standalone NTP servers
nic...@esperi.org
this URL: http://www.truetime.com/DOCS/networktime.html or
any other types, and have any comments they wish to share?
This seems to be easier and smaller than using a real computer to
do the job.
Nicolai
nic...@esperi.org
Peter Radcliffe <26$10$f3i...@pir.net> wrote:
><nic...@esperi.org> probably said:
>>Has anyone used a TrueTime standalone NTP server, for example those at
>>this URL: http://www.truetime.com/DOCS/networktime.html or
>>any other types, and have any comments they wish to share?
>
>
>http://www.coetanian.com/tss/tss100.htm
>
>for work. Not sure which would be better/cheaper.
$2000 is wince-provoking. I certainly don't need the Windows NT software
either.
>>This seems to be easier and smaller than using a real computer to
>>do the job.
But not cheaper; since posting the article, I got through to the UK
distributor of TrueTime products who quoted me a price of GBP 2395
(excluding VAT, I'm sure) for the cut-down NTP server with serial
configuration and NTP output only.
That's WAY too expensive for me.
Someone has mentioned a unit with similar abilities for perhaps $1000
which is much more useful, but the one who mentioned it to me couldn't
remember which it was.
Nicolai
Russ Allbery
> 3) Connecting a GPS unit to a real machine (in my case sparcs, since
> I run solaris boxes) is a royal pain in the neck. I've researched
> how to reliably connect a GPS clock and a sparc - it isn't pretty
> if it is actually possible.
It's definitely possible.
windlord:~> ntpq -p fathertime
remote refid st t when poll reach delay offset disp
==============================================================================
*TRUETIME(2) .GPS. 0 l 60 64 377 0.00 -4.771 6.07
time-C.timefreq 0.0.0.0 16 - - 1024 0 0.00 0.000 16000.0
clock.isc.org 0.0.0.0 16 - - 1024 0 0.00 0.000 16000.0
MightyDog.Stanf 0.0.0.0 16 - - 1024 0 0.00 0.000 16000.0
Avallone.Stanfo 0.0.0.0 16 - - 1024 0 0.00 0.000 16000.0
mashhad.Stanfor clock.isc.org 2 u 34 128 373 3.57 10.535 2.91
Argus.Stanford. 0.0.0.0 16 - - 1024 0 0.00 0.000 16000.0
EE.Stanford.EDU 0.0.0.0 16 - - 1024 0 0.00 0.000 16000.0
leland.Stanford afssvr3.Stanfor 4 u 301 1024 377 6.47 0.779 0.66
kerberos1.Stanf 0.0.0.0 16 - - 1024 0 0.00 0.000 16000.0
Seems to work reasonably well, but I'm not the person who had to put it
together either.
(Hm. leland is supposed to be stratum two, not stratum four. Must go fix
that.)
--
Russ Allbery (r...@stanford.edu) <URL:http://www.eyrie.org/~eagle/>
Thorfinn
> really should have it's own stratum 1 :/
Maybe I'm missing something, but what's wrong with a GPS unit as the
stratum 1 clock? xntpd seems to have a whole bunch of stuff that lets
you configure it to talk that, and surely a crap 386 with a GPS
attached should be pretty cheap?
ook,
Thorf
--
<a href = "http://netizen.com.au/~thorfinn/">thor...@netizen.com.au</a>
Beware of the Turing Tar-pit in which everything is possible but
nothing of interest is easy.
-- BSD fortune file
Thorfinn
Peter Radcliffe <26$10$f3i...@pir.net> wrote:
> Thorfinn <thor...@netizen.com.au> probably said:
> >Maybe I'm missing something, but what's wrong with a GPS unit as the
> >stratum 1 clock? xntpd seems to have a whole bunch of stuff that lets
> >you configure it to talk that, and surely a crap 386 with a GPS
> >attached should be pretty cheap?
> "consumer grade" and not server grade if it is affordable.
> I don't use PC hardware for anything but desktops (PC hardware
> does make good affordable desktop machines). All my work servers
> have maintenance contracts.
*nodnod* Well, I can understand that.
> 2) I want the stratum 1 to be very reliable. See 1)
> Hopefully the standalone stratum 1 units have no moving parts
> apart from possibly a fan.
Mmm... okay. I have this preference for reliability through multiple
doovers with medium MTBF, rather than single doovers with high MTBF.
Plug in 3 crappy 386en, netbooted and diskless, if you really care,
with a GPS plugged into each of them.
Then you've got 3 stratum 1s, and I think the odds of *all* of them
going down at once (barring meteor strike, or total power loss to the
machine room, etc) are pretty durn small.
> 3) Connecting a GPS unit to a real machine (in my case sparcs, since
> I run solaris boxes) is a royal pain in the neck. I've researched
> how to reliably connect a GPS clock and a sparc - it isn't pretty
> if it is actually possible.
There are GPS units that talk serial protocol... and sparcs generally
have a serial port or two, IIRC... and xntpd will happily talk serial
at a GPS unit to be a stratum 1.
> 4) I don't want to mix services between the stratum 1 and other things
> (I prefer to keep services as seperate as I can) running a real
> machine just to be a stratum one is overkill.
Okay. :) That's true enough, but see my multiple doover reasoning
above.
> 4) For many things I like the idea of appliances - an NTP appliance
> appeals to me.
*nodnod* I'm not going to argue with personal preference. :) Nothing
wrong with that.
> 5) I have no 5th point right now.
Yes you do. :) That 5th point is mislabelled, and the 6th point is
mislabelled...
ook,
Thorf
--
<a href = "http://netizen.com.au/~thorfinn/">thor...@netizen.com.au</a>
Beware of Programmers who carry screwdrivers.
-- Leonard Brandwein
Ingvar Mattsson
> Thorfinn <thor...@netizen.com.au> probably said:
> >Maybe I'm missing something, but what's wrong with a GPS unit as the
> >stratum 1 clock? xntpd seems to have a whole bunch of stuff that lets
> >you configure it to talk that, and surely a crap 386 with a GPS
> >attached should be pretty cheap?
>
> 1) I hate PC hardware. It generally blows and is definately
> "consumer grade" and not server grade if it is affordable.
> I don't use PC hardware for anything but desktops (PC hardware
> does make good affordable desktop machines). All my work servers
> have maintenance contracts.
>
> 2) I want the stratum 1 to be very reliable. See 1)
> Hopefully the standalone stratum 1 units have no moving parts
> apart from possibly a fan.
>
> 3) Connecting a GPS unit to a real machine (in my case sparcs, since
> I run solaris boxes) is a royal pain in the neck. I've researched
> how to reliably connect a GPS clock and a sparc - it isn't pretty
> if it is actually possible.
>
> 4) I don't want to mix services between the stratum 1 and other things
> (I prefer to keep services as seperate as I can) running a real
> machine just to be a stratum one is overkill.
>
> 4) For many things I like the idea of appliances - an NTP appliance
> appeals to me.
>
> 5) I have no 5th point right now.
We have something called either "Timeserve" or "Tymserve",
synchronises to the GPS time signal, has an LCD configuration
interface with some options (basically, "set IP address" and "Am I S1
or S2?", IIRC). Biggest problem is that if it's mounted somewhere *in*
the server room, it tends to need a prod now and then, since it
sometimes loses track of everything if the GPS coverage is bad.
//Ingvar (and they're Y2k safe...)
--
When in douFNORD! This signature has been hi-jacked by Fnord Information
systems, to fnordprovide you with unfnordlimited information.
Otmar Lendl
>
>4) For many things I like the idea of appliances - an NTP appliance
> appeals to me.
>
We had one of these here for testing.
* It crashed when sending it one of the typical attacks (land.c, winnuke, ...).
* It answered requests before it synced to GPS, giving out false time.
* It only spoke ntp, but not the simpler time and daytime service.
No, we didn't buy one.
