Sun-Spots-Request@RICE.EDU (William LeFebvre) (08/02/88)
SUN-SPOTS DIGEST Sunday, 31 July 1988 Volume 6 : Issue 158 Today's Topics: Re: login problem when NFS server crashes (3) Re: tn3270 on a Sun4 Re: Experiences with 4.0 Re: A question regarding setting subnet masks Re: Level 2 interrupt problems (3/280) Experiences with the Sun-4 Performance problem with 3/280 upgrade Problems with X11R2 xterm and SunOS 4.0 Changing Sun 3/110 LC to monochrome? Other NFS Servers? Disk for 386i? 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, 19 Jul 88 10:43:51 EDT From: nesheim@think.com Subject: Re: login problem when NFS server crashes (1) The other way you loose when logging into a machine with nfs directories of a crashed server mounted is if you have quotas turned on. Login calls /usr/ucb/quota, which attempts to contact the crashed server to find out if you're over your disk quota. This will happen even if your nfs directories are not mounted in root! root logins are not checked for quotas, so they go nice and fast. To avoid this mount all remote file systems not only "soft", but also "noquota". Or better yet rm /usr/ucb/quota and cp /bin/true /usr/ucb/quota! --- Bill Nesheim; Thinking Machines Corporation, Cambridge, MA +1 617-876-1111 nesheim@think.com, {mit-eddie,ihnp4}!think!nesheim [[ John Myers at CMU also pointed this out. --wnl ]] ------------------------------ Date: Tue, 19 Jul 88 13:15:23 EDT From: Mike Muuss <mike@brl.arpa> Subject: Re: login problem when NFS server crashes (2) The problem that you describe is quite real. The way we have avoided this problem at BRL depends on two factors: 1) We use the Doug Kingston NFS modification that allows us to mount an entire server machine's filesystem tree with a single NFS mount, and 2) We mount each system as /n/machinename, so that there are no NFS mounts lurking in the root directory. This keeps the root directory very small (for fast searching), and prevents getwd() from tripping over downed servers. If folks are unfamiliar with the Doug Kingston NFS modification, I would be willing to provide source mods to the Digest -- they are only a few lines long, and work fabulously. It's the only way we can survive with dozens of moby servers, here at BRL. Best, -Mike ------------------------------ Date: Tue, 19 Jul 88 02:04:01 EST From: Dan Trinkle <trinkle@purdue.edu> Subject: Re: login problem when NFS server crashes (3) There are two solution to the pwd stat problem. One is to put NFS mounts in a separate subdirectory. However, if you want to give precedence to one NFS mount, this does not do much good. The other method, as wnl pointed out, is to sort the actual entries in the root directory. Here is a (hideous hack of a) shell script which does just that automatically. It must be run in single user mode. -------------------- cut -------------------- #! /bin/sh # # Usage: shuffle-dirs [preferred mount points] # This script reorders the directory entries in /, putting all local # files (non-mount points), all local mount points, all preferred # mount points, then all remaining NFS mount points. For the local # mount points and the NFS mount points, the order is determined by # /etc/fstab. This script is highly dependent on the order in which # things are done. # # ENTRIES - list of current / entries minus ".", "..", and "lost+found" # PREF - preferred NFS mount points (command line args) # LOCAL - local mount points # REMOTE - NFS mount points # # TMPNAME - name of a non-existent file in / (the shorter the better) # # Daniel Trinkle # Computer Science Department # Purdue University # January 13, 1988 # TMPNAME=a PATH=/bin:/$TMPNAME export PATH MNTTAB=etc/fstab ROOTDIR=/ PROG=shuffle-dirs USAGE="$PROG: [-d <rootdir>] [preferred NFS mount points]" case $1 in -d) ROOTDIR=$2 shift; shift ;; esac if [ -n "$ROOTDIR" -a -d "$ROOTDIR" ]; then cd $ROOTDIR else echo "$PROG: Bad root directory: $ROOTDIR" echo "$USAGE" exit 1 fi ENTRIES=`ls -f | sed -e '/^\.