Sun-Spots Digest, v6n20
William LeFebvre
Sun-Spots-Request at RICE.EDU
Mon Feb 29 04:45:55 AEST 1988
SUN-SPOTS DIGEST Wednesday, 24 February 1988 Volume 6 : Issue 20
Today's Topics:
Corrected machine name and problems with "atrun"
Re: 800 bpi tape drive
Re: Lisp vs. C Floating Point Performance (Suns)
Re: sun-4 bcopy
Want to buy Motorola manuals? You can order direct!
Cheap(?) SCSI drives
More on NFS timeouts between Suns
Various problems
Mysterious ethernet misbehavior
Simulation package info wanted
Looking for sun TeX implementation(s)
nfsd problems/performance?
Availability of Used SUN-3's?
Alternate Pascal Compilers?
Artecon - Sun Clones - Has anyone experience with them?
DEC LAT protocol?
Scanned image into a SUN?
Send contributions to: sun-spots at rice.edu
Send subscription add/delete requests to: sun-spots-request at rice.edu
Bitnet readers can subscribe directly with the CMS command:
TELL LISTSERV AT RICE SUBSCRIBE SUNSPOTS My Full Name
Recent backissues are stored on "titan.rice.edu". For volume X, issue Y,
"get sun-spots/vXnY". They are also accessible through the archive
server: mail the word "help" to "archive-server at rice.edu".
----------------------------------------------------------------------
Date: Thu, 25 Feb 88 10:19:10 CST
From: William LeFebvre <phil at Rice.edu>
Subject: Corrected machine name and problems with "atrun"
In a recent article about C++ (v6n16), one of the machines at MIT was
referred to with an incorrect name: AI.PREP.EDU should really be
prep.ai.mit.edu. Sorry about that.
Also, I have heard through the grapevine (no, not the one at Xerox) that
/usr/lib/atrun is broken in both SunOS 3.5 and Sun4's sys4-3.2FCS. The
problem is caused by a change to sprintf which brings it more in line with
the ANSI standard, but also removes an undocumented "feature" that atrun
uses. For Sun3's, the fix is simple---use the atrun from 3.4. Those with
Sun4's are out of luck on this one until 4.0 is distributed. Sun is now
aware of the bug and will have it fixed in future releases.
How many people really use "at" anyway?
William LeFebvre
------------------------------
Date: Sat, 13 Feb 88 15:31:05 PST
From: hoptoad.UUCP!gnu at cgl.ucsf.edu (John Gilmore)
Subject: Re: 800 bpi tape drive
The TapeMaster controller board used in older Suns for 1600 bpi drives
will work fine with an 800 BPI tape drive, without driver changes as far
as I know. I did this myself on an ancient Sun release (e.g. probably
Sun Unix 1.0 or so). It's possible that more recent drivers do not work
at 800, but it seems a good thing to try first.
John Gilmore
------------------------------
Date: Sun, 14 Feb 88 19:32:21 PST
From: eggert at sm.unisys.com (Paul Eggert)
Subject: Re: Lisp vs. C Floating Point Performance (Suns)
In Sun-Spots v6n14 Malcolm Slaney (malcolm at spar.slb.com) concludes that
Lucid Lisp 2.1.1 generates Sun floating point code that runs as fast as
fast as C. I looked at his benchmark, which was Lisp code blindly
translated to C, and found that minor improvements to the translation
(please write for details) changed the results so that C is 1/3 faster:
Time (in seconds) to execute 10 iterations of a 1024 point FFT
Sun-3/160 68881
single double
C (SunOS 3.4) 3.5 3.7
Lucid 2.1.1 4.7 ?
I'm happy to see Lucid Lisp improving above its formerly woeful
performance, and it's important to say that floating point need not cause
one to shun Lisp. But I'm not yet convinced that Lucid Lisp and Sun C
have similar floating point performance, even ignoring the above results.
First, Slaney reported no Lucid times for FPA-equipped Sun-3s or for
Sun-4s, even though the hardware was available; what is the problem here?