/ol
--
/ Otmar Lendl (O.L...@Austria.EU.net) | Phone: +43 1 89933-0 (-533 fax) \
\ EUnet tech staff | Diefenbachgasse 35 A-1150 Wien /
Nicolai
Russ Allbery <r...@stanford.edu> wrote:
>Peter Radcliffe <26$10$f3i...@pir.net> writes:
>
>> 3) Connecting a GPS unit to a real machine (in my case sparcs, since
>> I run solaris boxes) is a royal pain in the neck. I've researched
>> how to reliably connect a GPS clock and a sparc - it isn't pretty
>> if it is actually possible.
>It's definitely possible.
>
>windlord:~> ntpq -p fathertime
> remote refid st t when poll reach delay offset disp
>==============================================================================
>*TRUETIME(2) .GPS. 0 l 60 64 377 0.00 -4.771 6.07
Can I sync to this? :-)
>Seems to work reasonably well, but I'm not the person who had to put it
>together either.
Indeed it can be made to work reasonably well; however, I'm reasonably
convinced that for the cost of an OEM GPS receiver (GBP 300 or so in .uk),
and a small embedded PC and some minimal core of a free UNIX, one could
make an ethernet-connected NTP box that's solid-state and reliable for
perhaps GBP600, which is not far off the cost of a retail GPS receiver and
a computer to attach it to.
Computers have discs, fans, and all that crap which fails, and something as
simple as an NTP server shouldn't have them. I seem only to find things that
are either overpriced, or also intended for purposes like providing
precision frequency standards or supporting 42 different time-distribution
protocols, and I don't want all that. I work for an ISP, not an electronic
engineering company.
Also, my manglement is more likely to go for a well-integrated box than a
custom hack, if the cost difference is not a factor of 5.
I just want one that costs less than my car.
Or alternatively suggestions for a GPS receiver known to be easy to
plug into a Sun SPARCstation.
Nicolai
nic...@esperi.org
Thorfinn <thor...@netizen.com.au> wrote:
>In bofh.general, on 19 Jan 2000 18:41:04 GMT
>Peter Radcliffe <26$10$f3i...@pir.net> wrote:
>> <nic...@esperi.org> probably said:
>> >Someone has mentioned a unit with similar abilities for perhaps $1000
>> >which is much more useful, but the one who mentioned it to me couldn't
>> >remember which it was.
>> I'd be interested as well, since an organisation as large as tufts
>> really should have it's own stratum 1 :/
>
>stratum 1 clock? xntpd seems to have a whole bunch of stuff that lets
>you configure it to talk that, and surely a crap 386 with a GPS
>attached should be pretty cheap?
Yes. So why can't I buy one already constructed, for less than the
cost of DIY? I'm sure there must be one somewhere.
For me, a crap PC is not an option; I want reliable hardware, which
means a server-grade PC or a Sun or something. They're all expensive
and complicated and have moving parts that will break.
Time-intensive [1] solutions like custom hacks, old and flaky hardware,
and so on to save money are not tenable at my place of ork, because we
don't have time to go back and fix things more often than we have to.
So if I have to go for the computer + GPS route, it'll be a real computer.
Unfortunately the serial ports on the real computers I have around are
not very good (all they do is provide 9600bps serial consoles, so this
doesn't matter in any other application we have).
Nicolai
[1] Boom, Boom. Sorry.
Russ Allbery
> Can I sync to this? :-)
No, all of our high-stratum servers are limited to on-campus connections.
I need to open up some stratum three servers for general access, though,
for our travelling users. On my to-do list somewhere. :)
Kai Henningsen
> Russ Allbery <r...@stanford.edu> probably said:
> >Peter Radcliffe <26$10$f3i...@pir.net> writes:
> >> 3) Connecting a GPS unit to a real machine (in my case sparcs, since
> >> I run solaris boxes) is a royal pain in the neck. I've researched
> >> how to reliably connect a GPS clock and a sparc - it isn't pretty
> >> if it is actually possible.
> >It's definitely possible.
>
> >Seems to work reasonably well, but I'm not the person who had to put it
> >together either.
>
> Oh, you can get it working, but because of how sparc (well, older
> sparcs I'm not 100% sure about Ultra serial) serial ports work and how
> long it can take to respond to a serial interrupt you get a random and
> up to 200ms delay. Since it's random ntpd can't correct for it ...
Well, use a real machine (such as a cheapo 386) then instead of toy
hardware ...
(Actually, I'm just about speechless to find out that there's serial
hardware around that is even more crap than on a PC, let alone in
supposedly-better boxen.)
Kai
--
http://www.westfalen.de/private/khms/
"... by God I *KNOW* what this network is for, and you can't have it."
- Russ Allbery (r...@stanford.edu)
ric...@illuin.org
> Kai Henningsen <kaih=7XHAH...@khms.westfalen.de> probably said:
>>Well, use a real machine (such as a cheapo 386) then instead of toy
>>hardware ...
>
> (and architechture, etc, etc).
> See my other post about not wanting unreliable machines and not
> using consumer grade hardware for servers.
>
so, buy a rack-mount industrial PC ; you could get a NVRAM disk to store
the OS but only if it's small. okay, so they are not cheap either,
but they will be cheaper than a whole server-like machine. (about $600)
There's a spacial offer on at http://gtweb.net/rack-bp.html
RjL
Kai Henningsen
> Kai Henningsen <kaih=7XHAH...@khms.westfalen.de> probably said:
> >Well, use a real machine (such as a cheapo 386) then instead of toy
> >hardware ...
>
> Cheapo 386s are not real hardware. They're crappy hardware
> (and architechture, etc, etc).
> See my other post about not wanting unreliable machines and not
> using consumer grade hardware for servers.
If you really want to get serious on this point, I'll have to say that for
a specific application, where you don't need lots of peripherals, there is
nothing particularly crappy or problematic about this architecture. It
just works.
Just make sure you don't get non-standard variations from some brand-name
(Compaq, for example) source. *Those* suck for anything that doesn't have
direct support (and sometimes even for stuff that does). Stay away from
people who think they're big enough to make up their own rules.
And use a real OS, such as Linux or one of the BSDs, not Wincrap or
Notware. Personally, I'd also stay away from commercial Unixen, Solaris,
SCO, whatever; previous experience with commercial x86 Unixen hasn't been
favourable (and that's an understatement).
My personal experience with Linux on clone-type hardware has been very
good. Well, maybe not exactly using the cheapest stuff around, but using
the oldest stuff from the known-good providers *is* both cheap and gives
reliable machines, in my experience. And if you don't need the fastest
machine around (and for NTP, you do not), it does what you need, too.
Old board from ASUS or Gigabyte, say; usually has 16550-equivalent serial
on board; small DRAM module (whatever the smallest available today should
still be enough); cheapest AMD CPU still available (plenty fast enough);
any small IDE disk to boot from (can you even get less than a gig today?);
any cheap video card and keyboard (only needed for setup anyway); maybe
cheap CD drive for install CD - floppy+net should work, but floppy drives
are among the most unreliable PC hardware around, and just about any board
available today can boot from an IDE CD; oh, a network card, of course -
just about any cheap card (such as an NE2000 clone) ought to do for
something like this. Case. Put together, throw bootable CD in, switch on,
install, configure, use & forget. Or get someone else to put it together
for you, though it's really not hard. Or even buy something like that
ready-made from nearest PC supplier, though that's bound to be
significantly more expensive, given that the mass market wants
"multimedia" pcs with real CPU power, soundcard, and so on.
Simon Lockhart
>Has anyone used a TrueTime standalone NTP server, for example those at
>this URL: http://www.truetime.com/DOCS/networktime.html or
>any other types, and have any comments they wish to share?
Yes, we have two of those (one GPS, one IRIG-B for connecting to "BBC Time").
They work well, although not cheap.