$/d' -e '/^\.\.$/d' -e '/^lost+found$/d'` PREF=$* LOCAL=`awk '{if ($3 == "4.2" && split($2,m,"/") == 2) print m[2]}' $MNTTAB` REMOTE=`awk '{if ($3 == "nfs" && split($2,m,"/") == 2) print m[2]}' $MNTTAB` if [ -f $TMPNAME -o -d $TMPNAME ] ; then echo "$0: $TMPNAME exists, I cannot continue." exit 1 fi echo -n "Removing mount points:" for f in $LOCAL $REMOTE ; do rmdir $f echo -n " $f" done echo "." echo -n "Shuffling remaining entries:" for f in $ENTRIES ; do if [ -r $f ] ; then mv $f $TMPNAME; mv $TMPNAME $f echo -n " $f" fi done echo "." mkdir $TMPNAME echo -n "Creating local mount points:" for f in $LOCAL ; do if [ ! -d $f ] ; then mkdir $f echo -n " $f" fi done echo "." echo -n "Creating preferred NFS mount points:" for f in $PREF ; do if [ ! -d $f ] ; then mkdir $f echo -n " $f" fi done echo "." echo -n "Creating remaining NFS mount points:" for f in $REMOTE ; do if [ ! -d $f ] ; then mkdir $f echo -n " $f" fi done echo "." rmdir $TMPNAME exit 0 -------------------- cut -------------------- And yes, even a man page. -------------------- cut -------------------- .TH SHUFFLE-DIRS 8L "July 7, 1988" "Purdue University" .SH NAME shuffle-dirs \- reorder directory entries in / .SH SYNOPSIS .B /etc/shuffle-dirs [ .B \-d .I root-directory ] [ .I pref-list ] .PP .SH DESCRIPTION .I Shuffle-dirs is a script that reorders directory entries in /, specifically designed to move NFS mount points to the end of the directory. This prevents NFS timeouts for remote servers when accessing local file systems. .PP It reorders entries by removing all mount points (determined by looking in /etc/fstab), moving (in turn) each remaining entry to a bogus entry ("a" by default) and moving it back (filling the first available directory slot), creating all local mount points, creating preferred mount points (specified by the white-space separated .IR pref-list ), and then creating any remaining NFS mount points. The order of the local and remain NFS entries are determined by the order in /etc/fstab. .PP Only top level mount points (i.e. those in /) are manipulated. .PP The \fB\-d\fR option can be used to specify an alternate root directory. All work is done relative to this root directory, including the search for /etc/fstab. This option is useful for manipulating the root directory of a diskless client from the server. .SH FILES .ta 3i /etc/fstab list of local and NFS mounts .ta .SH BUGS This whole things is a terrible hack (but it works) and is extremely dependent on the way directories are structured. .SH AUTHOR Daniel Trinkle, Computer Science Department, Purdue University -------------------- cut -------------------- Daniel Trinkle trinkle@cs.purdue.edu ARPA Department of Computer Sciences trinkle%purdue.edu@relay.cs.net CSNET Purdue University {ucbvax,decvax}!purdue!trinkle UUCP West Lafayette, IN 47907 (317) 494-7844 PHONE ------------------------------ Date: Mon, 18 Jul 88 08:11:48 EDT From: Mike Jipping <jipping@cs.hope.edu> Subject: Re: tn3270 on a Sun4 John M. Vogtle: > We're trying to bring up tn3270 on a Sun 4/110 runnning SunOS 4.0... Most "old" (pre-4.0) communication products that use low-level I/O will not work right with 4.0. This is because the new OS uses the SYSV STREAMS interface instead of the old style. Yea -- they'll compile, but they'll bomb! That's why Sun is way behind on all it's 4.0-compatible communications products. Mike Jipping Hope College Department of Computer Science jipping@cs.hope.edu ------------------------------ Date: Tue, 19 Jul 88 09:27:02 EDT From: dpz@njin.rutgers.edu (David P. Zimmerman) Subject: Re: Experiences with 4.0 Reference: v6n144 > From: Tim Bray <tbray@watsol.waterloo.edu> > > 2. Irritating nit with /bin/time > > Namely, it's