|
|
| Author |
Message |
| 25 new of 127 responses total. |
lilmo
|
|
response 100 of 127:
| 
Feb 21 02:28 UTC 1995 |
Speaking as a (relative) layman, I would just likee to say that I would be
leery of using ANY 586... I would feel much better about a 486/66 that had
been used and found reliable already (sorry about that line overload)
|
mdw
|
|
response 101 of 127:
|
Feb 21 03:49 UTC 1995 |
Actually, at home, I've been using a windows system to dial in - and,
actually, it's proven to be surprisingly reliable - no surprise crashes
yet. Of course, it's also not doing anything interesting - just
"terminal" (which sucks, but that's another story), and solitaire
(which, shamefully, I'm addicted to.)
At the same time, I know people who have had enormous and scary problems
getting PC machines to work reliably. And, I have noticed, without
exception, the poeple who care about reliability do not buy the cheapest
brand X hardware, toss it all together, and expect it to work. On the
contrary, they are extremely name and brand conscious. They know a
configuration that they trust, and they are extremely reluctant to
experiment. They have good reason for this fear: they can tell you
controllers, disk drives, and mothers boards that will not, for
instance, run netware reliably. They'll run DOS and windows just fine -
but they fall apart when asked to deal with the heavy multi-processing
load of netware.
I'm not saying PC machines are inherently unreliable. What I am saying
is that is that the *industry*, as a whole, is willing to tolerate
unreliability we would consider unacceptable. And, to get tolerable
reliability, we will have to spend more time & money, and accept more
risk.
But that was really only half of my argument. The other half concerns
people who have *unfriendly* interests in the system. The ability to
run netware doesn't count much at all here; netware servers generally
abend when an NLM doesn't play by the rules.
|
gregc
|
|
response 102 of 127:
|
Feb 21 04:31 UTC 1995 |
Ah, ok, I'll agree with you there. Let me restate my position:
It's possible to put together stable PC based hardware. From a design
standpoint, it's not inhereently unstable.(we'll leave the argument about
it being a crappy design from a performance standpoint to another day.)
But as long as you select *good* quality components, you can get a stable
system. Unfortuneately, there is alot of *crap* out there. Components that
don't conform to spec all that well. Put a system together with crap and
you *will* get a crappy system. You are right, this is alot of the problem.
Everybody touts the fact that their PC is cheaper than so-and-so's PC.
Cheaper isn't cheaper when you factor in lost data and lost time. That's
why, when I put together a PC based system, it ussually costs twice what
you'd pay for a similarly equiped system from Joe's discount computers,
but *my* system works, and it works with less common OS's like UNIX or
netware or what-have-you.
|
sidhe
|
|
response 103 of 127:
|
Feb 21 05:17 UTC 1995 |
So, are we saying I should reconsider this Packard-Bell P75 system,
here in this Best Buy ad, because it's available for US$1,995.00?
|
carson
|
|
response 104 of 127:
|
Feb 21 05:21 UTC 1995 |
That cost shouldn't be the prime motivating factor in your purchase.
|
steve
|
|
response 105 of 127:
|
Feb 21 06:28 UTC 1995 |
Definately. I would not recomend getting a Packard Bell system--
they are "flakey" just a little too often for my tastes. These computers
are sold very much on a volume basis, even moreso than say, Gateway. A
typical customer of Gateway has slightly more knowledge of things than
someone walking off the street into a Best Buy who accidently buys a
computer when they were originally looking for a vacuum cleaner.
I've helped three people now, getting their Packard Bells to work.
While I can't complain about the money I made from those consulting jobs,
half the reason I was there was to help with the "flakyness", which is
this ineffable quality that I can't quite describe, but can definately
ascribe to lots and lots of PC hardware.
|
gregc
|
|
response 106 of 127:
|
Feb 21 08:15 UTC 1995 |
I have several rules that I use when putting systems together:
1.) Buy recognized brand names for hardware, if it's an import knockoff
with either no name or some name you've never heard of, avoid it.
About the only exception to this is for serial/parallel I/O boards.
It's almost impossible to find a brandname for those items, the entire
market is almost 100% cheap foreign imports. Fortuneately, it's almost
impossible to screw up that kind of card.
