ephraim@techunix.BITNET (Ephraim Silverberg) (03/22/88)
As requested, I am summarising the responses to my original query.
For those who didn't see my first posting, here are the questions I
posed:
We are faced with a choice of getting either a Sun 3/260CXP or
one of HP 320SRX, HP 350SRX and HP 825SRX. From the
literature, it seems that the Sun 3/260CXP is comparable with
the 350 with the 320 being weaker and the 852 being twice as
strong (though weaker than a Sun 4); is this correct?
I would appreciate any input from people who have used these
machines regarding the quality of the HP-UX Unix system, the
graphics quality and speed and the computational speed
(particularly floating point), and any general impressions of
user friendliness, learnability, etc.
I received four responses. Since these responses were sent to me as
private e-mail, I am not including the names or e-addresses of the
respondants as the opinions they expressed may put them in a
problematic situation at their place of employment.
Response #1:
I haven't used HP systems much myself, but various people
around here have. They have gained a reputation for having
idiosyncratic software. They tend to end up as doorstops.
Response #2:
Article is from "ephraim@techunix.BITNET (Ephraim Silverberg)":
> We are faced with a choice of getting either a Sun 3/260CXP
> or one of HP 320SRX, HP 350SRX and HP 825SRX...
I am also interested in such information and I am sure other
people on the net will be too. Perhaps you could summarise any
responses you get in a posting to the net? Thanks.
One observation I had is that I assume the Sun uses a 4.2 file
system where as HP-UX probably uses the SYS 5 file system. For
large files the sys 5 file system can be 10 times slower than
4.2. (I have heard thi from three quite independent sources).
For small files its not as bad. I however am working on a
database project and we want a workstation to play with nicer
interfaces to the database so disk throughput is very
important.
Response #3:
In article <8802111716.AA29397@techunix.bitnet> you write:
>We are faced with a choice of getting either a Sun 3/260CXP
>or one of HP 320SRX, HP 350SRX and HP 825SRX. From the
>literature, it seems that the Sun 3/260CXP is comparable with
>the 350 with the 320 being weaker and the 852 being twice as
>strong (though weaker than a Sun 4); is this correct?
I agree with your observations. I can't give you alot of input
re: graphics comparisions because that's not my forte, nor is
it what we ~heavily~ do with our Suns and HPs. To me it comes
down to a question of BSD vs. SYSV. Do you like (or are you
used to) the look and feel of a BSD box or the smell (sic) of
SYSV box ? If networking at all plays a role I give the Sun
box the nod because of it's "state of the BSD" (4.3)
networking and it's networking extensibility. We often put new
hardware or software interfaces or protocol stacks into our
Suns boxes and kernels. Because of the SYSVisms on the HPs
this will be much more difficult. Also, HP is still
~basically~ running BSD4.1 networking code. HP promised 4.3
code in January, but it is now March. HP is still in it's beta
phase for NFS.
I personally dislike HP hardware packaging. So many "boxes"
with zillions of cables all over the place all using
proprietary busses and interfaces. HP disk throughput is VERY
slow because it's over GPIB. You also ~pretty~ much have to
buy all peripherials from HP because of their packaging and
interfaces. They even use some weird mouse that requires an
"adapter" box to hook it up to their system. In retrospect, I
like Sun hardware packaging, for desktop, deskside, and rack
mountable systems. Sun is also expensive with peripherials (if
you buy them from Sun), but we buy them much cheaper from
third party vendors.
Sun is very much on the leading edge in the UNIX kernel arena.
SunOS 4.0 will have an all new memory management scheme, file
mapping (great for those huge graphic data files), and shared
libraries. They also have a butt-kicking window based source
level debugger called dbxtool that make debugging a pure
pleasure.
I know BSD and SYSV are converging, but the last thing we need
is another angle. HP-UX has a bunch of HP specific stuff in
the kernel, utilities, and libraries that I think we could
live without. I find BSD to be a superset of SYSV, so to me
the logical approach is to take a BSD box and add SYSV
extensions (the way Sun does it), not the other way around
(the way ~I~ think HP has done it.)
I know Sun keyboards aren't great, but no one I know has ever
liked the HP keyboard. May seem trivial, but how do you think
I'm typing this in ?? Every second you're on the computer,
the keyboard is your most important tool.
