Sun-Spots-Request@RICE.EDU.UUCP (05/19/87)
SUN-SPOTS DIGEST Tuesday, 19 May 1987 Volume 5 : Issue 15
Today's Topics:
No more 'read error from network'
Re: IP fragmentation, and how to avoid it?
Putting 3 or 4 drives on a Xylogics 450
Re: acucntrl for Suns
Re: Duplicate IP address
Sun 3.2 /lib/ccom bug?
Hayes 2400 uucp -> Sun3.2?
speaking of swap space?
CD-I authoring system?
window systems?
Suntools memory behavior?
Tape drives for Sun?
4.3 timed for Suns?
FORTRAN I/O in mixed language environment?
SunView Question?
uucp over SunLink X.25?
Looking for a Visicalc clone?
------------------------------------------------------------------------
Date: 14 May 87 18:50:26 GMT
From: bcsaic!paula@june.cs.washington.edu (Paul Allen)
Subject: No more 'read error from network'
A week ago, I posted a query here regarding a problem I was having with
remote logins on our 3/280, duckabush. Breifly, these were the
symptoms:
An rsh or rlogin attempt from any Sun to duckabush after duckabush had been
idle for some time resulted in the message 'read error from network.
connection reset by peer.' at the requesting workstation. Rsh attempts
from our Ultrix VAX, however, always succeeded. After getting in from the
VAX, a time window would open up during which Suns could also connect to
duckabush. Duckabush has also been displaying console messages of the
form: 'duplicate ip address from <ethernet address>'.
I got mail from several people, the most helpful of which was from
Celeste Stokely at Sun and Bud Spurgeon at Stanford. Celeste suggested
calling 800-USA-4SUN. Bud suggested attempting to identify the
masquerading system by taking duckabush off the net and telnetting to
its IP address. Following Celeste's advice led me to Tony Brooks at
Sun, who suggested I follow Bud's advice. Sure enough, telnet connected
me to a TI Explorer called Huey, who was using the same address as
duckabush. Correcting Huey fixed the problem. Although we do have a
large ethernet spanning four buildings, Huey was just down the hall on
the same floor in my building!
Thanks for your help!
Now, can anybody explain why the VAX was able to get to duckabush when
the Suns couldn't?
Paul Allen
paula@boeing.com
...uw-beaver!ssc-vax!bcsaic!paula
------------------------------
Date: Fri, 15 May 87 09:14:44
From: jr@alexander.bbn.com (John Robinson)
Subject: Re: IP fragmentation, and how to avoid it?
IP reassembly is the responsibility of the destination host, not the
gateways. Note that not all fragments of a datagram neceessarily
traverse the same sequence of gateways on their way to the destination
when there are multiple possible paths, so IP reassembly in gateways
is impossible lacking gateway-to-gateway connection-oriented protocol.
Each gateway is "supposed" to know the largest possible legal datagram
for each net it connects to. This can be limited by a particuilar
network - Arpanet for example limits you to about 1006 octets - but
hosts on a network may have different restrictions. There used to be
a recommended maximum datagram size of something like 576 octets that
should avoid fragmentation anywhere on the Internet. All modern TCP
implementations "should" cope with fragmentation and reassembly just
fine, however.
So the problem could be a gateway that isn't fragmenting as small as
it should, or a host (or gateway) that doesn't accept the largest
possible datagram for the net it attaches to. If you can find out how
to set the Sun's max datagram, and can repeat the failures, you could
probably find a workable length by binary search. Unfortunately, I
don't think IP is chatty enough to help you find the problem except in
this indirect way.