now /usr/bin/time. This was the only place the 4.0 directory > reorg bit me... Uh, I don't see why this would break your validation suite. /bin is just a link to /usr/bin under 4.0. David P. Zimmerman Systems Programmer Rutgers University ------------------------------ Date: Tue, 19 Jul 88 09:38:52 EDT From: dpz@njin.rutgers.edu (David P. Zimmerman) Subject: Re: A question regarding setting subnet masks Reference: v6n142 > From: "R. Wayne Little, Sci. Pgmr/Analyst" <C0018@UMRVMB.BITNET> > > How does one set a subnet mask on a SUN? > ... > It doesn't appear in the ifconfig(8) man page in the SUN doc.... Sounds like you're using SunOS 3.2 or earlier. Ifconfig under 4.0 lets you do what you want. Since SunOS 3.3-3.5 have subnetting code, I assume they allow the same thing - I'm not sure, we stuck with the stable, unsubnetted 3.2 :-). Unless you upgrade to a higher version of SunOS, have source, or are *very* good at hacking binaries, the only out I see is possibly to have a smart gateway sit between your Suns and the rest of your net. David P. Zimmerman Systems Programmer Rutgers la dee dah ------------------------------ Date: Tue, 19 Jul 88 09:49:51 EDT From: dpz@njin.rutgers.edu (David P. Zimmerman) Subject: Re: Level 2 interrupt problems (3/280) Reference: v6n143 Supposedly there is a bug with the IPC board driver that requires you to always configure all 4 pc devices (pc0-pc3) into the kernel, even if you only have one board. Don't think it'll help your subnetting problem, but I suspect those spurious interrupts should go away. David P. Zimmerman System Programerre Universitie Rutger ------------------------------ Date: Tue, 19 Jul 88 16:00:17 EDT From: Dan Franklin <dan@wilma.bbn.com> Subject: Experiences with the Sun-4 We tried porting a roughly 150,000-line C program to the Sun-4 a couple of months ago, using the "Sun porting center" near us. It was running some variant of Sun OS 3.2 at the time. We found two portability problems (a failure to use varargs in one place, and an accidental structure alignment dependency) and two other problems: 1. As with Tim Bray, we had a lot of regression tests comparing floating point output values, and we were *quite* irritated to find them failing for values between 0 and 1 due to the absence of the leading zero (that is, getting ".1" instead of "0.1"). We weren't just irritated because of the test scripts; we were also irritated because we may have to go back and fix our floating-point print statements to insert a leading zero in this circumstance. I doubt very much that our customers will appreciate having the difference between "12" and "0.12" be reduced to a single speck on their listings! This output change is so stupid and gratuitous that I suspect Sun didn't decide to do it, but rather picked it up from some brain-damaged AT&T change. 2. There was a problem involving fprintf calls which we never did figure out. The symptom was a core dump whose stack had _doprnt at the top; it appeared that the arguments to _doprnt had gotten garbled. But all attempts to produce a small test case demonstrating the problem failed. We had a similar problem with the version of the RCS code we use here. Almost any change made the problem go away. Mostly it went pretty smoothly; it was less than a day's work to verify that we had no major problems facing us. But I sure wish I knew what was going on with these two problems. Sun was, alas, no help with either one. Dan Franklin ------------------------------ Date: Tue, 19 Jul 88 13:13:35 EDT From: doug@icase.arpa (Doug Peterson) Subject: Performance problem with 3/280 upgrade We have just upgraded our server from a 3/180 to a 3/280, and have seen a DECREASE in performance. We run Sun OS 3.5. The number of diskless clients in both cases is the same. The e-net collision rate is much higher with the new