SUN-Spots Digest, v5n3
Sun-Spots-Request at RICE.EDU.UUCP
Sun-Spots-Request at RICE.EDU.UUCP
Wed Jan 28 04:19:16 AEST 1987
SUN-SPOTS DIGEST Monday, 26 Jan 1987 Volume 5 : Issue 3
Today's Topics:
Network Simulation Tool available for Suns
Ether bridges may require rsize, wsize
Re: singleuser boot security (v5n1)
Disk speeds of Sun file servers (long)
Sun sl
Screen Saver for SUNs, answer! (2)
domain support in Sun 3.2
Re: Sun 3.0 /bin/login doesn't set tty ownership
3/260's are NOT vaporware (2)
rwhod fixes
RCS
Ethernet controller information wanted?
vectors unaffected by clipping?
hardware flow control?
Re: mail on diskless workstations?
1/2" tapes on Sun-3's?
C compiler bug?
nd and arp problems?
Sun terminal emulation?
------------------------------------------------------------------------
Date: Thu, 8 Jan 87 14:11:25 EST
From: <dupuy%amsterdam at columbia.edu> Alexander Dupuy
Subject: Network Simulation Tool available for Suns
Columbia University Computer Science Department
Distributed Computing & Communications (DCC) group
Network Simulation Testbed (Nest) version 2.3 Now Available
Nest is part of a collection of tools being developed at Columbia for the
design and analysis of large scale distributed systems. Nest allows rapid
prototyping of these systems by providing a simulated network environment in
which independent functions can easily be composed into distributed systems.
The simulated network can be modified while the system is running, allowing
performance tracing and measurement in a wide variety of simulated conditions,
including failures.
Nest implements this with a library that maintains a model of the simulated
network and multiplexes the independent functions within a single process. The
functions are associated with nodes in the network, and communicate with other
nodes through library facilities which can model many communication methods or
behaviors. This library also supports a user interface client which allows
runtime control and measurement of the simulated network.
The current release of Nest version 2.3 runs on Sun and Vax computers under
any 4.2bsd Unix derivative (including 4.3). The user interface client runs on
Suns with 2.0 or later release of SunView/SunWindows. Nest version 2.4 should
be ready by spring 1986, and will include an enhanced user interface client
with both SunView and X versions.
Nest is available to non-profit and educational institutions at no cost,
and at a nominal $100.00 support fee for others. Please contact Alexander
Dupuy or Jed Schwartz for details.
Alexander Dupuy Jed Schwartz
485 Computer Science Building 517 Computer Science Building
Columbia University Columbia University
New York City, N.Y. 10027 New York City, N.Y. 10027
(212) 280-8622 (212) 280-8192
arpa: dupuy at columbia.edu arpa: jed at cs.columbia.edu
uucp: seismo!columbia!dupuy uucp: seismo!columbia!cs!jed
------------------------------
Date: Thu, 8 Jan 87 20:01:32 PST
From: <speck at vlsi.caltech.edu> Don Speck
Subject: Ether bridges may require rsize, wsize
One fellow here was trying to run NFS across a network containing
Ungermann-Bass bridges on a 5 Mb/s backbone, and was getting wierd
errors. The errors went away when he changed /etc/fstab to specify
rsize=4096,wsize=4096. Apparently, the Ungermann-Bass bridge can't
take very many packets back-to-back, much like 3com Ethernet boards.
Don Speck speck at vlsi.caltech.edu {seismo,rutgers,ames}!cit-vax!speck
------------------------------
Date: Sun, 11 Jan 87 23:41:57
From: <lrj at lasspvax> Lewis R. Jansen
Subject: Re: singleuser boot security (v5n1)
Why not just place "login root" in /.profile? I imagine
it does present a window, if small, for a ^C, i really
haven't explored it much. The other problem is if you
like using the Bourne shell when root; we use csh.
This suggestion was taken from the BUGS section of the
SunOS 3.2 man page on init(8). It doesn't seem to appear in
the 3.0 version of that man page.