Second, many Lisp systems don't support fast double precision, which is
crucial for many applications. Can the question mark in the table above
be replaced by a hard number, so that we can see how well Lucid handles
double precision?
Floating point performance is just the tip of the C vs Lisp iceberg; for
more ice see Howard Trickey, "C++ versus Lisp: A Case Study", SIGPLAN
Notices 23, 2 (February 1988), 9-18.
------------------------------
Date: Tue, 16 Feb 88 05:28:43 EST
From: dan at bellcore.com (Daniel R. Strick)
To: sun-spots at rice.edu
Subject: Re: sun-4 bcopy
On the sun-4,
to[0] = from[0];
...
to[7] = from[7];
to += 8; from += 8;
is noticeably faster than
*to++ = *from++;
...
*to++ = *from++;
(I don't think there is an autoincrementing address mode in SPARC.
Standard load/store addressing does seem to include constant offsets.)
------------------------------
Date: Mon, 15 Feb 88 09:41:26 PST
From: dgh at sun.com (David Hough)
Subject: Want to buy Motorola manuals? You can order direct!
Motorola doesn't sell Sun workstations, and Sun doesn't sell Motorola
68020 and 68881 manuals. Fortunately if you want the latter you can buy
them fairly cheaply direct from Motorola. Apparently a single copy of
either manual sells for $7.20 including postage within US. The address
is:
Motorola Literature Distribution Center
PO Box 20912
Phoenix, AZ 85036
Phone number (602) 994-6561.
------------------------------
Date: Sun, 14 Feb 1988 11:18:56.76 EST
From: <droms at BKNLVMS.BITNET>
Subject: Cheap(?) SCSI drives
I recently heard a pitch about Trimarchi SCSI drives for Suns. They sell
a two drive enclosure for $750 and drives to plug into the enclosure that
run about $1250/100Mb (e.g., 320MB for $4k, 160Mb for $2k). A 640MB
system sells for $8700. There may be some educational discount available.
Does anyone have any experience with Trimarchi systems? Are these prices
competitive with other add-on SCSI drives? How much would a fileserver's
performance degrade with SCSI disks instead of SMD disks?
- Ralph Droms
Bucknell University
droms at bknlvms.bitnet
------------------------------
Date: Sun, 14 Feb 88 23:39:44 EST
From: dpz at aramis.rutgers.edu (David P. Zimmerman)
Subject: More on NFS timeouts between Suns
Chuck Hedrick recently pointed out our problems with Sun 4 NFS overrunning
Sun 3 NFS. I would like to place an addendum on that - we recently
installed our first Sun 3/60 here (after painfully ripping it off of my
"test desk" :-). It is running SunOS 3.2, and NFS/ND serving off of a
3/140 with SuperEagle, which has 10 clients and is also running SunOS 3.2.
The primary user of it started complaining of the server not responding
and then saying OK, and sure enough, the 3/60 is fast enough to overrun a
3/140 NFS server, even at low network and server load. Starting only 2 of
the 4 default biods seems to keep things reasonable. We point
/etc/rc.local off to /pub/etc/rc.local here, and a simple hack to look for
/etc/sun3-60 (a copy of /bin/true on that client) and then only start 2
biods made the setup fairly elegent and easy to maintain if we decide to
add more 3/60s to that server. The effect of multiple 3/60s hanging off
of that one poor server, though, may force us to still lower numbers of
biod or possibly higher product numbers of server.
------------------------------
Date: Thu, 11 Feb 88 09:48:18 CST
From: vuse!atk at uunet.uu.net (Alan Krantz)
Subject: Various problems
I am having several unrelated problems and was wondering if you
(networld?) might have some solutions:
1) I think this was recently mentioned - but can't find it in the back
issues: From time to time I do a remote login into a sun 3/xxx and my
page size is set to 37 (as if it were the console) when it should be 24
- (my term type is set to vt100). If I do a setenv TERM is indeed set
to VT100, though if I type set - term is set to su. But the problem
with the page size is not consistent though the manner of login and the
terminal being used is. First, does anyone know why this happens.
Second, once I'm logged in is there a way to reset the pagesize?