2.) Always buy Teac floppy drives.
3.) Buy a power supply with 300watts. Almost all the cheap foreign knockoffs
are 200 or 250watt supplys. A 300 watt unit is ussually made by a
better company. Make sure it has a UL listing. Even if your total system
requirement is 150 watts, the added buffer of a 300watt unit will buy
you increased reliability.
4.) Stick with known motherboards. I always use either AMI, Mylex,
or Micronics or Intel. Those brands are pricier, but they *work*.
5.) Use Northgate keyboards.
6.) Stick with name brand video. ATI, Diamond, Hercules, Number Nine, Matrox,
Orchid. These companys have been making video boards for years, it's
their main business and they know how to make boards that work.
7.) Use either an SMC or a 3Com ethernet board. Same reasons as #6.
8.) For disk controllers, use either Adaptec or Buslogic for SCSI, for
older MFM or ESDI, Western Digital boards were the *only* way to go.
Unfortuneately, for IDE controllers, there doesn't appear to be any
name brands. They appear to fit into the same category as item #1.
9.) Disk drives are harder. For SCSI the best rule is to be sure it has
a 5 year warrenty. Micropolis, Conner, Western Digital, Maxtor, and
especially Fujitsu and HP are good drives. Fuji's and HP's are rugged
drives, but you'll pay more for them.
A.) Memory is tough. Make *sure* you get the speed recomended for your
motherboard. Hitachi, TI, NEC, Toshiba, are names to look for.
B.) If you want a sound board, use a SoundBlaster, Adlib, Turtlebeach,
or MediaVision board.
C.) Stay away from QIC-40/80 drives for backup. Get a SCSI board and get
either an Archive QIC-525 drive or an Exabyte 8200/8205/8500/8505 drive.
D.) For CDrom, stay away from the drives with special interfaces. Go with
SCSI drives. NEC, Teac, Toshiba, and Panasonic are good drives.
E.) Buy a logitech, Mouse Systems, or Microsoft mouse.
F.) Don't use internal modems. Stick to external units, that's what serial
ports are for.
G.) If you're building a UNIX, Novell Server, or OS/2 box, go with SCSI
disk drives. If you're doing Windows, you'll probably be better off
with IDE. MFM, RLL, and ESDI are dead.
If you follow the above rules, it will take you longer to put the system
together and it will cost more, but you will have a reliable system.
|
ajax
|
|
response 107 of 127:
|
Feb 21 09:26 UTC 1995 |
Re #100, lilmo, you have a point, 486/66's are older and more tested,
as are pre-PCI buses...VLB is fading, but right now, VLB cards are better
refined than their PCI counterparts. Perhaps it would be a better pick for
Grex for now. I was surprised to learn M-Net, with decent performance, has
a 486/33. For points of comparison, M-Net, which seems to often have 60
users, has a 486/33 w/48MB RAM...I think its comm lines are its main
bottleneck. Another system I use, which often has 120 users, has a Pent/66
w/128MB RAM. So for Grex, at roughly our current usage, a 486/66 would
probably be plenty of power.
Re #101, marcus, the second part of your argument was about the security
of Intel chips. Some thoughts on this: all modern chips can have security
holes and the chip designer's girlfriend's name on the silicon, not just
Intel's. And Intel's handling of the division error was a fiasco, but I
think they've learned some lessons. Also, even with their bungling, they
eventually agreed to replace all affected chips. Even early on, they agreed
to replace chips for those to whom it "mattered." Assuming a serious
security flaw exists, Grex might get bitten by it, but once discovered,
Intel would likely correct the problem. In the interim, some irritating
workarounds might be needed on Grex, but this all seems like quite unlikely.
O/S security holes are much *much* more common than microprocessor security
holes, and Grex has survived those. One difference is that a microprocessor
problem *might* take longer to correct.
Re #106, gregc, that's a sound list of tips! One point I'd differ on are
that a lot of cheap motherboards work fine...though they could be oem'd
Intels for all I know. Some are bad, especially for recent or obscure buses,
but in my experience, their problems generally show up immediately (e.g.