/jr
------------------------------
Date: Wed, 13 May 87 07:10:01 PDT
From: cgl.ucsf.edu!hoptoad.UUCP!cfcl!rdm@sun.UUCP (Rich Morin)
Subject: Putting 3 or 4 drives on a Xylogics 450
One of the announced features of the 3.2 release of SunOS is the ability
to handle more than two drives on a Xylogics 450 controller. Unfortunately,
the manuals fail to give any configuration information for this. In calling
Sun and sniffing around, I found nobody that had actually tried such a setup,
but I did get some useful hints and information. Since my installation was
successful, I will share them:
1) Sun people make a great deal of the fact that running more than two
drives on a controller is a performance lose. When pressed, they
admit that the extra drive(s) do not slow things down unless there
is contention for the controller. Since my system is not being used
very intensively (e.g., as a file server), I decided to ignore these
warnings.
2) The release, as shipped, has sufficient table sizes, etc., to handle
a 3 or 4 drive configuration. The only work that is required is a
line or two in the configuration file, as:
disk xy2 at xyc0 drive 2
3) Since I am using Fujitsu 2284's, my line actually looks like:
disk xy2 at xyc0 drive 2 flags 0x1 # no overlapped seeks
I did a build, made the slots in /dev, and put up the new disk. After
a couple of days of hard thrashing, the system has shown no perceptable
problems. And:
Filesystem kbytes used avail capacity Mounted on
/dev/xy0a 7471 5683 1040 85% /
/dev/xy0g 99823 81000 8840 90% /usr
/dev/xy1c 123023 90600 20120 82% /1c
/dev/xy2c 123023 9 110711 0% /2c
(As can be seen, I add new disks in a 1c, 2c fashion...)
------------------------------
Date: Fri, 15 May 87 20:44:27 MDT
From: hi!kurt@hc.dspo.gov (Kurt Zeilenga)
Subject: Re: acucntrl for Suns
Acucntrl (4.3 BSD) is not needed on SUNs. Sun was smart enough
to support dialin/out in the tty drivers. Read the adding hardware
section of your manual. Also, docuemented in the communication
sections for installing tip(1) and uucp(1) (not the man pages
but the tutorials).
However, if you really want acucntrl, it is available in the public
UUCP sources archive on many Internet and UUCP sites (such as seismo).
--
Kurt Zeilenga (zeilenga@hc.dspo.gov) I want my talk.flame!
"So long, Mom, I'm off to drop the bomb..."
------------------------------
Date: Mon, 18 May 87 15:15:43 PDT
From: seismo!lll-lcc.arpa!lams!mudshark!lefko@sun.UUCP (Martin Lefkowitz)
Subject: Re: Duplicate IP address
Not one hour before reading Margaret Reek's article about a possible reason
for the dupicate IP address problem, it happened to us here at lams.
Between her reason for the problem and in talking to the last person to log
on, we found the problem. What happened is that a pc-nfs user mistakenly
tried to log in as the server host. It crashed the system. This is no
feature.
Marty Lefkowitz
------------------------------
Date: Thu, 14 May 87 19:00:57 EDT
From: smb@mimsy.umd.edu (Steve Burinsky)
Subject: Sun 3.2 /lib/ccom bug?
Does anyone have any ideas as to the cause of the following? I am running
3.2.
% cat c.c
main ()
{
unsigned u;
double d;
d * u;
}
% cc -c -f68881 c.c
"c.c", line 6: compiler error: expression causes compiler loop: try simplifying
%
However, the expression:
u * d;
does not result in an error.
This only occurs with the -f68881 option. "d" can be double or float, and
"u" can be unsigned int or unsigned long (of course).
Steve
smb@mimsy.umd.edu
------------------------------
Date: Thu, 14 May 87 17:19:57 PDT
From: trwrb!moline@ucbvax.berkeley.edu (Dan Molinelli)
Subject: Hayes 2400 uucp -> Sun3.2?
I have tried to use a hayes 2400 with uucp under Sun V3.2
The installation notes for 3.2 claim that it supports the 2400
but there are no coniguration instructions as ther are for the
1200. I got UUCICO to dial out on it but it hangs on the login
prompt. My guess is that UUCICO is waiting for status code 5
(connect 1200) while what's being returned is status code 10
(connect 2400). My dilema is this: I don't have source, I don't
have sun "software support" and my usenet news neighbor insists on
2400 baud. Anyone even know what to call the hayes in L-devices ?