[[ This has indeed been discussed recently. There are additional, poorly
documented, fields in the tty driver structures: rows and columns. The
command "stty everything" will reveal their values. Termcap will honor
these values before it looks in the corresponding termcap entry for "li"
and "co". To get the default values, "stty columns 0 rows 0". If they
are 0, termcap will look in the entry for the values. --wnl ]]
2) We have a tape drive - whose name/type/... I can't recall. In any event
the problem is that we want to read 800 bpi. What is confusing me is
that we can physically set the BPI on the tape drive - if we set it to
1600 all works fine (i.e., cat file > /dev/rmt8) - file can be read
from the tape. However, if we set the drive to 800 all we get is
I/O error from cat. On the console we get an error message to the
effect - error on device xt0 - ??: 0 err: 32. For some reason, I
was under the impression that BPI was a function of the device and
the driver simply dumped the data to the device (though I might need
to know how much data to dump per block - I'm not sure)... I realize
that more information is probably needed - but given the above does
anyone have some suggestions for using the drive at 800 BPI?
[[ Try using the device /dev/rmt0 instead. Rmt0 thru rmt7 are for "low
density" and rmt8 thru rmt15 are for "high density". --wnl ]]
3) We have an apple laserwriter and would like to dump RASTERFILE(5) to
it. However, adobe does not supply a filter to do such. Does anyone
have such a filter or suggestions as to where one can be purchased?
Somtimes the mailman fails to deliver these digests to me - so I would
appreciate it if replys could be CC directly to me.
Thank you,
Alan Krantz
------------------------------
Date: 15 Feb 88 16:01:42 GMT
From: ksr!benson at uunet.uu.net (Benson Margulies)
Subject: Mysterious ethernet misbehavior
We have an ethernet with on the order of 10 servers and on the order of 40
clients. All of the machines save a printer, a bridge terminal server, and
a kinetics are Suns.
Most of the machines are connected to ISO-lan fan-outs, which in turn have
vampire tranceivers on the cable.
About every two to ten minutes, there is a spike in net traffic that we
have not been able to correlate with valid packets. Machines sending
packets during the spike get collisions -- bursts of 500 to 2000 of them.
Our suns print any or all of:
ie0: no carrier
ie0: ethernet jammed
NFS: server foobar not responding
Sun swears that the first two have to mean cable problems, but the time
varying nature of the problem seems to contradict this. Also, the spikes
can be observed in the middle of the night, when nothing else much is
going on. (We don't run rwho.) Just in case, we are planning a TDM session
quite soon.
We have swapped out nearly all of our tranceivers, and will continue to do
so.
Are there any "old ethernet hands" out there to whom these symptoms point
a clearer finger? For that matter, does anyone out there make a business
of consulting on this kind of stuff? Replies to the this question by mail,
please.
Benson I. Margulies Kendall Square Research Corp.
harvard!ksr!benson ksr!benson at harvard.harvard.edu
------------------------------
Date: 12 Feb 88 0:28 +0800
From: "Mark J. McIntosh" <mmcintos%uvunix.uvic.cdn at UBC.CSNET>
Subject: Simulation package info wanted
I am looking for a simulation package to use on Sun-3/52M workstations.
We are currently running SUNOS 3.2. We have a Sun-3/280 which could be
used for the heavy computing part, relying on the 3/52M's for input and
output (is this possible?).
I am new to simulation, but I think I will require a discrete event
simulator. I wish to study the performance of several distribution
algorithms (protocols and topologies) for a distributed data structure.
This DDS will be used as a message store for communicating processes in a
distributed operating system we are developing here. Storage and
retrieval will be based on a key associated with the messages. For those
of you familiar with distributed systems, it is similar in concept to the
Linda language developed at Yale. We want to find out if (and how) the DDS
can be configured to achieve good performance (and reliability). Others
may want to use it for computer system design simulation.
There hasn't been much computer communications work done here, at least
not at a system level. So, we are in the market for a simulator which may
be used for computer communications (at a message level on a LAN) and
computer architecture research, as well as teaching.
I would appreciate receiving any recommendations (good or bad) you have
regarding simulators you have used that are available for Suns. Currently
we are considering the following. We aren't interested in GPSS.