server, which may account for some of the decrease. But I just received my copy of the May Sun STB, which has an article describing how the cache works on a Sun 3/200 series machine. On page 826, there is a statement to the effect that if the cache hit rate is effectively zero, then performance of a 3/280 system will be less than a 3/280. As a server, handling lots of random client requests, I would think that the cache hit rate would be very low. Has anyone observed similar phenomenon? Doug Peterson Systems Manager ICASE Mail Stop 132C NASA Langley Research Center Hampton, VA 23665-5225 (804) 865-4090 ------------------------------ Date: Tue, 19 Jul 88 17:25:08 CDT From: clyde@emx.utexas.edu (Head UNIX Hacquer) Subject: Problems with X11R2 xterm and SunOS 4.0 The X11R2 xterm fails when told to use any display other than 'unix:0.0'. This is on a SUN 3/[56]0, running SunOS 4.0. Witness the below debugging session... spica> dbx xterm ... (dbx) catch 1 (dbx) run -display spica:0.0 Running: xterm -display spica:0.0 stopped in main at line 244 in file "main.c" 244 int fd1 = -1; (dbx) c signal HUP (hangup) in close at 0x33ace close+8: rts (dbx) where close(0x52ad0) at 0x33ace yp_unbind(0x529f0) at 0x34063 yp_get_default_domain(0x529f0, 0xeffe980) at 0x34219 _yp_dobind(0x529f0, 0xeffe980) at 0x33eb5 yp_match(0x529f0, 0x4b887, 0xeffe9b4, 0x4, 0xeffe9cc, 0xeffe9c8) at 0x3574f getpwent(0x3ed, 0x51ea4) at 0x2fd59 getpwent(0xeffeb10, 0xeffef14, 0x3ed) at 0x2f973 getpwuid(0x3ed) at 0x2f36d spawn(), line 1046 in "main.c" main(argc = 0, argv = 0xefffc0c, 0xefffc18), line 549 in "main.c" (dbx) quit ... I have tried building xterm with and without optimization, the X libraries with and without optimization, dynamic and static linking, and I get the same error. Anyone have any ideas? I think that the Sun YP code is busted! -Clyde Hoover ------------------------------ Date: Tue, 19 Jul 88 14:44:30 EDT From: segall@caip.rutgers.edu (Ed Segall) Subject: Changing Sun 3/110 LC to monochrome? I have a Sun 3/110 LC whose display is too fuzzy to use. Smalltalk on the 13" color screen causes severe eyestrain. I wish to 'downgrade' to a monchrome screen, like the ones on on the 3/50 or 3/60. There doesn't seem to be a straightforward way to do this, since only RGB video outputs are available. Obvious possibilites are: Buy a whole new CPU and monitor that are set up for monochrome ---Too expensive "" "" and sell the current CPU and montor to recoup costs ---Seems difficult. Buy an RGB to composite mixer and a monochrome monitor ---I don't know if the image quality will suffer Sell the whole thing and buy a 3/60 ---This may be the best option if we can find a customer Any comments or suggestions? Thanks, Ed Segall uucp: {...}!rutgers!caip.rutgers.edu!segall arpa: segall@caip.rutgers.edu ------------------------------ Date: Tue, 19 Jul 88 06:34:31 EDT From: chuck@wooglin.scc.com (Charles Williams) Subject: Other NFS Servers? Does anybody have any experience using machines other than SUN's as NFS servers? For example, a VAX, or better yet a MultiMax from Encore Computers? If there is enough interest, a summary of all answers received will be posted. Please reply directly to: chuck@wooglin.scc.com or cwilliams@bluto.scc.com Chuck Williams Contel Federal Systems <Standard Disclaimer, etc.> ------------------------------ Date: Tue, 19 Jul 88 08:56 MDT From: "Samuel J. Cole" <SYSCOLE@chemistry.utah.edu> Subject: Disk for 386i? Greetings, Sun users! Does any of you know whether it is possible to use a 327-Mbyte disk/cartridge tape drive from a Sun 4/110 as a boot device for a RR-250? Thanks in advance for any help. Sam Cole Chemistry Computer Center University of Utah (801) 581-4696 Internet: cole@chemistry.utah.edu Bitnet: sjcole@UTAHCCA ------------------------------ End of SUN-Spots Digest ***********************