|
Grex > Coop8 > #35: Grex long-range budget meeting for Sunday, March 24 | |
|
| Author |
Message |
| 25 new of 94 responses total. |
steve
|
|
response 50 of 94:
| 
Mar 24 23:34 UTC 1996 |
True, heat is the mail problem with using a VME box. But as Rane
pointed out a while ago, we can fairly easily vent that heat into the
entire basement.
Steve, I'll give you a tour of either Suns sometime, and point
out differences between them and any AT box. Now, you may be correct
about it being overkill for a news system, but I know that in case
of a power hit, I'd rather trust a Sun VME box to survive it than a
PC. Even the newer Sun's aren't as good in this regard. The VME
chassis are *built*. There isn't really a better way to descrive
it with words. ;-)
Can we make a news machine based on an AT box? Certainly.
|
janc
|
|
response 51 of 94:
|
Mar 25 07:16 UTC 1996 |
VME chassis may be built, but last I heard people were speculating that there
had to be something wrong with that particular VME chassis, since we pretty
much replaced all the boards without stopping the crashes. Add to the cost
of the Sun 3 route the time and possible expense required to figure out what
that's all about.
If you're worried about power problems, put the Intel box on a UPS. Doing so
is much cheaper than putting the Sun on a UPS.
Venting the basement costs money up front to buy the fan, and money monthly
to power it. Add that to the equation.
Why will it make more sense to change to an Intel box later than it does now?
Grex is going to go out of business in the next year. We should be planning
for several years in advance when we spend money. Even if we decide in six
months that we don't want news after all, none of the money spent on an Intel
box will be wasted. We'll have uses for the machine (and certainly uses for
the disk). The Sun is already a white elephant. Investing more money in it
won't change that.
|
popcorn
|
|
response 52 of 94:
|
Mar 25 07:46 UTC 1996 |
Er, Grex *isn't* planning to go out of business in the next year.
That's what you meant to say, right?
|
davel
|
|
response 53 of 94:
|
Mar 25 10:02 UTC 1996 |
<looks worried>
|
ajax
|
|
response 54 of 94:
|
Mar 25 10:02 UTC 1996 |
I thought a 586/PCI system might be worth thinking about more because
of the PCI than because of the 586 part...slower 486 boards tend to be
ISA or VLB. A 486-100 PCI could be had for $70 less than a 586-133 PCI,
so that would mean $230 more than the 486-40 ISA estimates above. (Good
PCI SCSI cards are more than the motherboards w/CPU!). News servers
definitely do a lot of I/O. A faster bus and CPU could partially
compensate for the lack of a lot of RAM.
|
janc
|
|
response 55 of 94:
|
Mar 25 15:21 UTC 1996 |
Oops, misplaced a "not". My point was supposed to be that "let's think short
range now and put off thinking long range later" is not long range planning,
and Grex really should be doing long range planning.
I'd initially go with a somewhat moderate amount of memory. Since the machine
won't be doing anything but news, it may be able to keep up. Maybe we could
somehow give the responding-to-Grex-user-queries part priority over the
downloading-news-over-the-link part. If it doesn't do the job, we can always
add more memory.
|
gregc
|
|
response 56 of 94:
|
Mar 25 19:30 UTC 1996 |
You're estimates of the power consumed by the VME machine is way
off base. The current sun-3 had a CPU, 4 memory boards, an ALM-II,
2 scsi controllers, and a tape controller that wasn't being used.
Before the Sun-4 left my house, I made very accurate measurements of
it's power consumption. I don't know about the Sun-3 CPU card, but
if we switched to a 4200 CPU board like steve suggests, a VME box
with power supply and fan, and 1 4200 cpu card, 1 32meg memory board,
and 1 SCSI-3 board, uses 290 watts. If you figure that is 190watts
more than your hypothetical AT-bus machine that figures out to
$183.00 per year in extra power at $0.11 per kilowatt, not the $426.00
that jan suggests in #44.
For each additional VME 32meg memory board, add $60 per year.
With the sun, we already have memory, case, serial ports, scsi controller,
ethernet, keyboard and monitor/terminal. We could either stay with the
Sun-3 CPU for no cost, or spend $150 for a 4/200 CPU.
With an AT-bus box, I know we have the following items lying around
that belong to Grex:
1.) Case/power supply
2.) ethernet controller
3.) serial board.
4.) video board.
5.) Keyboard? (we may need to get one of these.)
We would need to buy:
1.) Motherboard
2.) 32meg (probably 64meg eventually)
3.) SCSI controller.
I think rob's estimate of $12-$13 for used SIMM ram is overly optimistic.
But I know we can get VME ram for $8 per meg.
