Sun-Spots-Request@RICE.EDU (William LeFebvre) (04/18/88)
SUN-SPOTS DIGEST Monday, 18 April 1988 Volume 6 : Issue 57 Today's Topics: Re: Can backups be done with system in multi-user mode (2) Re: 8mm video-tape for backups Re: More information on 892 MB drives (and a big problem) Memory Modules for Sun 3/60's Timeshared Suns - some experiences Interested in known DNI bugs. 3rd party SCSI hosts adaptors needed Sun 3/50 clock keeps resetting Sun4 Optimixer problems? Longer host names in rcp/rdump? 'cp' silently failing on NFS filesystems? mail from uuxqt? 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: Sat, 09 Apr 88 11:26:47 EST From: davy@intrepid.ecn.purdue.edu (Dave Curry) Subject: Re: Can backups be done with system in multi-user mode (1) Reference: v6n49 >From: montnaro@sprite.steinmetz.ge.com (Skip Montanaro) > the problem with multi-user backups is that a directory inode might get > deleted and then reallocated between dump passes. I posted mods to 4.3BSD dumptraverse.c about a year ago to the net to fix this so that you can safely dump live file systems (we do it all the time on 16 VAXes, 4 Goulds, and about 10 Sun servers, and have been for about 2.5 years with no problems). If there's enough interest, I can send them to Sun-Spots too; they're short. (and our illustrious moderator comments....) > I suppose it would be possible to write a signal handler that could write > unrecoverable information (such as open files, signals, etc.) out to a > file before dumping core... I did this once on our VAX 4.3BSD systems. It's in the form of a library called "chkpt" (checkpoint), which allows the user to call it and save his program, and also catches SIGTERM and saves the program automatically. The program is saved in executable form (0407 file, unfortunately) in the file "chkpt.dat" with it's start point set to an environment restoration routine. So, the user can simply execute "chkpt.dat" to restore his program. When the user executes "chkpt.dat", the program reopens all its files, seeks to the right places, resets its signal handlers, and so on, and then picks up execution right where it left off. Pretty much everything the user could possibly do system-call wise is restored. The idea was to allow long-running programs to make themselves relatively immune to shutdowns and system crashes. I've never ported it to the Suns because not too many people were ever interested in it on the VAXes. It should be fairly simple though; you'd need a source license (or need to know a lot about how the C library system call interface works) and have to write about 5 lines of assembly language to save and restore the stack pointer, frame pointer, and program counter. --Dave Curry Purdue University ------------------------------ Date: Mon, 11 Apr 88 11:13:09 -0400 From: dan@wilma.bbn.com Subject: Re: Can backups be done with system in multi-user mode (2) Using SIGQUIT won't work both because you can't continue programs you've core-dumped, only restart them (with undump), and because many programs ignore SIGQUIT. SIGILL might be better--assuming the program has no handler for it... Here's a different idea: what if you sent long-running processes SIGSTOP, did the dump, then sent SIGCONT? Then there should be no way a directory inode can be deleted and reallocated between passes. I've had a good deal of success with careful use of "kill -STOP" and "kill -CONT" when I wanted to stop some long-running massive job for a little while so I could use memory and/or CPU cycles. I always sent the signal to the youngest or almost-youngest child in the process stack, not to any of the others. I am not sure how easy it would be to automate this. But maybe sending SIGSTOP to everything except your own process tree would work ok. Dan Franklin dan@bbn.com ------------------------------ Date: Fri, 8 Apr 88 20:51:47 EST From: umix!lokkur!scs@rutgers.edu (Steve Simmons) Subject: Re: 8mm video-tape for backups Reference: v6n45 milano!sandel@tuvalu.sw.mcc.com (Charles Sandel) writes: >I have evaluated the Exabyte 8mm videotape backup system from Delta >Microsystems. It works very well.... and proceeds to give an thorough and largely positive review. I'd like to second his remarks, and add a few. We are using an Artecon-assembled 8mm drive. Like the one he describes the drive mechanism is by Sony, the controllers and other electronics are by Artecon. In all our research on 8mm drives they have all come down to these two manufacturers. The Artecon unit came with tape driver software for SunOS 3.4 and earlier. Our unit works with the standard 3.5 /dev/st device -- no driver needed. We *did* have to disable SCSI disconnect in the kernal (which degraded system performance) but this is fixed in a later release of the controller boards. Thus far our testing results match his, but we have only tested on the 3/50. Throughput is reasonable -- not great, not terrible. Our preliminary figures say we can back up a 1.4GB server between midnight and 7AM, which is all we're aiming for. Thus far it looks slightly faster than cartridge, slower than magtape. We're ordering a unit for our servers, and will post performance data when we have more. We expect backing up SMD disks to SCSI tape will