-- Lewis R. Jansen, LASSP Systems Grunt
lrj at lasspvax.tn.cornell.edu
------------------------------
Date: Mon, 12 Jan 87 00:28:43 PST
From: <speck at vlsi.caltech.edu> Don Speck
Subject: Disk speeds of Sun file servers
In hopes of gauging the differences in disk speed across the range
of Sun file servers, I ran some experiments measuring the minimum
rotational latency between successive reads on the raw device as a
function of blocksize. This is related to the "rotdelay" filesystem
tuning parameter, but is a little larger because the raw device has
higher overhead than the block I/O internals.
My method was to do reads some number of sectors apart, and push the
number down until I stopped gaining any speed from this (i.e. until
I started losing revolutions). The "gap" so obtained depended on the
size of the reads, so to separate the per-kilobyte setup time from
the per-kilobyte recovery time, I alternated large reads and small
reads. From this, I found that the setup time did not depend strongly
on the size of the read, but the recovery time did, and when the reads
were made large enough (i.e. recovery time got large enough) the disk
lost a revolution anyway.
I conclude that the DMA wasn't keeping up with the disk, and the large
recovery time was buffer draining. For moderately large transfers, the
buffer became full, and the disk had to "blow a rev" while the buffer
emptied.
The machines tested pretty much spanned the range of Sun file servers:
Name CPU OS controller disk sectors rps
rosie 3/260 3.2 Xylogics 451 M2333 67 60
stokes 3/160 3.1 Xylogics 451 M2333 67 60
jove 3/160 3.0 Xylogics 450 Eagle 46 66
sol 2/170 3.0 Xylogics 450 Eagle 46 66
Two rotational gaps were measured at each block size: the first is
the gap preceding the large read (setup time), and the second is the
gap after the large read (recovery time). Between each large read
was a small (1 KB) read to serve as a reference and timing point.
The measurements are in units of disk sectors:
1K 4K 8K 16K 20K 24K 25K 32K 43K
sol 25 25 26 29 28 33 32 41 36 50 37 53
jove 14 14 15 17 16 19 17 26 17 32 17 33
stokes 17 17 17 20 17 25 17 34 18 39
rosie 14 14 14 18 14 20 16 26 18 38 19 46
Attempts to do transfers larger than those in the table would lose
revolutions while the buffer emptied. Knowing the buffer size (8 KB) and
the average transfer rates of the disks, we can calculate the DMA rate:
sol 1.0 MB/s
jove 1.0 MB/s
stokes 1.2 MB/s
rosie 1.6 MB/s
(Life must be really miserable with ancient 450's that have only a
2K buffer; those can be expected to blow revs even on 8K transfers).
In no case was the DMA fast enough to keep up with the disk; you're
fooling yourself if you think you're gaining anything from transfer
rates higher than an Eagle's.
Reducing the "raw numbers" to milliseconds per transfer:
interblock time xfer time total for 8K
sol 7.9ms + 0.52ms/KB 0.66ms/KB 17.3 ms (2/170, XY450)
jove 4.5ms + 0.33ms/KB 0.66ms/KB 12.4 ms (3/160, XY450)
stokes 4.2ms + 0.30ms/KB 0.50ms/KB 10.6 ms (3/160, XY451)
rosie 3.4ms + 0.22ms/KB 0.50ms/KB 9.2 ms (3/260, XY451)
Surprisingly, the difference between a 3/160 and a 3/260 is not
large; it is exceeded by the difference between a 450 and a 451.
The 3/260 doesn't look like a cost-effective file server, given
its high price. Certainly one wouldn't want to hobble such an
expensive file server with the slower Xylogics 450.
Remember that this is *minimum* transfer time, such as encountered
when reading huge files on a properly tuned filesystem. ("Properly
tuned" means that the rotdelay parameter is set correctly, as in the
above table, instead of leaving it at the default setting, which is
far from optimal). Random traffic also encounters the average seek
time and average rotational latency (= 1/2 revolution time); both
figures are lowest in Eagles.
Don Speck speck at vlsi.caltech.edu {seismo,rutgers,ames}!cit-vax!speck
------------------------------
Date: Tue, 13 Jan 87 13:51:15 EST
From: <mark at mimsy.umd.edu> Mark Weiser
Subject: Sun sl
I recently had the 'opportunity' of putting SLIP into a sun kernel without
source, and discovered that my previously posted version depended on some
Maryland kernel hacks.