Simon
--
Simon Lockhart | Tel: +44 (0)1737 839676
Internet Engineering Manager | Fax: +44 (0)1737 839516
BBC Internet Services | Email: Simon.L...@bbc.co.uk
Kingswood Warren,Tadworth,Surrey,UK | URL: http://support.bbc.co.uk/
Simon Lockhart
Peter Radcliffe <26$10$f3i...@pir.net> wrote:
>3) Connecting a GPS unit to a real machine (in my case sparcs, since
> I run solaris boxes) is a royal pain in the neck. I've researched
> how to reliably connect a GPS clock and a sparc - it isn't pretty
> if it is actually possible.
I don't know if it's still available, but for one project I had an
involvement in, we used an SBus card which had IRIG-B in. Admittedly, you
do then need an external source of "time" in IRIG-B form - being the BBC,
we can find this pretty easily. :)
Kai Henningsen
> These days, DEC's putting PC serial ports into Alphaservers... and they're
> so screwed up you can't use them reliably at >9600 bps, AND they're
> jiggering tty00/COM1 so that they ignore modem control... they're looped
> back on the motherboard! Why? I assume that's so that you can boot off a
> dumb terminal with a 3-wire connection. Why not have SRM take care of it I
> don't know.
Hell.
These days, PC motherboards have on-bord serial ports that don't have
those problems. Well, all mobos I looked at, anyway.
If you need a fsking three wire connection, put the loopbacks in the
connector, where's the fscking problem here?!
Down. Not across.
[..]
> any small IDE disk to boot from (can you even get less than a gig today?);
ITYM "ten gigs", and no, not really.
-- Niels.
Scott Hazen Mueller
>>The 80386 came out in 1986.
>Were machines actually available then, though ? If so, I'm misremembering
>dates.
My employer bought a Compaq 386 machine - the first commercially available 386
PC - no later than 1987. If I could find an old enough comp.mail.maps
archive, I could nail it down better... Definitely no later than 1 September
1987, though, as I have a .pnewsexpert file with that mtime and that dates
from the original zorch, which I bought for personal use and loaded up with
files from that Compaq. Over a crappy serial connection, as it happens.
Microport SVR3 for the 386 had a truly abysmal serial driver.
>Perhaps not much faster in raw speed but they didn't degrade at higher
>load anywhere near as badly. The arch didn't suck anywhere near at much.
V7(?) Sparcs (e.g. 4/100s, 4/200 and SS1) had 8 or 64 (?) contexts in the PMMU
and sucked eggs under multitasking load. Our 4/470 was much more graceful in
that regard, though admittedly rather huge.
>My SS1 running NetBSD is still happily usable interactively when heavily
>loaded. A 386 running *BSD, even when built well from decent parts and using
>scsi disk, is not usually.
Must have a different definition of heavy load. *Anything* (Sparc, VAX, x86,
MIPS) runs slow when it swaps.
Sparcs are valuable because they're fairly well-balanced systems. The memory
bus, paging architecture, peripheral bus, disks and CPU are pretty well
matched. x86 machines tend to have enormously oversized di...er, CPUs, and
too little RAM and slow disks. Head-to-head, there's little in the modern PC
architecture (modern!) that is inherently broken. However, implementation is
a sticking point. I know if I buy a Sun that I'm getting a well-integrated
reliable machine with quality parts. I know if I buy a no-name clone that I'm
not. The hard part with PCs is finding systems that are built with the good
stuff, and with the *right* good stuff for the OS you'll be running. Shops
that work with PCs a lot are well-advised to culture a pet integrator and get
him/her to do that part for you.
\scott
Lars Balker Rasmussen
> And use a real OS, such as Linux or one of the BSDs, not Wincrap or
> Notware. Personally, I'd also stay away from commercial Unixen, Solaris,
> SCO, whatever; previous experience with commercial x86 Unixen hasn't been
> favourable (and that's an understatement).
FWIW, our main server is running Solaris 7 (used to be 2.6) on a dual
PII-233 (Asus LX97 board). Works very nicely, and the only reason I'd
exchange it with a Sun would be to solve some problems with application
support.
Yes, it means we have a 5" monitor and a keyboard in the rack, but it's
a small sacrifice for the ability to exchange hardware cheaply as the
need arises.
--
Lars Balker Rasmussen X500 *is* a Bag Thing happening.
-- Christi Alice Scarborough
nic...@esperi.org
Kai Henningsen <kaih=7XHAH...@khms.westfalen.de> wrote:
>26$10$f3i...@pir.net (Peter Radcliffe) wrote on 20.01.00 in <866dk6$7s5$1...@moek.pir.net>:
>
>> Russ Allbery <r...@stanford.edu> probably said:
>> >Peter Radcliffe <26$10$f3i...@pir.net> writes:
>> Oh, you can get it working, but because of how sparc (well, older
>> sparcs I'm not 100% sure about Ultra serial) serial ports work and how
>> long it can take to respond to a serial interrupt you get a random and
>> up to 200ms delay. Since it's random ntpd can't correct for it ...
>
>Well, use a real machine (such as a cheapo 386) then instead of toy
>hardware ...
Well, I think the UltraSPARC machines are somewhat better, and I've
discovered I can get to use one of those (even though it's total overkill).
I've also noticed that the Trimble Palisade has a feature whereby you
send it a signal transition on a control line and it responds sometime
later with the precise time you asserted the control line, which seems
to make things more accurate since you can assert a control line
much more precisely than service an interrupt from an incoming
control line or incoming data.
I wonder if ntpd gets more accurate if I give it realtime scheduling
priority.
Nicolai
Nicolai
Peter da Silva <pe...@taronga.com> wrote:
>Peter Radcliffe:
>>I was hoping a small ntp appliance would be cheaper than both ...
>
>You gotta be kidding. Volume on "thin servers" is low so the costs are way
>up there, and most of the time they're the industrial grade PC you just
>spurned in a black box with a flash boot disk.
Well, cost in .nl [1] of the Tymserve previously discussed is US$3600. I can
get a hell of a lot of PC, even server-grade Compaq PC, for that. I find it
hard to believe that is a low-margin item.
Nicolai
[1] Thanks Andre. I haven't got a .uk quote yet. But another NTP-only
box was quoted as GBP 2400 (== USD 3600).
nic...@esperi.org
>intel clone processor than a Sun workstation plugged into a rackmount shelf.
I seem to be banging on Peter (da Silva) a lot... sorry, but you're raising
the points I care about :)
How is remote manageability on these? Can I use a serial console, network
boot it, boot off alternate disc, boot off CDROM, etc? Does it have
environmental monitoring and a PROM or lowlevel monitor [2] that I can get
to with a key sequence to power cycle the box if it's locked up for some
reason?
The machine may be exiled to Manchester, which is a long way from where
I am, and the staff we have there may or may not be more intelligent
than an orangutan [1].
Nicolai
[1] I don't know; but some of our remote hands make my desk look like it
is filled with initiative and ability to solve problems, or in fact
speak intelligibly.
[2] eg, the LOM found on a Sun Netra 1, or the console thing on a VAX.
nic...@esperi.org
Simon Lockhart <sim...@rd.bbc.co.uk> wrote:
>In article <8647nf$d0u$1...@alfheim.satanic.net>, <nic...@esperi.org> wrote:
>>Has anyone used a TrueTime standalone NTP server, for example those at
>>this URL: http://www.truetime.com/DOCS/networktime.html or
>>any other types, and have any comments they wish to share?
>
>Yes, we have two of those (one GPS, one IRIG-B for connecting to "BBC Time").
>
>They work well, although not cheap.
Indeed not; I have been quoted GBP 2400 for the NTS-90, which does only
NTP. I'm afraid that strikes me as very overpriced, particularly since
it does not appear to be of stunning quality overall from what was
previously posted (bad at consistently reliable time when starting up,
TCP stack vulnerable to the latest crop of attacks).
That sort of thing is an issue if I ever plan to let people I can't
personally LART use the thing to sync with.