"Error: Can't access above 16MB"). Though I haven't done much with Unix
systems, as I'm sure you have. Of course, buying the ones you mention is
a good idea, but if money is an issue, I think many others are fine, too.
|
gregc
|
|
response 108 of 127:
|
Feb 21 11:48 UTC 1995 |
However, if you are building a UNIX system, using a no-name motherboard
that works fine for Windows, will be a complete crap-shoot for UNIX.
Consider: You could probably get a no-name 486/66 motherboard for 400, or
the AMI for $800. But if your no-name doesn't work and you go through 3
different motherboards before finding one that works, and you waste 20 or
30 hours in the process, was your $400 board *really* cheaper? I guess it
depends on how you value your time. In my case, it would definately be a
*loss*, a big one.
|
mdw
|
|
response 109 of 127:
|
Feb 21 12:27 UTC 1995 |
Greg's tips are I think sound, including the mother board tip. Unix is
much better at exercising the interrupt & dma hardware than is dos or
windows. Note that Greg is willing to pay *twice* what the el cheapo
house would charge for a system.
If you look back, historically, Intel has *always* had weird problems
with their chips. Their designs have generally not been as simple and
clean as one might wish. Their documentation has, quite frankly,
sucked. And, at least since the 286, they've had difficulty in getting
their more complicated chips right. Ie, vendors who have been the
pioneers in putting products out, have also had to suffer through an
embarrasing sequence of mask revisions. Do you remember when they were
offering cheap but slightly broken 8087 math processors?
This is in notable contrast to many other companies. Motorola makes a
convenient example: in their original 68000 design, they basically got
it right the first shot. They went through something like 2 mask
revisions total before they were into mass production. (Intel went
through dozens of revisions on the 286 and never did get it completely
right). IBM, with their 6000 series, has discovered design "oops" in
the motherboard on occasion. When they've done so, they send out
replacement boards, even when the problem hasn't yet been necessarily
detected at a particular site. What Intel did wasn't just dastardly, it
was also not even representative of the rest of the field.
With the 486, and even more with the 386, you had a lot of competition
for the motherboard market. Not so with the pentium. Intel probably
makes something like half the motherboards made. That's a recent, and
one might say, omninous shift in the market. That has not resulted in
higher prices for systems yet - but that's not because of Intel; that's
because of IBM, and Motorola, and Apple. Pentium systems hit the market
at least a year ahead of what Intel would have liked, because they are
scared of the PowerPC chip, and rightly so. It will be very interesting
to see how this goes in 3-4 years -- whether Intel will succeed in its
drive to monoplize the "pc" market, and if not, what the competition
looks like.
One of the nice things about the Sun platform we're on right now, is
they've been well broken in. We don't have to worry about debugging the
latest revision mask of the CPU, not only because the chip is better
engineered, but it also has nearly a decade of use behind it. That
means if there were any problems, they've been found and (we hope)
fixed, spare parts are plentiful and, on the whole, life is a lot more
boring. We also have a convenient upgrade path, all the way up to
systems that are very much leading edge, and far superior to even a top
of the line pentium.
That is not to say I don't see a use for intel based hardware, in grex.
We already run both kinds of hardware, and I see this persisting into
the foreseeable future. I see us using cheap redundant intel hardware
for routers, servers, and other back end support needs, while using
older higher end workstation equipement for login servers and the
visible ("programmable") components of grex. This gives us the best of
both worlds; the cheap I/O of intel, and the reliable trustworthiness
and redundancy of sun.
|
sidhe
|
|
response 110 of 127:
|
Feb 21 15:26 UTC 1995 |
What I was saying was is the p75 system I was eyeballing a bad
choice for my new net-cruiser, NOT whether or not I could run a grex
off of the damn thing! <anyway, why would I want to, when you guys do
as good a job as you do?>
|
gregc
|
|
response 111 of 127:
|
Feb 21 19:13 UTC 1995 |
An I think Steve already answered your question. Don't buy computers
at Best Buy. Buy computers at computer stores, or assemble them yourself
from trusted components.
|
ajax
|
|
response 112 of 127:
|
Feb 21 23:53 UTC 1995 |
Gregc, you're right, there's usually a $300-500 premium on name-brand
motherboards, and cheapos will take more time to fuss with on average.
