Sun-Spots-Request@Rice.edu (William LeFebvre) (09/07/88)
SUN-SPOTS DIGEST Tuesday, 6 September 1988 Volume 6 : Issue 219
Today's Topics:
Re: NFS write performance
Re: Feature (or bug?) in SunOS 3.5 Pascal Compiler (pc)
multicast packets
Stupid YP behavior (services.byname)
SUN I/O *support#%@@*
Who Lives At 0x3ff00000?
Notification on receiving new mail
problems suninstalling SunOS 4 clients
Problems installing AT&T C++
Request for information about Modula-2 compilers
Need an accurate real time clock
Text previewers (Postscript)?
nd across multi disks?
Send contributions to: sun-spots@rice.edu
Send subscription add/delete requests to: sun-spots-request@rice.edu
Bitnet readers can subscribe directly with the CMS command:
TELL LISTSERV AT RICE SUBSCRIBE SUNSPOTS My Full Name
Recent backissues are available via anonymous FTP from "titan.rice.edu".
For volume X, issue Y, "get sun-spots/vXnY". They are also accessible
through the archive server: mail the request "send sun-spots vXnY" to
"archive-server@rice.edu" or mail the word "help" to the same address
for more information.
----------------------------------------------------------------------
Date: Tue, 6 Sep 88 01:04:56 EDT
From: dan@wind.bellcore.com (Daniel Strick)
Subject: Re: NFS write performance
In addition to the two recent comments in Sun-Spots digest on my report of
poor NFS write performance (from brent@ginger.berkeley.edu and
sxn@sun.com), I got one piece of mail from someone at sun that verifies
the problem. I don't think any of these comments really explain the
problem in terms clear to people (like myself) who have not studied the
NFS protocols. Here is my attempt. It may be incorrect since it is based
only on observed behavior of NFS systems and common sense.
A Unix process does not usually wait for completion of a write to a normal
file. The process resumes execution as soon as the kernel copies the data
into an internal buffer and the process never receives any specific
acknowledgment of the successful completion of the write. This tends to
work acceptably well only because disk writes almost never fail and when
they do fail the failure is usually noticed (and the system probably
crashes).
NFS does fail (network packets get lost or ignored and servers do crash).
The reliability of writes to remote files depends on something (apparently
the biod processes) waiting for positive acknowledgment of completed
writes and resubmitting the write requests if the server does not respond.
In order to simplify the implementation, NFS apparently does something
like synchronous I/O. I don't know if this means that the biod processes
may block waiting for acknowledgment or that the user processes may block
waiting for idle biod processes (this is an implementation detail and the
slowest implementation looks like the easiest). It does seem to mean that
the server must complete related disk output (file retrieval info, etc.)
before it can acknowledge the write request (otherwise the data could be
lost after the client got the acknowledgment). Therefore:
(1) The file server keeps rewriting file system blocks that would
have been cached dirty and written to disk many fewer times
for local file system writes (retrieval info and such), and
(2) locality of disk access is reduced resulting in more head motion
and lower local i/o rates.
The bottom line is that peak remote file write rates (what you get writing
test files into empty file systems) vary from about 60 kb/sec to about 120
kb/sec (depending on the server and the disk) while local file write rates
vary from about 700 kb/sec to 1800 kb/sec.
(I feel compelled to note that remote file reads are much faster than
writes. If you have a machine without excessive main memory, you probably
do a lot more reads than writes. The limitation here may be the ethernet
and this may become painfully obvious when SMI extends the sun-4 product
line.)
------------------------------
Date: Tue, 6 Sep 88 08:37:40 EDT
From: Chuck Musciano <chuck@trantor.harris-atd.com>
Subject: Re: Feature (or bug?) in SunOS 3.5 Pascal Compiler (pc)
This is a feature. Pascal should always check all subranges, array
references, pointer usage, etc., either at compile time (if possible) or
at run time. Failure to catch such errors, especially with -C specified,
is a bug. There should be no argument about this behavior on any street
corner, especially when you have asked for explicit checking via -C.