Nicolai
Scott Hazen Mueller
>The machine may be exiled to Manchester, which is a long way from where
>I am, and the staff we have there may or may not be more intelligent
>than an orangutan [1].
Orangs are pretty intelligent. They've apparently been known to use planning
and foresight, which is a heckuva lot more than you can say for some humans.
That 'may or may not' should be taken at face value whether you really
intended it that way or not...
\scott
Kai Henningsen
> Peter da Silva <pe...@taronga.com> probably said:
> >You mean 8250. The 16550 is a lot better than any SS1 serial port.
>
> No, I meant 16550 rather than 16550A.
You mean to say you actually saw that beast? So far, it's existence hasn't
been more than theoretical for me. (The port tester I wrote can recognize
them, but I never saw one.)
Seen lots of crap PC serial ports with 8250; seen lots of better ones with
16550A or better. Never, ever seen a real 16550; I gather not all that
many were ever produced (the difference between the 16550 and the 16550A
is the first has a hardware bug that makes the FIFO unusable).
> >If you had to put something better than a 16550 in your 486 to deal with
> >POTS modem speeds you were doing something fundamentally wrong.
>
> 16550 (rather than 16550A) won't deal with 115k2 serial speed reliably.
*Real* 16550 is effectively just a weird 8250.
> The P133 I had until recently would die a croaking death (even with
> scsi disk) if I put a fraction of the load my SS20s deal with without
> complaining, and without the users even _noticing_. Hence many of
> my comments about the PC arch sucking so hard.
>
> CISC. Bleh.
Impossible to diagnose from here, but I can't imagine you not doing
something seriously wrong. It certainly doesn't match my experience.
If you had said IDE disk, maybe. Or if that PC had seriously less memory
than the Sparc.
> and I say again, raw processor power is not my main consideration.
But you make comments about CISC vs. RISC, which is exactly that type of
consideration. Make up your mind.
> Less degredation under load is far more important.
That is mostly an OS problem. If it's a hardware problem, you're doing
something stupid like using IDE disks. Or maybe one of the old ISA el-
cheapo SCSI controllers that did everything with the CPU (Adaptec 1520 or
something like that, beloved by sound card makers).
I see Linux getting faster on identical hardware with major releases, and
getting much slower during devel kernels when not really tuned. I see NT
being catastrophically slower than Linux on identical hardware, and OS/2
noticably slower. This is usually an OS issue.
Kai Henningsen
> In article <86i72l$n3e$1...@alfheim.satanic.net>,
> Nicolai <nic...@alfheim.satanic.net> wrote:
> >>You gotta be kidding. Volume on "thin servers" is low so the costs are way
> >>up there, and most of the time they're the industrial grade PC you just
> >>spurned in a black box with a flash boot disk.
>
> >Well, cost in .nl [1] of the Tymserve previously discussed is US$3600. I
> >can get a hell of a lot of PC, even server-grade Compaq PC, for that.
>
> "Server Grade Compaq PC" isn't exactly what *I* would use as my benchmark.
> Ick.
>
> And you're not buying the hardware. You're buying the software and the
> integration.
At those prices, one can't help but wonder if one ought to go into
production on these things.
I mean, you only figure out how to do it _once_, and after that all your
cost is throwing the hardware together.
Matt McLeod
that The Prince of Insufficient Light did write:
>> If you had said IDE disk, maybe. Or if that PC had seriously less memory
>> than the Sparc.
>
>default in workstation-class machines?
Not just workstations.
They gave me a quote for an Ultra-10 "server" (note that the
RFQ specifically stated that the box was to be used as a server)
with IDE disk.
I suspect this was just to make sure that the E450 they actually
want us to buy looked better.
--
"Don't put off 'till tomorrow, responsibilities.
They'll just come back to haunt you.
(Ignore them totally)"
-- TISM
Matt McLeod
>In article <slrn88t35f...@enzo.netizen.com.au>,
>Matt McLeod <use...@wibble.cx> wrote:
>>Yea, it is written in the Book of Cyril
>>that The Prince of Insufficient Light did write:
>>>> If you had said IDE disk, maybe. Or if that PC had seriously less memory
>>>> than the Sparc.
>>>
>>>Has anyone raised the point yet that Sun now ships IDE disks as
>>>default in workstation-class machines?
>>
>>Not just workstations.
>>
>>They gave me a quote for an Ultra-10 "server" (note that the
>>RFQ specifically stated that the box was to be used as a server)
>>with IDE disk.
>
>hardly-expandable tower case as a server... but I'll assume you
>know what you're asking for.
The plan was to whack a D1000 on the back of it. We'd not long
ago retired an SS20 with an array of some sort, and needed to
provide a lowest-cost option for completeness sake. We're
currently in the process of trying to justify getting more
disk.
Actually, I'm not at all happy with Sun about this one. We
got an SSA100 (from memory -- fibrechannel thing) about 18
months ago. Part of the reasoning behind this choice was
apparently that Sun had assured them that this would be a nice
expandable option (also leased an E4k, which is what the SSA
hangs off the back of).
The disks are now full, and Sun won't sell us 4G disks any more.
They refuse to support it if we take either of the alternatives --
source 4G disks elsewhere (if we can) or put larger disks in.
So we've got management who were sold on how wonderful this
system was going to be now being told that it isn't as good as
they'd been previously told, and being asked to hand over more
money to fix it. They'll have to swallow it, of course, but
it's leaving them a wee bit distrustful of sysadmins making
technical recommendations.
(That the sysadmins who did this are no longer around, and that
it wasn't just admins who fooled them but also one of their
own managers doesn't seem to make much difference.)
>>I suspect this was just to make sure that the E450 they actually
>>want us to buy looked better.
>
>What about an E250? A 450 is an altogether bigger beast (and really
>quick).
If we were to get an E250, we'd have to add an array to it (only
six drive bays). E450 plus disks works out to be not too much more
than just a D1000 with less disks. Strange, but true.
But as it happens, the original RFQ was indeed for U10 and E250.
The E450 thing came out of a later meeting with our Sun rep and
his SE.
<snip>
>But IDE isn't _that_ crap. It's only able to handle a few discs, they
>effectively have to be in the same box, I'm sure it uses more CPU
>cycles than SCSI, and you can't expand it much, and so on. However it
>works fine for a small server or a workstation. It's also a lot cheaper.
>
>Saying "IDE bad, IDE horrible, never going to use IDE, it will never work"
>is not a well-thought-out position.
I use IDE at home, in both the BSD box on the floor and the iMac on
the desk. Haven't done any proper testing, but experience has shown
that an SS5 with SCSI disk handles heavy IO load *way* better than
a U5 with IDE. Start untarring something large on a U5, and you may
as well go get some coffee while it runs, 'cause the machine will be
damn near unusable until it's finished.
I don't see that kind of thing on PCs running BSD (or Linux, for
that matter). So I'm taking a wild guess that Sun has screwed
something up in the IDE support on the U5.
--
"Two of my imaginary friends reproduced once ... with negative results."
fl...@interport.net (void)
Lars Balker Rasmussen
> The Prince of Insufficient Light <dot+.g...@atat.dotat.org> probably said:
> >Has anyone raised the point yet that Sun now ships IDE disks as
> >default in workstation-class machines?
>
I just put a SCSI disk in my U10, but haven't got around to testing the
subjective speed of the machine yet (read: use the machine. I used an
old slow disk to see if it'd work, now I'll exchange it with a newer
more spiffy one.)
When I switched my U1E with the U10, the machine felt a LOT slower when
I asked it to do more than one thing at a time, so it'll be fun to see
how this fares.
Matt McLeod
>dot+.g...@atat.dotat.org (The Prince of Insufficient Light) wrote:
>
>[SCSI v IDE]
>
>>.... you can put two disks of differing manufacturer on one bus
>>and reasonably expect them to work together....
>
>?! I've never encountered this as an IDE problem.