A new guaranteed to work for object only suns (at least for 3.2)
is available for anonymous ftp from mimsy.umd.edu, file sun_sl.shar.
I watched what I did this time, so there is also a README with step-by-step
installation instructions.
-mark
-----
Spoken: Mark Weiser ARPA: mark at mimsy.cs.umd Phone: +1-301-454-7817
CSNet: mark at mimsy UUCP: {seismo,allegra}!mimsy!mark
USPS: Computer Science Dept., University of Maryland, College Park, MD 20742
------------------------------
Date: Wed, 14 Jan 87 10:35:58 EST
From: <mckay at ee.ecn.purdue.edu> Dwight D McKay
Subject: Screen Saver for SUNs, answer! (1)
A while back I asked about a Screen Saver for SUN workstations. I just got
a message back from someone at SUN that this feature is part of their 3.2
UNIX release. Specifically:
screenblank (1) - turn off video when the mouse and keyboard
are idle
--Dwight Mckay, ECN Text & Workstation Software Support
[arpanet: mckay at ee.ecn.purdue.edu, usenet: ...ihnp4!pur-ee!mckay]
[Compu-serve: 75776,1521, office: EE 348B, phone: (317) 494-3561]
------------------------------
Date: 14 January 1987 1150-PST (Wednesday)
From: <wallen at nprdc.arpa> Mark Wallen
Subject: Re: Screen Saver for SUNs, answer! (2)
Here is a real simple minded screen saver that you can run from crontab--
it just uses lockscreen. The two weak points are:
1) the test for "idleness" is extremely simpleminded--
the console is owned by root.
2) You exit lockscreen with the "exit desktop" selection
but you get no login prompt; you have to type RETURN.
--------save_idle_screen -- run from crontab
#! /bin/csh -f
set DotSuntools = /local/bin/save_idle_screen.data
set llout = ( `/bin/ls -l /dev/console` )
# this is a really poor test of the idleness of the system
if ( X$llout[3] !~ Xroot ) exit
exec suntools -d /dev/fb -m /dev/mouse -k /dev/kbd -s $DotSuntools &
--------save_idle_screen.data -- use the icon of your choice
#
# dummy .suntools for save_idle_screens
lockscreen -e -WI /local/lib/icons/idle_sun1.icon
------------
Mark Wallen
Institute for Cognitive Science
UC San Diego
wallen at ucsd.edu
wallen at nprdc.arpa
------------------------------
Date: Wed, 14 Jan 87 12:48:31 est
From: <hedrick at topaz.rutgers.edu> Charles Hedrick
Subject: domain support in Sun 3.2
You might be interested to know that in 3.2 Sun more or less supports
the use of bind. They supply named (called /usr/etc/in.named). They
also supply an interface to the resolver. ypserv will call the
resolver if it looks up a name in the yp hosts database and doesn't
find it. To get this, you must run ypserv with the undocumented
option -i. In addition, you must create a file /etc/resolv.conf.
You should put in it the commands
domain foo.edu
nameserver 127.0.0.1
where foo.edu gets appends to any name that doesn't have a dot in
it. (Obviously you replace foo.edu with your own domain name.)
This file should not be needed, except that Sun's defaults seem odd.
By default it uses the domain taken from your yp domain. Maybe
that's OK for some sites. But we use rutgers.edu for all of our
host names, but each set of machines with a different passwd file
have their own yp domain. So for us it is wildly inappropriate
to use the yp domain as the default for hostname lookup. The
default nameserver is something completely off the wall, like
239.0.0.1. The example above assumes that you want to run named
on every machine that is a yp server. If not, then of course
the nameserver command can point to a machine or machines that you
want domain queries to go to. If you want to use only the domain
system, you will want to create a very short /etc/hosts, and then
to a make in the yp directory. We use /etc/hosts that lists only
127.0.0.1, as localhost and loghost. Apparently named needs that
entry in order to start.