perform better than SCSI disk to SCSI tape. To "torture-test" the drive we've been doing 3-4 backups per day for about a month, using the same tape. No problems so far. The unit we have is a standalone "shoebox" style. We hooked it in series with our existing cartridge drive and are able to use both with no conflicts. >I am convinced that the drive and the electronics work. My big concerns >here are all about the media itself (the tape)... >the low cost of the tapes ... at $8 apiece for 2Gb of storage This is the primary attraction for us. We can cut our media costs by at least 75%. On long-term storage: nobody knows. The vendors predict reasonableness, but the real answer will come when someone picks up a five year old tape and tries to read it. We use a vault for long-term storage so rust probably isn't a problem, but when was the last time you (or anybody else) retensioned a 5-year old tape from a vault? Or tried to read one, for that matter? Steve Simmons UNIX Systems Mgr. Schlumberger CAD/CAM scs@lokkur.uucp [[ Although I hate to admit that I have such experience, I used to put long term backups for a Heathkit computer onto cassette tape. Reliability of the medium was vastly improved when we started buying medium-high quality tapes (specifically, Maxell UD tapes instead of Radio Shack). In short, the better brand name tapes should last longer so the slightly higher price should be worth it. --wnl ]] ------------------------------ Date: Mon, 11 Apr 88 15:05:10 N From: Johan Widen <jw@sics.se> Subject: Re: More information on 892 MB drives (and a big problem) We have acquired two NEC D2363 drives together with an Interphase controller and driver routines from Holistic Technology. We seem to get considerably more than 892 MB from our drives. Our drives are formatted with 1021 cylinders + 3 alternates 70 sectors per track (no sector slipping) This comes out to (there are 27 heads): 1021*70*27*512 = 988001280 bytes while Sun (who do provide for sector slipping) according to the article in Sun-spots vol. 6.33 provide 964 cylinders + 2 alternates 67 sectors per track + 1 for sector slipping This comes out to 964*67*27*512 = 892864512 We too, got terrible kB/inode ratios. I have not yet got around to fixing the mkfs command. Johan Widen SICS, PO Box 1263, S-164 28 KISTA, SWEDEN Tel: +46 8 752 15 32 Ttx: 812 61 54 SICS S Fax: +46 8 751 72 30 Internet: jw@sics.se or {mcvax,munnari,ukc,unido}!enea!sics.se!jw ------------------------------ Date: Sun, 10 Apr 88 00:34:42 -0400 From: Henry B. J. Krempel <krempel@pacrat.npac.syr.edu> Subject: Memory Modules for Sun 3/60's I don't want to cause a rush here, but I got 4MB for my '60 from a company called Clearpoint. We got it in 2 weeks, put it in ourselves, and haven't had a problem yet. (a whole month) Henry B. J. Krempel <krempel@pacrat.npac.syr.edu> Northeast Parallel Architectures Center (NPAC) Syracuse University 250 Machinery Hall Syracuse, N.Y. 13244 ------------------------------ Date: Mon, 11 Apr 88 14:20:25 +0930 From: Kevin J. Maciunas <munnari!cs.flinders.oz.au!kevin@uunet.uu.net> Subject: Timeshared Suns - some experiences Since there have been many inquiries to Sun Spots about Suns (particularly 4's) as time-shared Unix systems, I thought I might share our experiences here at the CS dept, Flinders University. Background: We have had a 3/280 with 16MB, 32 lines (ALM-1's) and a Super Eagle/451 for about 18 months as a time-shared system for Undergraduate teaching. We have just installed two more Suns (both 4's, 32MB, 451's etc) to support all our Undergrad CS students. Experience: We were forced to buy the Sun-3 with the ALM boards - we wanted ethernet terminal servers (Encore Annexes).. We got the Annexes with the Sun-4's and I personally would highly recommend them. Last years load on the Sun-3 was quite high (it was used only by second year students) with about 24-30 active users doing program developement in Pascal and Modula-2 (term I), COBOL [yech] (term II) and SCHEME [MIT version] (term III). SCHEME in particular was a real test with only 16MB (how do you say "paging"?) At the same time we had a Pyramid 90x (8MB) and the same sorts of offered load resulted in load averages of 30+ (how do you say "S-L-O-W"!). While the Sun-3 certainly slowed down, it was by no means unacceptable (csh prompts came up virtually instantly). One interesting note was the fact that the students tended to use pi/px for their pascal code, one 1,100 line assignment could be compiled under pi in 4.3 secs (real time) with about 20 or so users on the 3. This tended to produce quick turn around on the edit->compile cycle, regrettably omitting the "think" part in between! Sun-4: The Sun-4's perform much as we expected. Both the 3 and the two 4's are I/O limited (wait for the 7053's - they should help things along nicely) but the 4's cope effortlessly given our job mix. There are a few teething problems though, three Highlights: o SunOS3.2L4 NO SUBNET SUPPORT! ARRGGHH. o " Bug in pty support o " No pi/px The subnet support comes in 4.0, the pty bug is that the kernel only recognises 32 pty's, which makes people with terminal servers unhappy! [Sun have supplied a fix - we now have 64 usable]. Pi/Px is now unsupported. This is a minor problem for us for the moment, as pi/px is quite nice - particularly for first--year pascal assignments etc - heck - even my final year compiler construction course have used it! Our (temporary) solution has been to port it from the 4.3 tapes - THIS IS VERY DIFFICULT TO DO. Any one wanting the "ported" version can contact me direct. Conclusion: While you certainly can't jam 100's of users on a Sun-4, given a hefty boost in the I/O dept, you could put 60 or so. We run very nicely indeed with 30->48 users with no problems. /Kevin Kevin J. Maciunas ACSnet: kevin@cs.flinders.oz Discipline of Computer Science, Internet: kevin@cs.flinders.oz.au Flinders University, UUCP: ..seismo!munnari!flinders!kevin South Australia, 5042. 