Chuck Musciano
Advanced Technology Department
Harris Corporation
(407) 727-6131
ARPA: chuck@trantor.harris-atd.com
------------------------------
Date: Tue, 06 Sep 88 09:05:13 EST
From: riedl@purdue.edu
Subject: multicast packets
In addition to changing the device driver to handle multicast packets, you
must also configure the ethernet interface to receive hardware multicast
packets. On the LANCE interface this consists of setting bits in a
multicast filter corresponding to the packets you want to receive. I
don't know what needs to be done for the Intel interface, but in the
header file (sys/sunif/if_iereg.h) there is a command type:
#define IE_MADDR 3 /* multicast address setup */
We have a modified version of the LANCE driver (sys/sunif/if_le.c) that
includes an ioctl to set the multicast packet filter. We're willing to
send diffs if you're also interested in multicasting on 3/50s.
John Riedl
{ucbvax,decvax,hplabs}!purdue!riedl -or- riedl@mordred.Purdue.EDU
------------------------------
Date: Tue, 6 Sep 88 12:18:08 EDT
From: Alexander Dupuy <dupuy@columbia.edu>
Subject: Stupid YP behavior (services.byname)
Sun had a chance to rationalize some of this when they came out with 4.0,
and they almost did, but they blew it.
Until 4.0, getservbyname(3) and getservbyport(3) both did complete scans
of the YP services.byname database, which was keyed by portnumber, not by
name. This was noted by people porting NFS to other systems, and in the
sun-source archives, a ypmake.diff posting by Dan Trinkle dated October
1987 contains fixes so that correct services.byname and services.bynumber
would be generated.
Now, in 4.0, Sun has finally stopped doing complete searches over the
services.byname map for getservbyport() calls. Unfortunately, the map is
still called services.byname (which it isn't), and the getservbyname()
function still does sequential searches of the database. This does help
programs like etherfind and nnstat, which do many, many getservbyport()
calls, but startup time for programs like telnet, which make calls to
getservbyname() is still quite slow (we have a fairly large services
file). It seems extremely unlikely that Sun will ever fix this
nomenclature problem, since it would cause incompatibilities between 4.0
YP clients and 4.1 YP servers.
So now that Sun is _depending_ on the dubious assumption that the
services.byname database is actually keyed by port number, Dan Trinkle's
fixes, which we had installed, actually breaks getservbyport(). What we
did, at Columbia, was to edit (with emacs _only_) /usr/lib/libc.{a,so.1.1}
replacing the character string "services.byname" with the more logical
"services.byport" (of exactly the same length). We then changed the YP
makefile to generate a services.byport map as well as a true
services.byname map. This has the advantage that a ypmatch does what one
might expect.
@alex
------------------------------
Date: 6 Sep 88 06:40:31 GMT
From: vsi1!tim@ames.arc.nasa.gov (Tim Richardson )
Subject: SUN I/O *support#%@@*
I've been reading (again!) about Sun's lack of "openness", or support for
3rd party peripherial vendors--especially disk controller mfg's.
For 21 months now, I have been fighting this problem in conjunction with
the people at Ciprico and Interphase. Meetings, phonecons, letters---
around and around and around!!! We just about get someone in decision
making authority to agree and before the decision can be rendered final,
BANG, we are back to square one. Three times people have been as far as
Sun's CEO but still no helpful decision for people like us who attempt to
supply custom "tuned" systems for our customers.
We've heard all their so-called arguments, especially the one about not
wanting field support problems. This doesn't hold any water with us,
since we are a board level OEM of Sun and they don't have ANY service
responsibility to/for our cusomters at all.