One area HP does shine in is sales and support. We've had
LOUSY Sun sales reps right from day one. You can count on HP
for that aspect.
Well, I've probably said too much, but these are my
thoughts/opinions.
Response #4:
I was a programmer/support person for some HP9000/320s at
Dept. And I've also worked a tiny bit on Suns, though mostly
Sun-2s. The one thing I have to say about both systems are
that the hard drives are incredibly flaky. Especially HP's
equipment, we and other sites were constantly rebooting
systems, and even replacing drives completely. Of about 50
drives in use, left powered up twenty-four hours per day, we
had a disk crash at least every other day. I'd guess we had to
replace a drive entirely every other week, perhaps more often.
I mean, we had bad drives stacked like cordwood.
One definite cause was we originally had them too close
to the workstations where users might bump them. Crashes
dropped dramatically when we placed them more out of the way,
though I still consider the failure rate absurd. Since the
department head once worked for HP, and was tight enough with
them to get all of that equipment donated to us, we were
encouraged to keep our heads down about the matter. After all,
HP did replace all of the bad drives free.
One other point, I have been told, though I'm not sure
how strongly I believe, that the flakiness of the drives was
because HP farmed the manufacture of the drives out.
Furthermore, a new subcontractor has taken over and is doing a
much better job.
Anyways, enough about them. On to Sun equipment. Sun too
isn't exactly the stablest stuff. Their drives seemed to have
too many failures, but what was really flaky were the video
boards. To the point that we (and other sites) depleted not
only our local supply of extras, but Sun ran short of them
too. This was all Sun-2 equipment, so they may have fixed the
problem in later models.
And on to what you probably really want to here:
performance. There, I can't really compare Sun-3s to HPs,
since I never really worked on the Sun-3 that much. I will
mention my thoughts on the HP, though. First off, the quality
of the displays was superb. When I first saw someone plot
something on these little CRTs that I *thought* was a standard
monchrome, my jaw dropped out. And then we got the big 19"
Trinitron color displays. Wow.
Of course, you'll have to consider your own particular
needs. We originally had four bitplanes of color. For CAD of
VLSICs, no go. We upgraded to six, and pounded on the software
a while, and voila'. I believe Sun has seven bit planes as the
maximum in their machines.
We also had a 68881 FP coprocessor in our machines, which
helped things along nicely. Speaking of which, both the Sun-3
and HP9000/320 have 68020s, but as anyone who has seen Hp's
code can tell you, HP can't write software worth beans. I
mean, they really stink at it. So I'm not suprised that their
machine isn't anywhere near Sun's. You of course have to
consider other things, ie. memory wait states, etc.
But back to HP-UX. It's sort of Berkeley 4.[12] and AT&T
SysV mashed togther, and not very well. Some of the code is
actually from SysIII. We at C-MU actually gave up, and ported
Berkeley 4.2 BSD Unix to the things. HP-UX is really pretty
brain-damaged.
For instance, they once decided for who knows what
reason, to put both the swap and root portions of disk on the
same partition. Well, the swap area is the stuff most likely
to get corrupted, ie. trashed in a power glitch. Being on
the same partition as root, instead of it's own, meant that if
it was corrupted, you lost the root information also. Which
means you've lost the entire disk. Time to reformat. Never
mind that you've still for the user information there, it's no
good without the root partition. Don't ask me why, I thought
we'd be able to get something back, but nope. No dice. It's
all gone.
Finally, I have to point out that I am a Berkeley Unix
fan. SysV disgusts me. And HP-UX really gets me going. So
consider all my comments with the knowledge that I prefer Sun
simply because I hate SysV. And that is finally all I have to
say. Sorry it's so long, but I figure you'll just pitch
anything you aren't really concerned about.
Hope this summary is useful.
---------------------------------------------------------------------------
Ephraim Silverberg,
Faculty of Biomedical Engineering,
Israel Institute of Technology,
Haifa, Israel.