It's an older problem. Haven't seen it in the last few
years, but it was very common for IDE disks to not work
with those from other manufacturers, and sometimes even
just various combinations of disk from the same...
>OTOH, we've got M$ all over the place, so if it was causing monthly
>failure, we'd probably never notice. <sigh>
I never saw it cause an actual failure as such, more just
stopping anything from working at all until you sorted it
out (generally by not getting caught in the first place)
--
Never count your chickens before they rip your lips off
Lars Balker Rasmussen
> I use IDE at home, in both the BSD box on the floor and the iMac on
> the desk. Haven't done any proper testing, but experience has shown
> that an SS5 with SCSI disk handles heavy IO load *way* better than
> a U5 with IDE. Start untarring something large on a U5, and you may
> as well go get some coffee while it runs, 'cause the machine will be
> damn near unusable until it's finished.
>
> I don't see that kind of thing on PCs running BSD (or Linux, for
> that matter). So I'm taking a wild guess that Sun has screwed
> something up in the IDE support on the U5.
Uh-huh. Also try copying a lot of stuff from CD to disk/nfs while doing
something on the U5/U10... Or preferably, do it on someone elses U5.
Matt McLeod
>use...@wibble.cx (Matt McLeod) writes:
>> I don't see that kind of thing on PCs running BSD (or Linux, for
>> that matter). So I'm taking a wild guess that Sun has screwed
>> something up in the IDE support on the U5.
>
>Uh-huh. Also try copying a lot of stuff from CD to disk/nfs while doing
>something on the U5/U10... Or preferably, do it on someone elses U5.
I have. She turned round and yelled at me. :-)
Am really looking forward to getting my hands on that SS5 again.
Been stuck with a PIII/450 for months...
--
"In those days he was wiser than he is now -- he used to frequently
take my advice." -- Winston Churchill
Jeff Wolfe
>> the desk. Haven't done any proper testing, but experience has shown
>> that an SS5 with SCSI disk handles heavy IO load *way* better than
>> a U5 with IDE. Start untarring something large on a U5, and you may
>> as well go get some coffee while it runs, 'cause the machine will be
>> damn near unusable until it's finished.
>>
>> that matter). So I'm taking a wild guess that Sun has screwed
>> something up in the IDE support on the U5.
>
>Uh-huh. Also try copying a lot of stuff from CD to disk/nfs while doing
>something on the U5/U10... Or preferably, do it on someone elses U5.
Any idea if priority_paging was set?
-JEff
--
Jeff Wolfe College of Earth and Mineral Science - Penn State
Kai Henningsen
> In article <7XWpk...@khms.westfalen.de>,
> Kai Henningsen <kaih=7XWpk...@khms.westfalen.de> wrote:
> >Seen lots of crap PC serial ports with 8250; seen lots of better ones with
> >16550A or better. Never, ever seen a real 16550; I gather not all that
> >many were ever produced (the difference between the 16550 and the 16550A
> >is the first has a hardware bug that makes the FIFO unusable).
>
> I've had a bunch of machines with 16450 parts, I don't know what the exact
> difference is between the 8250, 16550A, and 16450, but IIRC the 16450 was
> better than the 16550(non-A).
The 16450 is supposed to be pretty much exactly an 8250, IIRC.
Hmm. I ought to have a file on that somewhere ... right:
14.4. UART Model Numbers
Here's a list of UARTs. TL is Trigger Level
+ 8250, 16450, early 16550: Obsolete with 1-byte buffers
+ 16550, 16550A, 16c552: 16-byte buffers, TL=1,4,8,14
+ 16650: 32-byte buffers. Speed up to 460.8 Kbps
+ 16750: 64-byte buffer for send, 56-byte for receive. Speed up to
921.6 Kbps
+ Hayes ESP: 1K-byte buffers.
The obsolete ones are only good for modems no higher than 14.4k (DTE
speeds up to 38400 bps). For modern modems you need at least a 16550
(and not an early 16550). For V.90 56k modems, it may be a several
percent faster with a 16650 (especially if you are downloading
uncompressed files). The main advantage of the 16650 is its larger
buffer size as the extra speed isn't needed unless the modem
compression ratio is high. Some 56k internal modems may come with a
16650 ??
Non-UART, and intelligent multiport boards use DSP chips to do
additional buffering and control, thus relieving the CPU even more.
For example, the Cyclades Cyclom, and Stallion EasyIO boards use a
Cirrus Logic CD1400 RISC UART, and many boards use 80186 CPUs or even
special RISC CPUs, to handle the serial IO.
Most newer PC's (486's, Pentiums, or better) come with 16550A's
(usually called just 16550's). If you have something really old the
chip may unplug so that you may be able to upgrade by buying a 16550A
chip and replacing your existing 16450 UART. If the functionality has
been put on another type of chip, you are out of luck. If the UART is
socketed, then upgrading is easy (if you can find a replacement). The
new and old are pin-to-pin compatible. It may be more feasible to
just buy a new serial board on the Internet (few retail stores stock
them today).
from the Linux Serial HOWTO.
Kai Henningsen
> When I switched my U1E with the U10, the machine felt a LOT slower when
> I asked it to do more than one thing at a time, so it'll be fun to see
> how this fares.
That's exactly the expected SCSI<->IDE result. IDE degrades pretty badly
under real multitasking - it's too CPU intensive.
Of course, modern DMA versions are supposed to solve that, so it just
becomes a strange version of SCSI that's limited to two (usually) or four
devices per bus. (Modern IDE seems to prefer using SCSI data formats.
That's why everyone and his brother does an IDE-as-SCSI emulation driver.)
<shrug> My home machine uses SCSI for disks, and I don't plan to change
that anytime soon. Though I do now have a DVD drive on IDE.
Kai Henningsen
> In <389dea48....@news.lspace.org> al...@lspace.org (Alan Bellingham)
> >writes: ?! I've never encountered this as an IDE problem.
>
> I refuse to buy Quantum disks thanks to this problem. Whilst at
*I* refuse to buy Quantum disks because they keep dying on me. *Every*
Quantum disk I used at home (three or four) has developed serious
problems; westfalen.de had two quantum disks die in two days (the second
being a replacement for the first), and I'm sure there was something at
work as well. Oh yes, a bigfoot died.
I'm currently a fan of IBM disks. Fairly cheap, too.
Ben
dot+.g...@atat.dotat.org (The Prince of Insufficient Light) writes:
> In article <86mlet$2bk$1...@alfheim.satanic.net>,
> <nic...@esperi.org> writes:
> > What about an E250? A 450 is an altogether bigger beast (and really
> > quick).
>
> We just took delivery of two 450s and six 250s as the new compute
> array. Nice kit, but the rackmounting leaves a lot to be desired.
>
>[snipped and]
>>[also snipped]
Rackmounting 450s. Don't get me started. We had a 19" rack, we had two
450s. Have you ever tried to rack mount a 450 on your own?
Don't.
Have you ever had help rack mounting the top 450, finished, then
realised you had to slide it out to get inside. Have you then found the
rack had not been attached to the ground with anything other than
gravometric forces?
Have you then resolved to mount 450s in pairs front-and-backwards, so
that when you pull one out, the other one comes out the opposite way to
counterbalance the turning forces.
Fully loaded 450s are beasts.
--
"I'm sorry, you must be confusing | For your top-notch JenniCam
me with someone who gives a damn." | parody: http://bofhcam.org
Abbot, Cambridge Chapter of the Monks of Cool since MCMXCVI a.d.
Lars Balker Rasmussen
It wasn't/isn't. I've been meaning to try it out in my CFT.
HiRez
[ Rackery ]
>Fully loaded 450s are beasts.
Thank you for reminding me of my youth.
In them days, even the most badly constructed kit (Midlectron, for instance)
came with those slide-out bars to deal with the leverage of (say) a CDC
Phoenix (80Mb. Now that was *storage*) being slid out so you and three
other chaps could swap it.