For business purposes, it's usually not worth the hassle; time is money.
But for Grex's purposes, it *may* be worth added hassle to save some bucks.
I don't mean to devalue staff time, but it's a tradeoff Grex does make;
we could often spend some money to save some staff time. Where to draw
the line is a good question. On this motherboard issue, if I were doing
the setup, I'd opt for the cheapo route, but whoever does the work should
have the most significant say in the matter.
Marcus, all chips can have security holes, but for the sake of argument,
I'll concede Intel might be more likely than others. Still, given that,
don't you think that (a) it's much less likely than o/s and program security
holes on open unix systems like Grex, and (b) if a chip hole is found, that
it would be fixable, albeit inconveniencing?
Out of curiosity, any idea how the Intel 486/586 clones do with BSDI or
other unixes? I have an AMD 486 at home that runs netware, dos, win, and
linux without probs, but for light usage. I've heard that not all the
clone chips are compatible with all o/s's and programs. And Marcus, if
they were compatible (big if! :), would you be more comfortable with a
clone 486 chip than with Intel's?
|
steve
|
|
response 113 of 127:
|
Feb 22 02:20 UTC 1995 |
The tradeoff isn't so much staff time, as downtime. No matter what
motherboard we chose, we could get Grex up on it, assuming that a UNIX
would work at all.
The (killer) problem would be if that motherboard was generally flakey
as motherboards sometimes are, and Grex died once a day, because "something"
hung. Everyone would then pay, and it might just kill this organization.
So spending another $400 or so on a better motherboard doesn't seem so
bad, espically since it would be something on the order of 10% of the cost
of the system.
With regard to the reliability of Intel CPU's, I think there are
enough 386/486/Pentium UNIX boxes out there that we'd have heard about
them in general as being horrid if they really were. But we haven't, and
if {3,4,5}86 boxes aren't the most popular UNIX platform in the world yet,
they soon will be.
Grex could even use one of the "flawed" Pentiums. We don't do a whole
lot of floating point divides in the course of things here, and that which
we do, we can do in software.
|
mdw
|
|
response 114 of 127:
|
Feb 22 08:35 UTC 1995 |
I think if we found a mistake in the protection model, Intel would laugh
in our faces. And even *That* is assuming we can find somebody at Intel
willing to take the time to understand whatever we discover. Protection
model problems are hard to understand, and even harder to "sell". I
remember talking to sales people at Alpha Micro (a low cost mini
computer maker of a few years back) and they were quite proud of *not*
having any significant protection hardware. The 5%-10% speedup in the
hardware was something they could sell to customers. A more rugged
software development environment was something they really didn't give a
damn about. In the 386/486/586 marketplace, protection is NOT there to
provide a bomb-proof software environment: it is there to allow for
storage relocation and virtualization.
I've seen quite a few intel based Unix systems over the years - from an
early 8088 based system, through a multitude of Xenix, and freebsd
today. All of these were useful products for their intended
applications. All of them shared one common characteristic: holes in
the protection model that would allow a knowlegable malevolent person to
crash the system.
Here, on grex, most people are well disposed towards the system. The
number of people who are not favorably disposed is in fact astonishingly
small. Nevertheless, there *are* people like that out there. That
makes these people an operational problem: so, just like regular
backups, power conditioners, and good quality modems, we also need to
worry about making the system as "bomb-proof" against malevalent persons
as possible. Certainly there plenty of opportunity for mischief in the
OS and configuration - and indeed, the SunOS config we run today has
many changes because of just this problem. But that concern needs to
extend all the way down to the chip level. With the 68020, and with
SunsOS 4.1, we have good evidence (8 years of history) to suggest we
don't have a problem. With the intel 386/486/586, over a decade of
history suggests there are certain to be problems, and the youth of
pentium & 386bsd systems is in itself further cause for concern. I
think it can be pretty much taken as a given that there *are* problems,
and if we ran 386bsd, we'd find out about them, the hard way.