BITNET : ephraim@techunix
ARPANET : ephraim%techunix.bitnet@cunyvm.cuny.edu
CSNET : ephraim%techunix.bitnet@csnet-relay.csnet
UUCP : {ihnp4, allegra, rutgers}!psuvax1!techunix.bitnet!ephraim
---------------------------------------------------------------------------rml@hpfcdc.hp.COM (Bob Lenk) (03/25/88)
I'd like to correct a few points. Some may be based on old information. I will not argue with anyone's experiences or opinions. > One observation I had is that I assume the Sun uses a 4.2 > file system where as HP-UX probably uses the SYS 5 file > system. The HP-UX machines use a 4.2BSD-based file system. It has been modified In a few ways to give an application appearance of a Sys V file system, but is equivalent to 4.2 in performance. The main modification was in file name length, since many applications have a 14 character limit hard-coded. However, the recent 2.0 release for the series 800 permits a choice of 14-character or 255-character file names by file system. > this will be much more difficult. Also, HP is still > ~basically~ running BSD4.1 networking code. HP promised 4.3 > code in January, but it is now March. HP is still in it's > beta phase for NFS. There has never been 4.1 networking code in HP-UX. The current releases are all based on 4.2 (with various enhancements and bug fixes). I don't know anything about 4.3 promises, and comment on that. NFS is released for both the series 300 (December) and 800 (March), but may not be in customer hands in all cases. > the logical approach is to take a BSD box and add SYSV > extensions (the way Sun does it), not the other way around > (the way ~I~ think HP has done it.) HP-UX was implemented by starting with 4.2BSD. We chose to modify it to look indistinguishable from System V by default. Sun chose to be pure BSD and support System V via libraries. Both approaches have merit. > For instance, they once decided for who knows what > reason, to put both the swap and root portions of disk on the > same partition. Well, the swap area is the stuff most likely > to get corrupted, ie. trashed in a power glitch. Being on > the same partition as root, instead of it's own, meant that > if it was corrupted, you lost the root information also. > Which means you've lost the entire disk. Time to reformat. The disk is laid out with a file system at the front, and the unused portion beyond that is swap space. The swap space is not part of the file system, and its content is of no interest across reboots or power failures (or to fsck). Of course I speak for myself, not HP. Bob Lenk {ihnp4, hplabs}!hpfcla!rml
irf@kuling.UUCP (Bo Thide) (04/01/88)
In article <8803221250.AA02228@techunix.bitnet> ephraim@techunix.BITNET (Ephraim Silverberg) writes: > One observation I had is that I assume the Sun uses a 4.2 > file system where as HP-UX probably uses the SYS 5 file > system. For large files the sys 5 file system can be 10 times > slower than 4.2. (I have heard thi from three quite > independent sources). On their 9000/800 and 9000/300 series HP uses the BSD "High Performance" file system for all supported OS's (HP-UX, Pascal, BASIC, 4.3 BSD). > is another angle. HP-UX has a bunch of HP specific stuff in > the kernel, utilities, and libraries that I think we could > live without. I find BSD to be a superset of SYSV, so to me Without the "HP specific" Real-Time Enhancements and Device I/O Library I would be out of business ... > Sun-2s. The one thing I have to say about both systems are > that the hard drives are incredibly flaky. Especially HP's > equipment, we and other sites were constantly rebooting I personally have used various HP computers and disk drives for more than five years under very extreme environmental conditions running field experiments in the arctic (cold and damp weather, 220V diesel generators that just barely work) and in tropical jungles (where just ANYTHING can happen). Not one single second have I had any problems whatsoever with any HP hardware, including disk drives! I can tell you that the way the equipment is being handled during air and truck transportation is just a shame. So, I very much doubt that just the placement of the disks near the computers has anything to do with the problems reported. I might add that I have used various computers for scientific/technical applications for more than 20 years (and used them A LOT!), that I am just an ordinary assistant professor but with special responsibility for computer soft- and hardware in our institute (we have non-HP computers too; needless to say, we feel they are inferior to our HP's :-> ) and that my only connection to HP is that of being a (quite pleased) user of various HP equipment. -Bo -- >>> Bo Thide', Swedish Institute of Space Physics, S-755 90 Uppsala, Sweden <<< Phone (+46) 18-300020. Telex: 76036 (IRFUPP S). UUCP: ..enea!kuling!irfu!bt