For sites that have a number of multi-homed hosts, Sun provides
an extension to the resolver to let you select which address
you prefer. /etc/resolv.conf can have a line like
address 128.6.0.0 128.121.0.0
The resolver will sort the list of addresses by looking for any
address on network 128.6 and moving it to the beginning of
the list, then any address on 128.121, etc. However this is
controlled by a conditional in gethostnamadr.c, in
..src../usr.etc/in.named/res. So to get this, you'll have to
turn on that conditional in the Makefile, rebuild libresolv.a
in that directory, and then rebuild ypserv in ..src../usr.etc/ypserv.
We have an additional problem, because we have machines that serve
clients on two different networks. So we have an additional
sorting that depends upon which network the request arrives from.
SInce Sun specifically prohibits such configurations, we assume
that this change will not be relevant to many other people.
------------------------------
Date: Wed, 14 Jan 87 12:58:11 CST
From: <phil at titan.rice.edu> William LeFebvre
Subject: Re: Sun 3.0 /bin/login doesn't set tty ownership?
That's funny. As far as I know, I am on a Sun 3 running 3.0, and my
tty is owned by both my user id and group id. Perhaps your system was
not installed correctly? Perhaps /bin/login is not setuid to root like
it is suposed to be?
William LeFebvre
Department of Computer Science
Rice University
<phil at Rice.edu>
------------------------------
Date: Wed, 14 Jan 87 19:42:37 EST
From: <mark at markssun.cs.umd.edu> Mark Weiser
Subject: 3/260's are NOT vaporware (1)
Well, I heard from at least 3 people within 24 hours that had received 3/260's
as regular customers. All had received them in the last 2-3 weeks.
I also heard from other people who had been told by Apollo reps that
no one was shipping 68020's. This seems to be a nasty rumor (or
perhaps it was true when it was started k months ago.)
So, when your friends from Apollo start trashing 68020's, point them at Sun.
-mark
------------------------------
Date: Thu, 15 Jan 87 11:17:56 EST
From: <mo at seismo.CSS.GOV> Mike O'Dell
Subject: 3/260's are NOT vaporware (2)
If 260's are vaporware, I am not exactly sure just what the 3 machines next to
me really are, then. My site has three 3/260HM's delivered and running like
nothing I've ever seen (fast, Fast, FAST!). Further, my 3/280 server with
16 Megs, a 9-track, and 2 Eagle XP's just came in the door yesterday.
Oh yes, we have had the 260's since mid-November.
So if anyone says 260's are vaporware, tell 'em they're only blowing smoke.
Smoke from the flaming tires on a 260.
-Mike O'Dell
PS - Jumpdemo on a 260, in a dark room, with your face
close enough that the image fills your field of vision
is a Real Trip.
------------------------------
Date: Thu, 15 Jan 87 12:05:08 EST
From: <mark at markssun.cs.umd.edu> Mark Weiser
Subject: rwhod fixes
It has been pointed out to me that changes to rwhod were
posted to Jim Knutson about a year ago (to this list, V 4, I 4, dated
Saturday, 8 Feb 1986). His approach is the same I recently mentioned and
offered diffs for.
I have no reason to think Jim's isn't a good approach--I simply forgot it
was there.
-mark
------------------------------
Date: Thu, 15 Jan 87 10:45:48 cst
From: <ables%pp at mcc.com> King Ables
Subject: RCS
We just recently (finally) got RCS fixed on Sun-3s. The solution is to use
diff from 4.3BSD as rdiff in RCS. diff didn't have the right parameter so
they wrote rdiff. However, they hacked diff to get rdiff. There was a bug
deep in diff that made it not work when dereferencing null pointers (which on
a Vax, you can do). They fixed this basic bug in diff in 4.3BSD in addition
to adding the needed parameter. RCS under 4.3BSD doesn't even need rdiff since
diff works like it needs to!
Also, just FYI, Jim Knutson of U. of Tex. (knutson at ngp.utexas.edu) posted
almost the same fix to rwho that was posted recently at least a year ago
(I *think* to this very newsgroup/mailing list!). I've been running it that
long and it does a great job of cutting down net traffic!