Fortunately for us, there *are* alternatives, and what's more, we also
have plenty of other reasons to get experience with 386 hardware. So
it's not really an either/or kind of situation, and if the situation
changes down the line -- if Intel, with future chips, really does turn
around; if the PC marketplace recognizes the value of protection
hardware as an anti-virus defense, then we'll be in a good position to
switch.
|
gregc
|
|
response 115 of 127:
|
Feb 22 10:45 UTC 1995 |
Actually, Marcus, in case it's not obvious, I'm not arguing for Grex to
become Intel based. I was just playing devil's advocate and trying to
describe how a stable {3,4,5]86 system *can* be built.
I would very much prefer for Grex to stay on Sun based 68020/68030 and
Sparc based architectures. Sun built hardware designed to be servers
and designed to be part of TCP/IP networks, and that's what we need.
|
sidhe
|
|
response 116 of 127:
|
Feb 22 13:43 UTC 1995 |
This is all highly confusatory. The question I had was if a HOME
system,
not a system that was going to come under the workload that a net does,
could be a packard bell with a fair amount of reliability..
I've seen two different thoughts on these lines, and it isn't
clarifying anything.
|
popcorn
|
|
response 117 of 127:
|
Feb 22 15:13 UTC 1995 |
If you're not doing anything heavily intensive on your home computer,
a Packard Bell should be OK.
To see the original topic of this item, type "only 0" at the "Respond
or pass?" prompt. If people aren't answering your home computer
question, it's probably because they're trying to stick to the original
topic, I'd guess.
|
ajax
|
|
response 118 of 127:
|
Feb 22 20:26 UTC 1995 |
Re 115, I think if someone gave a cost analysis, with a rough "power"
comparison to 486s, Sparcs would be more attractive. If I could offer a
comparison, I would, but as I said before, I'm only familiar with PCs.
I do think the ideas people are sharing are helpful, as good info will
allow better decision-making.
Based on suggestions made, here are some updated prices. The components
include a good Adaptec disk controller (of the appropriate bus), 3Com
ethernet card, and Fujitsu disk. Term server is a used 16-port. The 64MB
RAM is paritied, though we might get by with 48MB RAM for a while, with
M-NET as a point of comparison. The base system includes a floppy disk,
mono monitor, serial port, and decent power supply.
Base system total: $500 to 1500 (the $900 option sounds likeliest)
Cheap 486-66/VLB base system $500 -or-
Good 486-66/VLB base system $900 -or-
Cheap P90/PCI base system $1050 -or-
Good P90/PCI base system $1500
Component total: $4500 (includes $1400 in parts Grex needs anyway,
and we might get by with $500 less RAM)
64MB RAM $2100 BSDI binaries $550
SCSI2 card $200 YAPP $100
2GB disk $600 Term server $800
Ethernet $100 Shipping $50
As an optimistic view, a good 486-66 with 48MB RAM would cost $3500
beyond what Grex already plans to spend (on a Term server & disk).
|
sidhe
|
|
response 119 of 127:
|
Feb 22 22:04 UTC 1995 |
Hm. This is educational, at least. Why YAPP, though? Won't PicoSpan
run on an intel based system?
Popcorn- I know my question was drift, but if it was to be answered
at all, I wanted something a bit more concrete, like what you said.
I can now "cancel" my drift-inquiry, and get back to the main topic.
|
steve
|
|
response 120 of 127:
|
Feb 23 12:38 UTC 1995 |
YAPP was used as an example because its distributed in source form,
and thus can be compiled on a Sparc or 486.
|
sidhe
|
|
response 121 of 127:
|
Feb 23 14:57 UTC 1995 |
Oh, all right.
|
ajax
|
|
response 122 of 127:
|
Feb 23 16:49 UTC 1995 |
Actually, I put it in there because I thought someone said it would be
harder to port to BSDI, but looking back, chelsea mentioned merely that
M-Net got YAPP because a BSDI Pico wasn't easily available and they were
in a hurry, and jfk said that at least two BSDI ports of Pico exist, so
maybe that's the easier/cheaper way to go.
|
sidhe
|
|
response 123 of 127:
|
Feb 23 21:13 UTC 1995 |
I see.. facinating.. <sidhe goes to get his newbie guide to net
hardware terminology..>
|
tsty
|
|
response 124 of 127:
|
Feb 28 22:27 UTC 1995 |
<<You're here for the education, right? WEll ... here you get it>>
|