Should the line read ACUHAYES cua0 cua0 2400 or maybe something
like HAYES2400 cua0 cua0 2400? Sun refuses to tell me the answer
because I won't pay $6000 a year for phone support to answer a
question that should have been answered in the manual.
thanking in advance.
Bob Agnew
TRW MEAD
UUCP: {ucbvax,ihnp4,decvax}!trwrb!trwrc!agnew
------------------------------
Date: Thu 14 May 87 23:01:26-PDT
From: Bryan@sierra.stanford.edu (Doug Bryan)
Subject: speaking of swap space?
anyone understand this swap space stuff? how much swap space is
'sufficient' for a 3/110 with 8mb? how about a 3/50 with 8mb?
we give them 16mb of swap and they behave nicely. but since we have
12 diskless workstations on one file server, it would be nice to save
disk space where possible.
doug
-------
------------------------------
Date: 15 May 87 15:04:46 GMT
From: cadnetix!sharp@seismo.css.gov (Jim Sharp)
Subject: CD-I authoring system?
In the May issue of Computer Graphics World there is a news
blurb saying that Sun and Philips have agreed to work together to
produce a CD-I (Compact Disk - Interactive) authoring system.
Does anyone out there in newsland have any further information
about this?
Thanx in advance.
------------------------------
Date: Fri, 15 May 87 14:37:17 EDT
From: sundar@hermes.ai.mit.edu (Sundar Narasimhan)
Subject: window systems?
I had a couple quick question regarding Sun's window systems:
I know that sunview doesn't support multiple tty windows in a single
application -- is this fixed in news? (Is there a mode in which sun
terminal emulator windows can be operated in the same manner as xterm
slave windows)? (The program example that comes with the Documentation
seems to indicate slave mode operation may be possible, but the
documentation is sketchy enough so that I havent been able to figure
it out).
-Sundar
------------------------------
Date: Fri, 15 May 87 17:44:06 edt
From: mike@thumper.bellcore.com (Mike Caplinger)
Subject: Suntools memory behavior?
I'm working on a C garbage collector based on the "conservative" approach;
that is, it examines the entire data, bss, and stack areas of the program
looking for possible pointers, marks all of the objects so referenced,
then looks in each marked object for pointer references to others, and
so on. It replaces malloc, realloc, and free with new versions that use
a controllable format for allocated segments (with mark bits, and so on).
When I apply this to a Suntools program, it finds a large amount of storage
apparently unused and frees it. Things go subtlely wrong from there on out,
though -- pixwin operations like blt and text drawing start to leave bits
of crud on the screen, and the machine eventually panics and halts.
The question is this: Suntools is obviously doing funky things with memory
mapping and stuff, but isn't it the case that any malloc'ed object still
has a pointer to the base of that object somewhere in user data space?
If that assumption is wrong, then the whole GC scheme will have to be
tweaked (probably by locking down Suntools-allocated storage explicitly,
somehow or another.)
A few technical details -- my malloc puts everything on four-byte boundaries
like it's supposed to, and a pointer has to lie on a two-byte boundary
and have its low two bits clear to even be considered as possibly valid.
To answer my own question partially, Suntools calls valloc early on, and
valloc works by asking for enough space to meet alignment requirements
and then returns a pointer to the middle of the space actually requested,
so this chunk of malloced space looks unused to my sweeper. I'm not
yet sure what the right way to handle this is.
VAX valloc says under BUGS that "vfree" is not implemented. Sun valloc
(or maybe memalign) allows the resulting space to be freed with a
simple "free" by smashing a segment header into the middle of the space
allocated by valloc. This explains why my malloc didn't work with
Sun's valloc/memalign: the format of a segment header is different
between my code and theirs.
All this is completely inferred from behavior and the VAX source, but
I'm pretty sure it's correct.
I was surprised to see how many changes had been made to Sun's malloc
and related calls from the 4.3 versions; a look at the manual pages may
be in order if you haven't seen them lately.