'phone: +61 08 275 2129 ------------------------------ Date: 11 Apr 88 21:56:16 GMT From: laic!darin@decwrl.dec.com (Darin Johnson) Subject: Interested in known DNI bugs. I would like to know of bugs with the Sunlink/DNI package (DECnet). It is possible that it may have crashed one of our vaxes awhile back. (does anyone know of a 'user' at sun.com to mail questions like this to?) Darin Johnson (...ucbvax!sun!sunncal!leadsv!laic!darin) (...lll-lcc.arpa!leadsv!laic!darin) ------------------------------ Date: 11 Apr 88 17:21:59 GMT From: dck@mtunn.att.com (David Kwong) Subject: 3rd party SCSI hosts adaptors needed Does anyone know of 3rd party SCSI host adaptors one can plug into the SUN-3s and SUN-4s ? (I have SUN-4/260s and SUN-4/110s, but I understand that they use the same cabinets and backplanes as the SUN-3s.) Yes, I know that they have announced a SCSI shoebox option, but it doesn't allow one to easily plug another SCSI disk in and get it to work. Thanks in advance. David C. Kwong (201)957-2657 ------------------------------ Date: 11 Apr 88 21:28:30 GMT From: ndmath!ndcheg!evan@iuvax.cs.indiana.edu (Evan Bauman) Subject: Sun 3/50 clock keeps resetting We've had a Sun 3/50 for just about a year, and lately, when we reboot we get a message that the clock has lost 29 days. It also says to please check the clock and reset. This has only just started happening. My first guess is that there's a battery that needs replacing. But before I rip the thing apart, I'd like someone to confirm my diagnosis. If it is a battery, where should I get a replacement? BTW, we're running SunOS 3.5. I looked at the patches posted back in volume 6 #1 of sun-spots and I'm pretty sure we have a different problem. Thanks in advance. Evan Bauman University of Notre Dame ..!iuvax!ndcheg!evan [[ I think you do have a different problem. The patches in v6n1 fixed the "creeping clock" bug---the clock would be running too fast or too slow and would gain/lose a significant amount of time while the system was running. You should install the patches anyway, if you haven't already. --wnl ]] ------------------------------ Date: Mon, 11 Apr 88 13:19:24 EDT From: schwartz@gondor.cs.psu.edu (Scott Schwartz) Subject: Sun4 Optimixer problems? Several months ago I noticed that nroff and troff didn't work on the Sun4's here. I submitted a bug report to Sun and to sun-spots. (The service order number is 143015). Moreover, as was reported to Sun, compiling the 4.3BSD troff (which runs fine on a Sun3 or a Vax) fails in EXACTLY the same way the SunOS4-3.2 troff does. Conclusion: a compiler problem, obviously. I have received no followup information, so I guess they can't fix it :-) Sigh. Anybody want to port GCC? -- Scott Schwartz ------------------------------ Date: Mon, 11 Apr 88 17:08:41 EDT From: weltyc@csv.rpi.edu (Christopher A. Welty) Subject: Longer host names in rcp/rdump? Is there an easy way one can modify rcp (and rdump) to take host names longer than 10 characters. This is a really annoying bug for people without src.... Christopher Welty --- Asst. Director, RPI CS Labs weltyc@cs.rpi.edu ...!rutgers!nysernic!weltyc ------------------------------ Date: Tue, 12 Apr 1988 08:29-EDT From: Ralph.Hyre@ius3.ius.cs.cmu.edu Subject: 'cp' silently failing on NFS filesystems? I was wondering if anyone else has encountered problems with 'cp' on NFS mounted filesystems. In a few cases, it seems to update the access time of the destination file, but the contents are not overwritten. I don't believe any of the servers or networks was particularly busy at the time. Any clues or workarounds appreciated. Thanks. - Ralph ------------------------------ Date: Sat, 9 Apr 88 21:07:00 EDT From: brad@cayman.cayman.com (Brad Parker) Subject: mail from uuxqt? I'm on a 3/50 with SUNOS 3.2. Every night root gets mail from uuxqt as "daemon". We fire up several uucp polls each night from cron. I've also seen uucp get mail from "daemon". Anyone else seen this or know what's causing it? (ps: I know that uuxqt is part of uucp. I understand how and why it is started - I just don't know why I'm getting mail from it) From daemon Sat Apr 9 13:46:35 1988 Received: by cayman.Cayman (3.2/smail2.5/12-02-87) id AA10319; Sat, 9 Apr 88 13:46:33 EDT Date: Sat, 9 Apr 88 13:46:33 EDT From: root (Operator) Message-Id: <8804091746.AA10319@cayman.Cayman> To: root uuxqt cmd () status (exit 0, signal 0) -brad brad@cayman.com ------------------------------ End of SUN-Spots Digest ***********************