What further burns me up is that the solution is so simple---basically all
they have to do is to allow their boot-prom object-code to be linked with
someone else's code. However, as a recent net respondent wrote..."Sun is
busy hiring ex-DEC people." Simple but elegant solutions to problems has
never been one of DEC's fortes and Sun has seemingly decided to emulate
engineering and sales philosphy of a company which is still attempting to
apply 1960's solutions to late 1980's type problems and technology!
Today's customers demand hardware and systems which can be continuously
improved (or tuned if you wish), by the customer. Often customers will
opt for OEM supplied upgrades, but equally often they will want their own
solutions. An enormous amount of "tinkering" goes on out in the field,
which often develops into products later on down the road. No company, no
matter how big or successful they think they are can afford to long ignore
a reasonably sized number of customers. Each one of us which is today
somewhat offended or "put-off" by an attitude or problem, tomorrow is a
potential new customer for another company's product.
Tim Richardson VP Engineering
VICOM Systems Inc. sun!pyramid----\
2520 Junction Ave. uunet!ubvax-----!vsi1!tim
San Jose, CA 95134 ucbvax!tolerant/
+1-408-432-8660
------------------------------
Date: Tue, 6 Sep 88 13:23:02 CDT
From: Scott Guthery <spar!ascway!guthery@decwrl.dec.com>
Subject: Who Lives At 0x3ff00000?
Here's a little C program that hard crashes my 4/110 with a watchdog reset:
main()
{
float *f = 0x3ff00000;
*f = 1.0;
}
Anybody have any ideas what's living at 0x3ff00000 and why he gets so mad
when you wake him up?
Cheers, Scott
------------------------------
Date: Tue, 6 Sep 88 17:41:50 IST
From: Shlomo Goldberg <shlomo@TECHSEL.BITNET>
Subject: Notification on receiving new mail
Here I am running several Sun 3/50's being served by a Sun 4 all using Sun
OS 4.0. In addition, I have a VAX 780 running Mt Xinu 4.3 with NFS.
Using NFS I made one mail directory "/usr/spool/mail" for all the machines
located on the VAX.
This has the nice feature that a user may use the "mail" command on any
machine to read and handle his mail. He may also send mail to any user
without the need for an updated alias file, as all the mail in the system
is held in the same place. Unfortunately, the "comsat" program is started
by "inetd" deamon only on the machine to which the mail is directed.
Thus, if a user is on another machine although he can handle his mail he
is not informed of its arrival.
Has anybody solved this problem? Thanks.
[[ The c-shell has builtin mail notification, but it only checks
immediately before it prompts you (in other words, it is synchronous).
What about Mailtool? It checks for new mail by periodically looking at
/usr/spool/mail/<user> and doesn't need comsat. For generic terminal
users, there's "sysline" from BSD Unix. If you don't want to use Mailtool
on your Sun, it would be nice to have some sort of mail watcher tool that
would notify you when /usr/spool/mail/<user> changed. To my knowledge,
no such program exists. Although it wouldn't be hard.... --wnl ]]
Shlomo Goldberg
Manager, Systems Group - Israel Institute of Technology, Computer Science Dept.
CSNET - shlomo@sel.technion.ac.il
BITNET - SHLOMO@TECHSEL
------------------------------
Date: 6 Sep 88 00:47:26 GMT
From: eric@eddie.mit.edu (Eric Van Tassell)
Subject: problems suninstalling SunOS 4 clients
After booting from le(). I run suninstall to configure the tapeless local
machine as a standalone machine. Both the server and client are sun
3-50's. Running suninstall under adb shows that it is trying to
fopen("/usr/etc/install/files/sys_info","r")
This file doesn't exist and the file system is read-only. There is also a
message that /usr/etc/install/files/suninstall.log can't be opened. None
of this is a surprise. My question is how do these files ever get there
and how did I mess up installing the server? Does anyone understand this
and why "assigning host info fails"? Thanks in advance. Please e-mail me
directly since I sometimes can't get to a machine with news.