I'm not ruling out the possibility of bugs in the code, just wondering
if anybody has any insight on this topic.
Mike Caplinger
mike@bellcore.com
{decvax,ihnp4}!thumper!mike
------------------------------
Date: Fri, 15 May 87 21:01:02 EDT
From: km@emory.arpa (Ken Mandelberg)
Subject: Tape drives for Sun?
How good is Suns 6250bpi tape drive? Does a Sun3 make it
stream on dumps?
Are there much better outside alternatives for the same cost?
------------------------------
Date: Mon, 18 May 87 12:50:21 EDT
From: mlandau@diamond.bbn.com (Matt Landau)
Subject: 4.3 timed for Suns?
I've seen some mention lately of porting the 4.3 timed facilities to
SunOS 3.2 to compensate for clock skew in NFS clusters. Has anyone
actually done this? If so, can you make the required changes available?
(Copies of our 4.3, SysV, and Sun source licences can be supplied if
necessary...)
------------------------------
Date: Mon, 18 May 87 12:23 EDT
From: "SDRRTR::PSI%HDSRTR::HDSRTR::MRGATE::\"MRGATE::PSI%SCRVX2::SIMON\"%sdr.slb.com"@relay.cs.net (Simon Barnes)
Subject: FORTRAN I/O in mixed language environment?
We would like to call FORTRAN routines (for which we do not have source)
from a main program written in another language, which could be C, but
happens to be Ada. The problem is that FORTRAN I/O referencing logical
units 5 and 6 is directed under these circumstances to the files
fort.5 and fort.6 instead of standard input and standard output.
How can we arrange for FORTRAN I/O to be directed to standard input/output,
where it belongs? Note that standard output does not necessarily mean
the terminal, /dev/tty.
Thanks to any contributors...
Simon Barnes.
------------------------------
Date: Thu, 14 May 87 09:19:40
From: nancy%WISDOM.BITNET@wiscvm.wisc.edu (Nancy Yavne)
Subject: SunView Question?
Is there any method to drag an icon over the canvas without having to redraw
the whole canvas (or even the portion covered by the icon). I know that this
can be done easily with cursors but my bitmaps are much bigger. Did anyone
implement such application and has suggestions and/or solutions for me?
Alex Polyak
BITNET: nancy@wisdom
ARPA: nancy%wisdom.bitnet@wiscvm.wisc.edu
UUCP: .. !decvax!talcott!WISDOM!nancy
P.S.
Please mail me personally, I promise to post summary of responces
(if there is enough interest).
Thanks.
------------------------------
Date: Fri, 15 May 87 15:31:37
From: mcvax!tut.fi!jh@seismo.css.gov (Juha Heinanen)
Subject: uucp over SunLink X.25?
We have tried to get uucp f-proto running over SunLink X.25 by making
uucico first to log in to the same Sun through a ptty and then to
invoke SunLink's pad command with the number of the remote uucp site.
This works so far that uucico is able to log in and start f-proto but
the actual file transfer fails probably because of a bad check sum.
Has anybody got any ideas why the transfer fails or, better yet, has
been able to solve the problem.
One solution we are considering is to make uucico open an X.25 socket
instead of playing the game with the pad command. Effectively this
would mean writing a new dialer for Sunlink X.25.
Juha Heinanen
Tampere Univ of Technology
Finland
jh@tut.fi (Internet), tut!jh (UUCP)
------------------------------
Date: Tue, 19 May 87 09:31:46 pdt
From: ssc-vax!ssc-bee!bossert@beaver.cs.washington.edu
Subject: Looking for a Visicalc clone?
I'm looking for a visicalc clone to run under 3.3. I have a copy of "vc" from
a User Contributed Software tape, but it generates compiler errors. Does anyone
have either a working version of vc or a good alternative?
John Bossert
UUCP: uw-beaver!ssc-vax!bossert
Internet: ssc-vax!bossert@beaver.cs.washington.edu
------------------------------
End of SUN-Spots Digest
***********************