--
J "Monty Python's field circususus" H-R
Paul Tomblin
>On Thu, 27 Jan 2000 09:16:09 GMT, Ben <B...@lspace.org> wrote:
>[ Rackery ]
>>Fully loaded 450s are beasts.
>came with those slide-out bars to deal with the leverage of (say) a CDC
>Phoenix (80Mb. Now that was *storage*) being slid out so you and three
>other chaps could swap it.
I remember watching our 6 foot high Sun3 cabinet falling over on the field
circus guy as he slid out the Fujitsu Eagle drive that was down on the bottom.
I don't remember how much that drive held (I suspect it was around 100Mb or
so), but it must have weighed 50 pounds.
--
Paul Tomblin, not speaking for anybody.
"I find your lack of clue...disturbing" - Sithadmin Vader.
SETI@Home: http://www.setiathome.ssl.berkeley.edu/
Me: http://xcski.com/~ptomblin/
Jeff Wolfe
Lars Balker Rasmussen <la...@balker.org> wrote:
>wo...@ems.psu.edu (Jeff Wolfe) writes:
>> In article <0f901dc...@humulus.daimi.au.dk>,
>> Lars Balker Rasmussen <la...@balker.org> wrote:
>> >Uh-huh. Also try copying a lot of stuff from CD to disk/nfs while doing
>> >something on the U5/U10... Or preferably, do it on someone elses U5.
>>
>> Any idea if priority_paging was set?
>
>It wasn't/isn't. I've been meaning to try it out in my CFT.
Turn it on.. It will make a big difference if you do any kind of IO on
the box. We had all sorts of problems with early U5s and 10s because
it's painful to swap on the IDE drive. Once we got enough memory in
them and turned on priority_paging, things got almost reasonable. When
the 440s started coming in U10s with the 7200RPM 9gb drives,
they performed well enough that I decided I might want one of those on my desk.
(I'm stuill using a 170E, though.. :) )
Hou you could build somthing and call it a workstation with a 4500 RPM disk
and a straight face is beyond me.
Jan Chrillesen
> Rackmounting 450s. Don't get me started. We had a 19" rack, we had two
> 450s. Have you ever tried to rack mount a 450 on your own?
Not that the larger Enterprise machines are better. Who the hell decided to
build the 4000 and 4500 a little wider than 19"? We had to buy custom build
racks to mount them.
/Jan
David Iain Greig
>In article <86mna3$fm7$1...@rivendell.apana.org.au>,
>The Prince of Insufficient Light <dot+.g...@atat.dotat.org> wrote:
>>In article <86mlet$2bk$1...@alfheim.satanic.net>,
>> <nic...@esperi.org> writes:
>>> What about an E250? A 450 is an altogether bigger beast (and really
>>> quick).
>>
>>We just took delivery of two 450s and six 250s as the new compute
>>array. Nice kit, but the rackmounting leaves a lot to be desired.
See E220R and E420R servers at www.sun.com
--D.
Tony Finch
>
>Of course, modern DMA versions are supposed to solve that, so it just
>becomes a strange version of SCSI that's limited to two (usually) or four
>devices per bus.
And only one outstanding request on the but at any time (i.e. no
intelligent disk head movement scheduling and no multi-splindle
concurrency).
Tony.
--
i am the dot at dotat dot at
remorse code
>>*I* refuse to buy Quantum disks because they keep dying on me.
>10 years. In fact I've never had an early Quantum SCSI drive go bad on me,
>though they did have some truly horrid MFM drives.
My experience with Quantum Fireball drives (good name, oh yeah! watch
your data go up in an incandescent blob!) has been dismal. Bad
firmware.
rone
--
Insultant: n. Contract worker who gets paid an obscene hourly wage to insult
full-time company employees. <ro...@ennui.org>
remorse code
Peter Radcliffe <26$10$f3i...@pir.net> wrote:
>The Prince of Insufficient Light <dot+.g...@atat.dotat.org> probably said:
>>ridiculously high cost of black goats and sacrificial daggers these
>>days....
>Have you seen the price of a pint of chicken blood these days ? Yeesh,
>it's a wonder we get anything done.
chickenblood.com: i see a great need.
Paul Tomblin
>So what does priority_paging do?
It makes sure your PFY gets paged before you do.
Matt McLeod
>In article <slrn89152o...@darwin.ediacara.org>,
>the Netra telco-certified kit so we can install in colocate space in
>telephone exchanges with impunity.
Ah yes. I have two of those arriving last week.
Or at least they were supposed to. When queried about this,
our Sun rep got the runaround from the factory who didn't
want to talk to him. Eventually he found out that one is
shipping from .us tonight, the other some time next week.
We're not really impressed. I got a bit nasty and mentioned
I was having a meeting with a NetApp rep. About an hour
later we started getting information.
Don't recall exactly what we're getting (I didn't do the
ordering, it got handballed my way later), but they're
both fairly grunty. And probably under-specced for the
task, which is to serve as development/model boxes for
a telco service with some "interesting" design decisions.
(E.g., we have a table with 1,000,000 records, each
approx 6k. We need to go through this table and take
an action on each record. Do we use cursors or other
tricks, or do we try to read the whole fscking lot into
memory? Why they bothered using an rdbms backend I've
*no* idea -- flat files would've been just as effective,
and a lot cheaper...)
Jan Chrillesen
> If rackmounting them you are supposed to take the side panels off; they're
> mostly decorative. Then the things fit precisely in a 19" rack.
Maybe if you use front mounted brackets but we bought the original drawer[1]
brackets from Sun.
/Jan
[1] They come with a base plate that fits on the bottom of the Sun and 2 side
brackets that is tightened to the rack.
Neil Hoggarth
> something on the U5/U10... Or preferably, do it on someone elses U5.
Anyone got any views on the Ultra 10 if they use SCSI exclusively?
I've got a couple of Ultra 1/140 doing a mixed bag of server type
stuff, which we were thinking of replacing with Ultra 10s @ 440MHz,
with SCSI cards. I've got disks in desktop multipacks (6 x 18Gbyte,
fast/wide SCSI, 7200rpm) attached to the current systems, and if I got
a replacement with an IDE drive then I'd ignore the internal disk and
find space for the system partitions and swap on the SCSI disks.
Regards,
--
Neil Hoggarth, BOFH, University Laboratory of Physiology, Oxford, UK
<neil.h...@physiol.ox.ac.uk> http://www.physiol.ox.ac.uk/~njh/
Nicolai
Neil Hoggarth <n...@physiol.ox.ac.uk> wrote:
>> Uh-huh. Also try copying a lot of stuff from CD to disk/nfs while doing
>> something on the U5/U10... Or preferably, do it on someone elses U5.
>
>Anyone got any views on the Ultra 10 if they use SCSI exclusively?
>
>I've got a couple of Ultra 1/140 doing a mixed bag of server type
>stuff, which we were thinking of replacing with Ultra 10s @ 440MHz,
>with SCSI cards. I've got disks in desktop multipacks (6 x 18Gbyte,
>fast/wide SCSI, 7200rpm) attached to the current systems, and if I got
>a replacement with an IDE drive then I'd ignore the internal disk and
>find space for the system partitions and swap on the SCSI disks.
We use Ultra 5s with SCSI cards and SCSI disc (fast/wide, 7200rpm) for
things like webcaches; they don't swap, the OS loads off the IDE disc,
the cache goes on the SCSI discs and it all goes very nicely thank you.
Putting the cache on the IDE disc, particularly when it is the crappy
4500rpm [1] disc, does not work, at all. Not a surprise. Running the
OS and other programs off the IDE disc doesn't seem to matter much;
if your machine pages heavily then you want to do that to the SCSI
discs, obviously.
Nicolai
[1] Just what the F*#@)(%*#@)( do Sun think they were playing at?