But I'm fairly certain that an AT class news machine will cost us at
*least $650.00 in new hardware. A new Sun-4 will only cost $150. That
gap would widen if we decided to go to 64meg. Figure $1100.00 for the
AT machine or $400.00 for a Sun-4.
We could run the Sun-4 for 2.7 years (or 3.8 years) on the difference.
At the end of that 3 years we would have an obsolete machine either way,
but with a sun-4 we pay most of the money out in installments over
a 3 year period, not a big lump sum at the beginning.
|
steve
|
|
response 57 of 94:
|
Mar 25 19:41 UTC 1996 |
Thanks Greg, that sums up my feelings pretty well.
I knew that Greg had made power measurements that were better than
what I said, so the cost factor for the Sun has just fallen.
The possible problem with the Sun-3 chassis is real; forunately,
we have spares. Thats why we picked up all the Suns General Dynamics
would let us have--for spares.
Somewhere, maybe in another item Jan asked why not move now
to an AT box? Well, its pretty simple: if we can avoid them we
will be better, at least in terms of the main Grex machine. I'd
like to eventually jump to a SparcStation of some sort; that will be
possible, as prices start comming down.
But remember this people: The Sun-4/260, and espically the
Sun-4/470 were designed to be *servers*. The 16 hardware contexts on
the 4/260 card we're running now is reasonable for what we're doing,
and the 4/470's *64* hardware contexts would be just great to have,
as Grex gets bigger with more dialins/ptys.
AT boxes are not designed for this type of thing. They simply
aren't. Oh, they can be used in such a fashion, but they aren't
designed for it. Actually, the concept of more special purpose
computers seems to be fading, unforunately. But we're using
equipment that was meant for server type of use now, and thats
helped us out, a lot.
|
gregc
|
|
response 58 of 94:
|
Mar 26 00:11 UTC 1996 |
OH! One other thing I forgot to mention, and I think this is actually
much more important than power usage. One of the prime reasons I would
like to see the News machine be a Sun-4 VME machine is that it can also
act as a hot backup for the primary Grex machine.
If something bad happens to the current Sun-4: Power supply dies,
backplane cracks, CPU blows up, or the machine just developes a
general undiagnosable flakiness, we could just shut both machines
down, move the ALM-II card and 2 memory boards to the News machine,
disconnect the news machine's disk enclosure, move the main Grex disk
enclosure over to the news machine, and we'd be back up and in business.
News would always play a secondary role to the prime function of grex itself.
The above has already happened to the Sun-3. In addition to our long
running problem with The-Disk-Bug, the Sun-3 developed a spontaineous
reboot failure mode about 3 or so weeks ago. For the last 2 or 3 weeks
of it's life, the Sun-3 was rebooting 4 times a day on average.
Steve and I tried a couple of quick fixes, but nothing seemed to help.
It was difficult to trry to do any detailed analysis on a running,
production, machine. Since then, I've concluded the power supply
on the Sun-3 has developed a glitch. Changing out a power supply
on one of those cabinets is not an easy task, but if we'd had
an entire backup system available, we'd have switched to it weeks ago.
,
|
janc
|
|
response 59 of 94:
|
Mar 26 04:50 UTC 1996 |
I haven't advocated moving the main machine to an AT machine anytime recently,
and don't expect to start any time soon.
If these power figures are accurate, then the case for a Sun news machine
has certainly been greatly improved.
If we go with the Sun, plans for improving dungeon cooling *must* be included
in the plan. It doesn't appear that we are going to be able to run both
machines in there in any kind of warm weather, much less in the summer. Did
the fan plan calculations count on two Suns?
|
ajax
|
|
response 60 of 94:
|
Mar 26 08:04 UTC 1996 |
Indeed, $200 a year in power is much more palatable. Greg, if you haven't
followed memory prices for the last month, you'll be surprised! This week
I've seen even new 4 meg SIMMs for under $12/meg, though $15/meg is more
common. The table next to us at the JCC sale was selling some new 72-pin
SIMMs for around $15/meg. Sun memory prices seem to drop in proportion to
new memory prices, so I imagine your $8/meg estimate is on the high side.
I saw some 1 meg 100ns SIMMs yanked from Sun 3/60s selling for $6 each, for
example.
|
steve
|
|
response 61 of 94:
|
Mar 26 15:24 UTC 1996 |
Right--as new memory costs plummet, so will Sun memory cards,
although probably not as fast as the new stuff. But the Sun cards
have been cheaper for quite some time, and is likely to remain so
for a while.