dlcdev!eric@eddie.mit.edu
mit-eddie!dlcdev!eric
------------------------------
Date: Tue, 6 Sep 88 15:35:13 CDT
From: "Eric C. Pearce" <pearce@tycho.yerkes.uchicago.edu>
Subject: Problems installing AT&T C++
Has anyone out there successfully installed AT&T C++ (release 1.2) on a
SUN? My problem is that CC dies compiling lib/static/munch.c while
building the C++ translator from the C++ source:
cp libC.a /usr/local/lib/libC.a
CC -O -DBSD -o munch lib/static/munch.c
CC lib/static/munch.c:
cc -o munch -O munch.c..c -lC
./CC: 24293 Memory fault - core dumped
rm munch.o
rm: munch.o: No such file or directory
*** Error code 1
I have run bsd.fix before starting to bootstrap the translator. The
library also seems to be ok (after fixing a few makefile filenames).
AT&T has been no help. Does anyone have any ideas? Please respond by
e-mail to pearce@tycho.yerkes.uchicago.edu if you can help. Thanks.
- Ecp.
------------------------------
Date: Tue, 06 Sep 88 11:49:11 +1000
From: Alistair Moffat <munnari!munmurra.mu.oz.au!alistair@uunet.uu.net>
Subject: Request for information about Modula-2 compilers
We are looking for a Modula-2 compiler or interpreter to run on networked
3/50's to teach introductory programming to first year students. I am
aware of two compilers, one from ACE Corporation, and one as part of the
Amsterdam Compiler Kit, both from the Netherlands. Does anyone have
experience with either of these two compilers in a student environment?
Does anyone have information about other compilers that may be suitable?
Thanks for the help. Alistair Moffat, alistair@munnari.oz.au The
University of Melbourne Australia (+61 3 344 5229).
------------------------------
Date: 6 Sep 88 14:03:01 GMT
From: Jonathan Knight <mcvax!cs.kl.ac.uk!jonathan@uunet.uu.net>
Subject: Need an accurate real time clock
I'm posting this for a friend, but post email to me.
I have a friend who needs a very accurate (ie better that that from
standard sys calls) real time clock. Is their any place on a sun 3/50
where he can obtain an accurate form of timekeeping? Failing that are
there any cards that would do the job?
Details of cost and accuracy would be appreciated.
Jonathan Knight, JANET: jonathan@uk.ac.keele.cs
Department of Computer Science UUCP: ...!ukc!kl-cs!jonathan
University of Keele, Keele,
Staffordshire. ST5 5BG. U.K.
------------------------------
Date: Tue, 6 Sep 88 13:05:31 EDT
From: beth@tcgould.tn.cornell.edu (Beth Elias)
Subject: Text previewers (Postscript)?
Hello and Help?
We have just installed SunOS 4.0 on our fileserver and now have people
actually using the workstations that are plugged into it.
One question that comes up more that others is: Is there a previewer
(postscript most wanted) that would allow the fully formatted text to be
displayed to the workstation monitor?
Since we are new to the SunOS maybe someone can tell me if there is such a
program available and if so how to get a release of it. Other than asking
Santa...
P.S. the kicker is that users are also requesting that is display correct
font and pitch.
Please reply directly to me and I will summarize for posting to the net...
Thanks Very Much!
[[ Previewer for what text formatter? Troff? TeX? --wnl ]]
Beth Elias
Electrical Engineering
Cornell University
Ithaca, N.Y.
elias@tesla.ee.cornell.edu
------------------------------
Date: Tue, 6 Sep 88 17:26:05 EDT
From: telenet!venus!emast@uunet.uu.net (Eldon Mast)
Subject: nd across multi disks?
I recall a contribution posted to ss's giving a detailed explaination on
configuring a 3.x server with nd's across multiple 450/451 controllers. I
have searched back issues to March with no luck. Can anyone tell me x and
y?
Thanks,
Eldon R. Mast
uunet!telenet!emast
------------------------------
End of SUN-Spots Digest
***********************