-king
ARPA: ables at mcc.com
UUCP: {gatech,ihnp4,nbires,seismo,ucb-vax}!ut-sally!im4u!milano!mcc-pp!ables
------------------------------
Date: 7 Jan 87 4:50 +0800
From: <see%ubc.csnet at RELAY.CS.NET> Helen Lim See
Subject: Ethernet controller information wanted?
I'm currently writing Ethernet device drivers for a distributed operating
system (Team Shoshin) for Sun-2/50s which use the Intel 82586 LAN controller
chip and Sun-3/50s which use an AMD LANCE (Local Area Network Controller for
Ethernet) Am7990 chip. The specifications and users manuals for these chips,
and the 3Com chip as well, would be extremely helpful to me.
I would appreciate it if someone could send me any of this material or tell
me where I can obtain it. Even better, has someone written a driver for
either the Intel or AMD device that they can send to me?
Thanks very much for your assistance.
Helen L. See | {ihnp4!alberta,uw-beaver}!ubc-vision!ubc-cs!see
Dept. of Computer Science | see at cs.ubc.cdn
University of British Columbia| see%ubc.csnet at csnet-relay.arpa
Vancouver, B.C. | see at ubc.csnet
Canada V6T 1W5
------------------------------
Date: Fri, 2 Jan 87 11:01:21 EST
From: <mark at markssun.cs.umd.edu> Mark Weiser
Subject: vectors unaffected by clipping?
On page 18 of the Pixrect Reference Manual (Version A of 17 February
1986), near the bottom, it says: "The vectors are balanced according to
their endpoints as given and not as clipped, so that the same pixels
will be drawn regardless of how the vector is clipped."
The following little program illustrates that, sadly, this statement is false.
Compile the guy below with
cc -o vectortest vectortest.c -lsuntool -lsunwindow -lpixrect
Run it, then draw a vector with the left mouse button, and try setting
different clipping regions with the right button. The vector is
redrawn each time you change the clipping region, and the pixels are
NOT necessarily the same each time. Now, I'm not a graphics person--is there
something I misunderstand? Perhaps pw_vector doesn't use pr_vector??
----cut here for program------
#include <suntool/sunview.h>
#include <suntool/canvas.h>
#include <suntool/panel.h>
Pixwin *mainpw;
void main_event_proc(), draw_proc(), clear_proc();
int cx = 100, cy = 100, vx, vy;
main(argc, argv)
char **argv;
{
Frame mainframe;
Canvas maincanvas;
Panel mainpanel;
mainframe = window_create(NULL, FRAME,
FRAME_LABEL, " left to set vec, right to set clip",
0);
maincanvas = window_create(mainframe, CANVAS,
WIN_EVENT_PROC, main_event_proc,
0);
mainpw = canvas_pixwin(maincanvas);
window_main_loop(mainframe);
exit(0);
}
void
main_event_proc(window, event, arg)
Window window;
Event *event;
caddr_t arg;
{
int id = event_id(event);
if (event_is_button(event) && event_is_down(event)) {
switch(id) {
case MS_LEFT:
vx = event_x(event);
vy = event_y(event);
break;
case MS_RIGHT:
cx = event_x(event);
cy = event_y(event);
break;
}
draw();
}
}
draw()
{
Pixwin *pw;
pw = pw_region(mainpw, 0, 0, cx, cy);
pw_vector(pw, 0, 0, vx, vy, PIX_SRC, 1);
pw_close(pw);
}
------------------end of program--------------------------------------
------------------------------
Date: Fri, 02 Jan 87 20:04:31 -0500
From: <mark at markssun.cs.umd.edu> Mark Weiser
Subject: hardware flow control?
What is anyone doing about hardware flow control (using DTR, etc.
instead of XON/XOFF)? Anyone implement this in a version I can borrow?
(We have Sun Source). Does this come with the appropriate Sun networking
product? There have been two previous requests on this list for
this information, but I saw no replies. Thanks in advance...
-mark
------------------------------
Date: Fri, 9 Jan 87 10:00:00 PST
From: <ted at braggvax.arpa> Ted Nolan
Subject: Re: mail on diskless workstations?
>From: Doug Bryan <Bryan at SU-SIERRA.ARPA>
>Subject: mail on diskless workstations?