- NETWORK II.5 and a teaching interface, by CACI in California
- PAWS and its graphical interface GPSM, by Information Research
Associates of Austin, Texas
- Simscript
Please reply by e-mail. I will post a summary.
Thank you very much.
Mark J. McIntosh <mmcintos%uvunix.uvic.cdn at ubc.CSNET>
University of Victoria, ECE Dept.
Box 1700, Victoria, BC, Canada
V8W 2Y2 (604) 721-7211
UUCP: ...!{ucbvax!uw-beaver,ubc-vision}!uvicctr!mmcintos
------------------------------
Date: 15 Feb 88 22:25:43 GMT
From: sdsu!turtle at sdcsvax.ucsd.edu
Subject: Looking for sun TeX implementation(s)
Does anyone know of any ports made of Donald Knuth's TEX program for any
kind of Sun computer? If so, where could we (SDSU) get hold of them?
Andrew W Scherpbier
San Diego State University
San Diego, California
ihnp4!jack!sdsu!turtle
[[ The University of Washington distributes a tape called "the Unix-TeX
tape". It includes support for many Unix implementations, including Suns.
My current recommendation along these lines is to use Tim Morgan's
"TeX-to-C" package to convert the Pascal form of TeX into C, then compile
and use the executable from the C code. This version of TeX is
substantially faster than the straight Pascal version. I belive that
"TeX-to-C" comes on the Unix-TeX tape. I think that the appropriate
person to contact these days about Unix-TeX is Pierre Mackay,
<mackay at ward.cs.washington.edu>. I hope Pierre forgives me if I am wrong!
--wnl ]]
------------------------------
Date: Thu, 11 Feb 88 21:57:36 +0100
From: mcvax!olsen!nagler at uunet.uu.net (Robert Nagler)
Subject: nfsd problems/performance?
The nfs daemons running on one of our 3/180 servers has been having very
intermittent problems. After some undetermined set of circumstances, all
of the nfsd processes hang in the "D" state forever. This problem has
only occurred about 6 times in the past 6 months (rough numbers here).
After the first few failures, we decided to try to fix the problem by
changing the conditions. The initial configuration of the machines was:
3 380MB Eagles
1 Xylogics controller (yes, we know this is bad.)
4 /etc/nfsd processes
First, we added another disk controller. This seemed to help
(psychologically at least), but we had another hang. So we tried to bump
up the number of servers from 4 to 8. Some time later the processes hung.
So we tried to decrease the number of servers to 3, but the problem
recurred.
The number of disks could be the problem, but then we have another server
with 3 disks (using Ciprico controllers) which has had no problems. We
have several other servers with 2 disks, but have never seen the problem
on these machines. Has anyone ever seen this behavior before or know how
we could go about repairing it?
After looking into this problem for a while, we've got some general
queries about the relation of the nfsd processes to request/response time.
There would seem to be two bottle necks in the NFS:
- Network request/response time
- Disk request/response time
If this is the case, then one should modify the number of nfsd processes
based upon the network loading factor and the number/speed of the disks
(plus caches...). In other words, one would assume that the number of
nfsd processes should be tailored to the particular environment and
machine. However, the nfsd man page speaks of magic and the number 4.
There doesn't seem to be any other reference that says anything different.
After doing some experiments, the performance seemed to vary when the
number of processes was changed, but there was nothing conclusive. We saw
a slight degradation (5%) going from 4 nfsd processes to 1, but, in one of
the tests, the greater number of processes (4) caused reduced performance
on the clients.
Has anyone done comprehensive testing of NFS performance? If so, why is
"4" magic or is it? Why is 4 the magic number for "biod" processes? Is
there any relationship between the number of biod and nfsd processes?
More generally, if we wanted to improve our performance (not that we know
it is bad), how we would we go about it? Any help would be appreciated.
Rob
------------------------------
Date: Fri, 12 Feb 88 19:20:54 PST
From: Richard H. Miller <RMILLER at sri-nic.arpa>
Subject: Availability of Used SUN-3's?