Something to think about in terms of the cost of running things:
I don't care to think about the future in excess of more than two
years down the road, because of one simple fact: that *whatever*
we do now, in two years other things are going to be sufficiently
cheaper that it is likely to blow away what we're doing now. If
we spend an extra $600 for an AT box for news to save $20/month on
electricty, in that two year span of time we haven't really saved
anything. I think that, in the end of a two or three year time
frame, we're going to wind up spending basically the same amount
of money: more up front at first for an AT box using less power,
or less up front and more monthly cost with a Sun.
What I'd *really* *like* *to* *see* is our getting little
low power SPARC machines and use those for routers, kerberos,
MX forwarders, mail machines, etc. There is comming the time
when things like SparcStation 1's, Sun-4/110's and other things
will be given away. At that point I would dearly love to have
us get our hands on them. The advantage would having a single
type of software platform that runs everything. We're getting
close to the point where we can do that.
|
srw
|
|
response 62 of 94:
|
Mar 27 03:19 UTC 1996 |
STeve, what Rob is saying about lower memory prices for SIMMS chops
a great deal off the difference in up-front cost. I'm really surprised to
hear Greg say we can run a Sun-4 News machine for $243/year at 32MB in power
costs. I trust Greg to be right on these things, but I am surprised it is so
low.
I trust Rob Argy on the SIMM memory prices though, and they bring the amount
to be made up back to a smaller amount.
If we found that 64MB was needed, this means an even faster payback.
I am concerned at our choice to spend our money on Detroit Edison instead of
hardware. I have no argument to counter the hot backup argument except that
the backup machine could be left powered off. It wouldn't be as hot, but it
would probably be just as effective when needed, if tested occasionally.
Admittedly this adds the cost of a redundant Sparc card if we run news on
intel.
I guess I want to see the numbers one more time, after we can agree to them,
before I can say that I'm totally satisfied.
|
steve
|
|
response 63 of 94:
|
Mar 27 03:41 UTC 1996 |
Well, it all fluctuates up and down. SIMMs do seem a little cheaper
now, but then again I've just heard of sun 8M cards for $70, and 32M
cards for $150. I'll have to inquire about them.
But still, beyond the electrical costs, there are other factors
such as software plaforms, and their stability. SunOS is among
the best, and while I know its easy to overlook that, thinking
that just becuase there aren't any human users on the machine doesn't
mean that it won't take a heck of a beating in the daily discourse
of giving people news and taking in new things constantly.
Remember, if we don't pay Edison more money, then we'll be
paying a computer store (or whoever) more. I do think that in
the end the costs are going to be fairly similar. Also, in
going with the Sun, once things like SparcStations are cheap
enough, we won't have to recompile the news world--we can just
like a new CPU beneath the disks and be running once again.
|
janc
|
|
response 64 of 94:
|
Mar 27 06:26 UTC 1996 |
Basically there is almost nothing Grex spends money on that doesn't
immediately go up in smoke. Computers become worthless so fast that spending
money there is not noticably better than spending it on electricity. The UPS
is one of the few things that we are likely to buy that will be worth anything
a few years from now.
I remain unconvinced by the stability arguments. I think if you compare the
reliability and stability of M-Net's hardware to Grex's hardware over the
last few years, you are going to have a hard time convincing anyone that
we did better with Sun/SunOS than they did with Intel/BSDI, especially if you
factor in the hours of staff time spent on each system. I don't think,
however, that stability is the driving issue here though. Economics probably
is.
For both the UPS project, and the cost assessment of the usenet project, we
need more reliable figures on power consumption. It would be really good if
someone could make some real measurements sometime soon.
|
gregc
|
|
response 65 of 94:
|
Mar 27 07:06 UTC 1996 |
Ok, a bunch of points here:
1.) Srw, we currently own 5 32meg memory boards, of which 4 are in the Sun-4.
So, for the initial machine, going with a Sun-4 for news means 0 cost
for memory.
2.) Also srw: If we agree that it's a Good-Thing to buy a spare 4200 CPU
board for Grex, then that spare CPU can be *used* in the news
machine instead of collecting dust on the shelf. This also means
that if the board has to be bought anyway, then the CPU upgrade
for the news machine effectively becomes a 0 cost item if we go
the sun-4 route.
3.) Jan, you probably don't remember this but, durring the time period
from January 1994, when we first went to the Sun-3, until July 27th
1994, when we installed the big /home disk, the Sun-3 was *very* stable.
We had uptimes up over a *month*. SunOS is normally a very stable
platform. All of our problems have been realated to *1* nasty bug:
The-Disk-Problem. That problem is gone now. A Sun-4 running 4.1.3
is even more stable. The current machine has *NEVER*, since the day
I first installed the OS back in August of 1995, experienced a panic,
a spontaineous reboot, a hang, or any kind of crash whatsoever.