>My question is: Is there an easy way to set up mail so that the users
>can log into any of the 110s and read their mail sent to them at the
>file server?
>doug
We have been successfully running the POP protocol provided with UCI's MH 6.5
for several months now. It provides what you want, the only disadvantage
being that you can no longer use the regular Mail command to read your mail,
you must use the MH commands (we already did, so had no problem with that).
You have to add one demon to the fileserver, but you can turn off sendmail
on all the clients. MH is available free by anonymous ftp.
Ted Nolan
ted at braggvax.arpa (a POP host)
------------------------------
Date: Fri, 9 Jan 87 11:56:22 EST
From: <smb at mimsy.umd.edu> Steve M. Burinsky
Subject: 1/2" tapes on Sun-3's?
I've seen people discussing several drives running on several controllers.
I am interested in Xylogics controllers because Sun will support them.
The 451 and 472 have both been suggested as tape controllers. Can someone
summarize which drives go with which controllers? Also, I have a DEC TU80
that I'd like to put on a Sun, but can't figure out who makes it. Bottom
line is, which controller do I need for a TU80?
Steve Burinsky
smb at mimsy.umd.edu
------------------------------
Date: Sat, 10 Jan 87 22:13:57 -0500
From: <ken at rochester.arpa> SKY
Subject: C compiler bug?
Sorry if this is the wrong list, but could people tell me if this bug
is still in the C compiler? The error message is:
"foo.c", line 21: compiler error: no table entry for op REG
typedef struct int_board_pos { /* internal rep of board */
int white, red, kings;
int border; /* which squares have highlight */
} int_board_pos;
display_board(current)
int_board_pos *current;
{
register int changes;
register int white, red;
register int white_kings, red_kings;
register int prev_white_kings, prev_red_kings;
static int_board_pos previous = { 0, 0, 0, 0 };
/* locate all squares that have some change */
changes =
current->border ^ previous.border |
current->white ^ previous.white |
current->red ^ previous.red |
white_kings ^ prev_white_kings |
red_kings ^ prev_red_kings;
}
Thanks, Ken
------------------------------
Date: 12 Jan 87 21:54:02 GMT
From: <jjg at linus.UUCP> Jeff Glass
Subject: nd and arp problems?
I have a SUN-3/75 which boots via nd from a SUN-3/180. While booting, the
ARP table entry for the diskless node changes from 8:0:20:1:3d:ee (good)
to da:0:0:0:0:0 (not good) on the diskless node and on the nd server. For
some reason, this works fine, until the diskless node is rebooted. Then
rarpd on the nd server does not recognize the diskless node until an
"arp -d disklessnode" is done on the server, so the diskless node repeatedly
gets "tftp timeout" messages. I have a SUN-3/110C which experiences exactly
the same problem, only its ARP table entry gets changed to ca:0:0:0:0:0.
However, there are seven other SUN-3s booting from the same nd server which
do not experience this problem.
I presume that nd is at fault because the ARP table entry for the diskless
node on the nd server does not change until just before /etc/rc is executed;
that is, it does not change until late in the boot. If I boot the diskless
node in single user mode, the ARP table entry becomes incorrect as usual.
If I then manually delete that ARP table entry, it remains deleted as long
as the diskless node is inactive. Once I execute a command on the diskless
node, an incorrect ARP table entry is created. No nfs filesystems are
mounted, and yp is not executing.
I periodically run "arp -d disklessnode" from cron on the nd server so that
the diskless node can reboot if need be, but I would like to know what is
happening, and how to fix it. Thanks/jeff
------------------------------
Date: 14 Jan 87 18:50:04 GMT
From: <galvin at uo.utexas.edu> Peter Galvin
Subject: Sun terminal emulation?
I'm using a sun 2 running sun unix v3.0, and I need to talk to Dec-20's.
The problem is the terminal type of a shelltool is not quite a vt100, and
so using editors et al on the 20's produces wierd results in my windows.
Is there any way to do real vt100 (or other terminal) emulation in a window?
Sure would make life fun.
--Pete
------------------------------
End of SUN-Spots Digest
***********************
More information about the Mod.computers.sun
mailing list