures: sun-3/sparc for the login machine, and 486 for most of
the servers. The reason to do that is the sun-3/sparc are power sucking
monsters, and peripherals, while of better quality, are more expensive.
Since the servers would not be accessible to the unwashed masses,
protection defects in the MPU or loopholes in rdist are much less
important.
That will greatly simplify the job of system software maintenance
on those machines.
|
steve
|
|
response 24 of 171:
|
Dec 23 05:31 UTC 1994 |
Thanks for the comments, Chuck. The advantages of common hardware
aren't lost on me--thats why I listed $1000 for two SPARC cards, just
to keep the two possible Sun-3 cum 4 systems the same.
I'm curious as to your comments on backups though: From all the
evidence I've seen, the biggest problem with exabyte were in early
model heads, which did have some reliability problems. But the later
units have provem themsevles reliable, and certainly they are the only
tape units around that can demonstrate 7 year storage abilities.
Remember that cost is a big factor here--the 8mm exabytes are being
pushed out in favor of things like WORMs and 4mm units, so they're
finally turning into table scraps, which we can afford. $10 a tape
we can afford.
|