I think you are going to be amazed at the uptime we will experience
over the next months.
4.) Also jan: Maybe you missed this in the other item here, but I alread
*made* "real measurements" of the Sun-4's power consumption. A number
of different ways, with a number of different instruments. You can
bank on those numbers. What do you mean by "more reliable"?
Do you want a signed affidavit from UL labs or something?
|
steve
|
|
response 66 of 94:
|
Mar 27 14:44 UTC 1996 |
The future uptimes are going to be very much influenced by the
times we have to shut things down, not how often we die. There are
reasons why the "Use SunOS 4.1.3 or die camp" have caused enough
troubles in Sun that Sun finally released SunOS 4.1.4, even after
supposedly switching to Solaris.
|
n8nxf
|
|
response 67 of 94:
|
Mar 27 15:22 UTC 1996 |
(UPS's have change quite a bit in the last few years. There are now
inverters [part of a UPS] on the market that reproduce 60 Hz sine-wave
power very well at a cost less than square wave inverters of 10 years
ago. Some electronics will not tolerate the output of square wave
inverters. More than one laser printer power supply has been destroyed
by square wave inverters for instance. Most of the cheap UPS's use
square wave inverters and most *modern* computer equipment, with their
switching power supplies, can tolerate it.)
|
gregc
|
|
response 68 of 94:
|
Mar 27 18:23 UTC 1996 |
By the way, it's pretty silly to put a laser printer on a UPS.
Both from a technical standpoint, they have *big* current spikes, and from
an operational standpoint, is it *really* necasary to keep printing
during a blackout?
|
steve
|
|
response 69 of 94:
|
Mar 27 19:21 UTC 1996 |
Some people think so--I've met them. They were silly however.
|
srw
|
|
response 70 of 94:
|
Mar 28 01:33 UTC 1996 |
I don't dispute the stability of the Sun platform as a fact. I dispute it
as a good reason to use a sun platform for a news machine. By saving on
capital expenditures you ensure that the operating costs will remain higher
throughout the life of the server. The justification for the extra cost is
to have an extra stable news machine. I question that logic.
I don't want a machine that crashes all the time, but I am not convinced that
an intel machine running inn will crash as much as gryps, which is running on
an old version of FreeBSD. Greg uses FreeBSD a lot for routers, why not also
for inn? I am not convinced by the hardware contexts argument here. I am
certain a 486DX33 can handle this job far into the future.
If it costs more in the long run to run on a sun, I question whether that
cost is well spent on a news server. We can use the money for other things.
I would rather justify the cost of a hot backup machine separately, but the two
are tied together. It might be cheaper to let the backup cpu stay turned off,
because of the money you save in electricity by running news on intel.
I don't have time to go back through this item and extract the costs we decided
on (or did we decide). I mean Greg's power numbers and Ajax's memory prices.
I wonder where we stand in the comparison based on these.
|
janc
|
|
response 71 of 94:
|
Mar 28 02:43 UTC 1996 |
Re #65.4: The reports I heard on your measurements sounded like you'd
acturately measured things in a configuration different than what is being
used, and people were filling in the the gap with more wild-assed guessing.
I may be wrong.
|
gregc
|
|
response 72 of 94:
|
Mar 28 06:28 UTC 1996 |
Ok, jan, I understand. You're sort of right and wrong.
1.) I do have accurate measurements for the configuration that would
be used for a news server.
2.) People have also been doing some wild-assed guessing.
|
n8nxf
|
|
response 73 of 94:
|
Mar 28 14:55 UTC 1996 |
I did some serious testing of UPS's for a medical application several
years ago. Power Var was the best when it came to handeling weird line
transiants without causing the CPU in our instrument to reset. They also
make a non medical line of UPS's and were of the switch over type. They
were also one of the cheaper models I tested. Unfortunatly I don't have
a phone number for them. The Power Var unit outpreformed a high end unit
costing more than twice as much. (I'll never forget the look on the sales-
mans face when I described my test to him and the results. I never heard
from him again.)
|
steve
|
|
response 74 of 94:
|
Mar 28 17:08 UTC 1996 |
As I said before, I posted electrial values that I knew to he
high, knowing that Greg would enter more accurate values later.
Steve, maybe I haven't said this well enough in the past, but
the cost differences between an AT box and a Sun box shouldn't be
looked at farther out than about two years. Becuase at that point,
no matter what we've done, news is going to be so much bigger than
it is now that we'll have to do something else in order to keep up
with it.
If news growth wern't an issue, then yes: the AT box would win
out, assuming it was stable enough, just over the electrical costs
alone. But we don't have enough AT parts around at the moment, and
we do for the Sun.
|