I could spin that disc faster with a piece of string around
the spindle !
remorse code
Alan Bellingham <al...@lspace.org> wrote:
>"remorse code" <^#*&$@ennui.org> wrote:
>>My experience with Quantum Fireball drives (good name, oh yeah! watch
>>your data go up in an incandescent blob!) has been dismal. Bad
>>firmware.
>with a Seagate, and it's given me no problems whatsoever.
Sorry, i meant the SCSI Fireballs. I've not used the IDE variety.
David Iain Greig
>
>When did they happen?
Very Recently. Last month or so I could order them. The PCI bus slot
count is a bit low. Coming soon with be the UltraSPARC III cpu's
and Solaris 8. (excepted from our onsite presentation by Sun Canada)
--D.
Jeff Wolfe
Nicolai <nic...@alfheim.satanic.net> wrote:
>In article <86shpa$sjr$1...@news.ox.ac.uk>,
>Neil Hoggarth <n...@physiol.ox.ac.uk> wrote:
>>> Uh-huh. Also try copying a lot of stuff from CD to disk/nfs while doing
>>> something on the U5/U10... Or preferably, do it on someone elses U5.
>>
>>Anyone got any views on the Ultra 10 if they use SCSI exclusively?
>>
>>I've got a couple of Ultra 1/140 doing a mixed bag of server type
>>stuff, which we were thinking of replacing with Ultra 10s @ 440MHz,
>>with SCSI cards. I've got disks in desktop multipacks (6 x 18Gbyte,
>>fast/wide SCSI, 7200rpm) attached to the current systems, and if I got
>>a replacement with an IDE drive then I'd ignore the internal disk and
>>find space for the system partitions and swap on the SCSI disks.
That's more or less what we do. We have a few U5s and 10s with SCSI
cards and 10krpm disk packs. Seems to be a reasonable compromise
between somthing like a 450/250 with a small budget.
>We use Ultra 5s with SCSI cards and SCSI disc (fast/wide, 7200rpm) for
>things like webcaches; they don't swap, the OS loads off the IDE disc,
>the cache goes on the SCSI discs and it all goes very nicely thank you.
>Putting the cache on the IDE disc, particularly when it is the crappy
>4500rpm [1] disc, does not work, at all. Not a surprise. Running the
>OS and other programs off the IDE disc doesn't seem to matter much;
>if your machine pages heavily then you want to do that to the SCSI
>discs, obviously.
About the only thing we do with the IDE drives is boot from them.
>[1] Just what the F*#@)(%*#@)( do Sun think they were playing at?
> I could spin that disc faster with a piece of string around
> the spindle !
Heh! The early U5s were just cheap. Once the 2MB cache and higher
speed processors became available along with a 7200rpm disk, things got more
reasonable.
-Jeff
Matt McLeod
>In article <slrn89431m...@darwin.ediacara.org>,
> gr...@ediacara.org (David Iain Greig) writes:
>> Very Recently. Last month or so I could order them. The PCI bus slot
>> count is a bit low. Coming soon with be the UltraSPARC III cpu's
>> and Solaris 8. (excepted from our onsite presentation by Sun Canada)
>
>those bastards at Sun point us at these?
First we heard of these was while I was talking to one of their
engineers, joking that they could always just take the troublesome
E4k back and give us a pair of E450's...
I may have to remind $BOSS about these before she finalizes
the new purchasing proposal.
--
"A satisfied customer? We ought to have him stuffed!"
-- Basil Fawlty
David Iain Greig
>In article <slrn89431m...@darwin.ediacara.org>,
> gr...@ediacara.org (David Iain Greig) writes:
>> Very Recently. Last month or so I could order them. The PCI bus slot
>> count is a bit low. Coming soon with be the UltraSPARC III cpu's
>> and Solaris 8. (excepted from our onsite presentation by Sun Canada)
>
>Fuck. Last month was when we ordered the 450/250 array. Why didn't
>those bastards at Sun point us at these?
Dunno, they've been on www.sun.com longer than they've been available.
We just got the 'what's new in 2000' talk. They UltraSPARC III will
be brought out in a desktop box first, then the E250/E450 class, then
finally the enterprise servers. I gather. The plan for the E10000
is murky, cuz apparently 64 UltraSPARC III's will pump out more data than
the E10000 bus architecture can move... so they need to redesign the bus.
Oh, the prices for the E250/220 and E450/420 look about the same, so
they didn't avoid them to screw you on price.
E450 is nice, but toooo big. Tho I like 10 PCI slots:
Typical E450 server cared mix for me is:
Ethernet
Spare Ethernet (in case NIC #1 fails, we're mildly paranoid)
FWD SCSI #1 for EMC/whatever/JBOD/etc
FWD SCSI #2 for EMC/whatever/JBOD/etc
ATM Card for Netbackup LAN
PGX frame buffer
So I use 6 slots on a typical machine bus. Too many servers with only
4 pci slots in the world. Bah.
If its a VCS or Sun HA server, add a qfe card.
Oh, I resigned today too. :) New job in 2 weeks.
Al Castanoli
>In article <86degk$24pb$1...@citadel.in.taronga.com>,>>I'd much rather trust an industrial PC running QNX on a low-power IBM/Cyrix
>>intel clone processor than a Sun workstation plugged into a rackmount shelf.
>
>I seem to be banging on Peter (da Silva) a lot... sorry, but you're raising
>the points I care about :)
>
>How is remote manageability on these? Can I use a serial console, network
>boot it, boot off alternate disc, boot off CDROM, etc? Does it have
>environmental monitoring and a PROM or lowlevel monitor [2] that I can get
>to with a key sequence to power cycle the box if it's locked up for some
>reason?
It's because of their remote manageability that industrial - grade PC
boxen running QNX are used by one of our local hospitals [1] for ambulatory
EEG monitoring. These aren't department store PC's, but have several extra
features such as heat/humidity and case airflow (those cheetah disks run hot)
monitors available to the OS. If one of the power supplies smokes, the 2ndary
kicks in autotragically. I've seen Sun kit used in such installs, but this
QNX set has been chugging along for about 5 years now, with only memory, disk,
and processor upgrades.
[1] Baptist Hospital in San Antonio, Texas, USA
--
Al Castanoli | afc...@ucan.foad.org | {cheap}
| satlug.org | You can have it {quick} Pick any 2.
| texas.net | {right}
Kai Henningsen
> In article <7Xd9r...@khms.westfalen.de>,
> >*I* refuse to buy Quantum disks because they keep dying on me.
>
> I've got some Quantum LPS* series drives that are still going strong after
> 10 years. In fact I've never had an early Quantum SCSI drive go bad on me,
> though they did have some truly horrid MFM drives.
Yes, I think the _early_ Quantums were nice. That was when they made their
name.
Matthew Byng-Maddick
>In article <38a46d64....@news.lspace.org>,
>Alan Bellingham <al...@lspace.org> wrote:
>>"remorse code" <^#*&$@ennui.org> wrote:
>>>My experience with Quantum Fireball drives (good name, oh yeah! watch
>>>your data go up in an incandescent blob!) has been dismal. Bad
>>>firmware.
>>Hmm, different. I've got a Fireball in this machine, on IDE together
>>with a Seagate, and it's given me no problems whatsoever.
>Sorry, i meant the SCSI Fireballs. I've not used the IDE variety.
I heard that they were all pretty bad - a friend of mine did some
climate testing on the drives, and the Quantum ones, especially the FBs
failed most of the tests...
I refuse to buy Quantum discs now.
MBM
--
/* Matthew Byng-Maddick */char m[8192];y(int n){m[n>>3]|=1<<(n&7);}x(int n
/* m...@colondot.net */){return m[n>>3]&(1<<(n&7));}main(int g,char **s
){if(--g){int a=atoi(*(++s)),i= 2,j;memset(m,85,8192);*m=83;for(;i<a;i++){
j=i;if(!x(i)){while(j<a)y(j+=i);printf("%d\n",i);}}exit(0);}else exit(1);}
Lionel
>In <389dea48....@news.lspace.org> al...@lspace.org (Alan Bellingham) writes:
>>?! I've never encountered this as an IDE problem.