Please excuse a possibly laughable question:
Is anyone in this community aware of a source for USED SUN-3's? For
example, it's not unusual in other environments to find leasing companies
turning over their 2- and 3-year old systems.
We are a small consultancy, very involved in OSI and conformance testing,
with a requirement for at least two SUN-3's in a hurry. SUN-3/50's are
welcome, but I'd also like to find a 160 or 180.
On the chance that some of you have encountered sources, can anyone point
me to a reasonable organization that is currently offering SUN's on lease
or lease-to-purchase terms?
Any assistance is appreciated.
Rich Miller RMiller at SRI-NIC.ARPA
Telematica, Inc. Tel: (415) 329-9259
505 Hamilton Ave., Suite 200
Palo Alto, CA 94301
[[ If you only have information on specific machines for sale, please
contact Rich directly. If you have more general information (say, about
leasing or resale companies) then feel free to send the information in to
Sun-Spots. I just don't want this list to get in to the "swap shop"
business. --wnl ]]
------------------------------
Date: Thu, 11 Feb 88 15:19:00 EST
From: Marty Hall <hall at alpha.ece.jhu.edu>
Subject: Alternate Pascal Compilers?
I am posting this for a friend who is looking for alternate pascal
compilers for the Sun. He is used to Turbo Pascal, and is looking for
some of the same extensions such as direct bit manipulation, Hex
arguments, direct access to physical addressing of memory with variables,
and other such bells and whistles.
However, he is not interested in working in C, and was wondering if anyone
had comments/suggestions on other Pascal compilers available. Any ideas?
I will pass on any email that comes here, or you can contact him directly:
Mike Bagley
AAI Corp, MS 100/805
PO Box 126
Hunt Valley, MD 21030
(301) 628-3074
Thanks!
- Marty Hall
(hall at alpha.ece.jhu.edu, hall at bravo.cs.jhu)
------------------------------
Date: Mon, 15 Feb 88 08:25:00 PST
From: weiser.pa at xerox.com
Subject: Artecon - Sun Clones - Has anyone experience with them?
These are not Sun clones, exactly, they are repackaged Suns. That is,
they buy Sun CPU's (and perhaps other parts) and put them in their own
box. Sun lets them do this presumably because Artecon offers some value
added that Sun doesn't, such as different form factor, different disks,
etc. I have heard that Artecon office boxes are quieter, wood-grained,
and have up to a gigabyte of disk, but use only one pedestal, a
combination not available from Sun. I do not have direct experience with
Artecon, but third hand information says they are good.
-mark
------------------------------
Date: Mon, 15 Feb 88 15:42:33 CST
From: vuse!backup!drl at uunet.uu.net (David R. Linn)
Subject: DEC LAT protocol?
Does anyone have any leads on software that will allow a (nonDEC) UNIX
machine (specifically a SUN3/160) to talk both host and client side of the
DEC LAT protocol? Vanderbilt has a campus backbone shared by TCP/IP,
DECNET, and LAT packets and my users wish to be able to talk to machines
that speak only one side (or the other) of the LAT protocol.
David Linn
INET: drl at vuse.vanderbilt.edu CSNET: drl at vanderbilt.csnet
UUCP: ...!uunet!vuse!drl BITNET: linndr at vuengvax
AT&T: (615)322-7924
------------------------------
Date: 15 Feb 88 09:32:51 GMT
From: hans at sics.se (Hans Eriksson)
Subject: Scanned image into a SUN?
Any tips how to get a picture (paperbased) scanned into a SUN. We are
running SUNview. We also have macs and thunderscan so if someone has a
conversion program that also would be a great help.
Thanks
Hans
Hans Eriksson
Swedish Institute of Computer Science, Box 1263, S-163 13 SPANGA, Sweden
Tel: +46 8 752 15 27 Ttx: 812 61 54 SICS S Fax: +46 8 751 72 30
UUCP: hans at sics.UUCP or {seismo,mcvax}!enea!sics!hans EAN: hans at sics.sunet
------------------------------
End of SUN-Spots Digest
***********************
More information about the Comp.sys.sun
mailing list