>
>I refuse to buy Quantum disks thanks to this problem. Whilst at
>the time (about 5-6 years ago) Quantum disks were quite fast,
>you could have *real* problems getting them to be friendly
>with any other disks on the IDE chain. This was often true even
>if you had two identical Quantum disks (ie same revisions, etc).
>If you *did* get them to work together then it was normally
>thanks to dark rites.
*grin*
Specifically; tweaking the master select & spindle sync control jumpers,
amongst others.
The hardest part was often figuring each manufacturer's acronym for
their proprietry name for any of the controls.
That said, I never did manage to get a couple of specific drive
combinations to work, although my success rate was better than 95%.
--
W
| ,. w , "Some people are alive only because
\|/ \|/ it is illegal to kill them." Perna condita delenda est
---^----^---------------------------------------------------------------
Victoria O'Staff
>In article <rone.86t09s$mqe$1...@ennui.org>, remorse code <^#*&$@ennui.org> wrote:
>>In article <38a46d64....@news.lspace.org>,
>>Alan Bellingham <al...@lspace.org> wrote:
>>>"remorse code" <^#*&$@ennui.org> wrote:
>>>>My experience with Quantum Fireball drives (good name, oh yeah! watch
>>>>your data go up in an incandescent blob!) has been dismal. Bad
>>>>firmware.
>>>Hmm, different. I've got a Fireball in this machine, on IDE together
>>>with a Seagate, and it's given me no problems whatsoever.
>>Sorry, i meant the SCSI Fireballs. I've not used the IDE variety.
>
>I heard that they were all pretty bad - a friend of mine did some
>climate testing on the drives, and the Quantum ones, especially the FBs
>failed most of the tests...
We used to have a bunch of Quantum Grand Prix here at work, and they
would crash and burn on a regular basis. It actually became a sort
of grim and bitter joke. :-/
OTOH, my (IDE) Fireball at home I've had since mid-college; it's never
been problematical.
--
Victoria Swann O- | "'Nice' in a bodyguard is about as useful as
Panix Support Wench | the ability to regurgitate whole lobsters."
Work-stuff: v...@panix.com | --Marquis de Carabas, from _Neverwhere_
Other-stuff: to...@panix.com | Outlaw spam! http://www.cauce.org
Georg Bauer
In article <7XWpk...@khms.westfalen.de>, Kai Henningsen wrote:
>26$10$f3i...@pir.net (Peter Radcliffe) wrote on 23.01.00 in <86g0jd$sn6$1...@moek.pir.net>:
>
>> No, I meant 16550 rather than 16550A.
>You mean to say you actually saw that beast? So far, it's existence hasn't
>been more than theoretical for me.
It was used in early IBM PS/2 machines (model 60 for example). No sane PC
builder ever used them, except IBM. But then, you could argue days for or
against the sanity of IBM ...
>Impossible to diagnose from here, but I can't imagine you not doing
>something seriously wrong. It certainly doesn't match my experience.
Yep. I do use fairly old and slow machines as servers - I started with an
386/25 as our main proyxy at home and upgraded that to a 486/66 and just
recently to a Pentium 120. Even the really old machines where quite useable
when under high load (ok, not for kernel compilation, but you better use very
fast machines for that).
If you happen to run an OS written by people who know what they are doing
does help a lot.
BTW: I am currently using my Sun 3/50, running NetBSD. It is far more useable
than when running it under SunOS. And now it even doesn't have a disk of it's
own. It boots and swaps over NFS.
Using the right OS often helps in overcoming the problems of sucky hardware.
Although the 3/50 isn't too sucky. (Except the really bad serial ports, but
the only thing I do over serial line nowadays is using the RCX IR tower).
bye, Georg
Georg Bauer
In article <7Xd9r...@khms.westfalen.de>, Kai Henningsen wrote:
>si...@bpfh.net (Simon Burr) wrote on 26.01.00 in <86n2bd$juj$1...@triad.bpfh.net>:
>I'm currently a fan of IBM disks. Fairly cheap, too.
But provide the right cooling, or they will die faster than any Quantum ...
Kai Henningsen
> In article <7Xd9r...@khms.westfalen.de>, Kai Henningsen wrote:
> >si...@bpfh.net (Simon Burr) wrote on 26.01.00 in
> ><86n2bd$juj$1...@triad.bpfh.net>: I'm currently a fan of IBM disks. Fairly
> >cheap, too.
>
> But provide the right cooling, or they will die faster than any Quantum ...
Depends on which ones. As always, the thicker the disk, the more heat it
produces.
They don't seem to die on me, anyway. (Guesses as to how long until I'm
proven wrong?)
Jasper Janssen
>I've had a bunch of machines with 16450 parts, I don't know what the exact
>difference is between the 8250, 16550A, and 16450, but IIRC the 16450 was
>better than the 16550(non-A).
I'm not entirely sure about the 16550 (nonA), as I've never seen one
in the wild, but 16450 sure has the damnedest time doing 115k2 serial.
As a result, my cablemodem has been running at half maximum speed for
a year now. Poo on serial-port cablermodems.
Jasper
Jasper Janssen
>(At least I think it's a Seagate - it just says it's an ST34342A. The
>hard drive on SCSI says it's a Seagate ST31230N, and those are very
>similar numbering systems.)
That'd be a Seagate. If you want, I can go look up in my "upgrading
and repairing PCs, third edition" exactly what they mean. IIRC:
STnxxxxl,
where: ST always, n is a number for the form factor (3 means 3.5 inch,
etc.), xxxx is capacity, l is a letter signifying type. Type can be
IDE, MFM, SCSI, SCSI with kit, etc., each with a capital letter
assigned to them.
Jasper
Jasper Janssen
>Using the right OS often helps in overcoming the problems of sucky hardware.
>Although the 3/50 isn't too sucky. (Except the really bad serial ports, but
>the only thing I do over serial line nowadays is using the RCX IR tower).
And you hang that off Sun serial ports? Shame on you -- your RIS
should strangle you.
Jasper
Matthew Byng-Maddick
> Simon Burr <si...@bpfh.net> writes:
>> at http://www.seagate.com/support/kb/disc/discmodelinter.html for the
>> full details on Seagate drive model numbers.
>OK, who else attempted to parse this as "disc mode linter"?
<fx: raises guilty hand>
MBM
--
/* Matthew Byng-Maddick */char m[8192];y(int n){m[n>>3]|=1<<(n&7);}x(int n
/* m...@colondot.net */){return m[n>>3]&(1<<(n&7));}main(int g,char**s)
{if(--g){int a=atoi(*(++s)),i=2,j;memset(m,85,8192);*m=83;for(;i<a;i++){j=
i;if(!x(i)){while(j<a-i)y(j+=i);printf("%d\n",i);}}exit(0);}else exit(1);}
Jasper Janssen
>what disk you're holding. I just wish others did the same (or do
>they, and it's just do obfuscated that I can't figure it out?)...
Most of them seem to do, except that the defintions change more.
Frex, I have WD caviars named "Caviar 280", a few named "Caviar 2850",
and also a few named "Caviar 34000". 80, 850, and 4000 megabytes
respectively. Not entirely sure what the 2s and 3s mean, since they're
all IDE disks. Maybe IDE sec vs PIO4, or something like that.
Jasper
Tony Finch
>[ Followups to the place of discussion of notworking matters ]
Hmm, I think my feed is inadequate.
>Of course, in the third world Fascist country of the United Kingdom,
>we just about have phone lines for our internet connectivity and ADSL
>is just being trialled.
I've been using ADSL for two years at home. It's nice to work for an
ISP that does that sort of thing for its employees. (Fuck-all else,
though.)
Tony.
--
** **
*** *** *** **** *** ******* *** *** **** *** ****
*** *** * *** *** *** * ***