barrett@jhunix.HCF.JHU.EDU (Dan Barrett) (07/20/90)
I am having severe problems attempting to boot a DS3100 as a diskless node, using a DECsystem 5400 as the file server. A number of very helpful DEC employees are helping us try to solve this problem, but I am throwing the question out to you Net folks, since nobody has a solution yet (and we've been working on it for 3 days). If you have successfully done a diskless boot using "/etc/dms" under either Ultrix 3.1D or Ultrix 4.0, please tell me what we have done wrong!! We experience the same behavior under Ultrix 4.0 and Ultrix 3.1D. Our DS5400 is running Ultrix 4.0; the prerelease version that is identical to the "release" version but still has the SMP bug. Our server's name is "server" and our client's name is "whatever". (Yeah, cute; I know.) Here is what we did: (1) Did a fresh install of Ultrix 4.0, including the mandatory patch tape. We installed ALL SUBSETS, supported and unsupported. (2) Used /etc/dms to: 1. Create a /dlenv0 partition, by choosing 'c' and then 'd'. 2. Create a /dlclient partition, by choosing 'c' and then 'c'. 3. Install all the mandatory subsets from the 4.0 boot tape (tape 1 of 2), by choosing 'i'. 4. Add "whatever" as a client, by choosing 'a'. We told it to swap on local disk (partition rz3c), with no crash dump. 5. Booted "whatever" by typing: >> boot -f mop() PLEASE NOTE THAT, UP TO THIS POINT, WE DID NOTHING OUT OF THE ORDINARY. WE CHANGED NO CONFIGURATION FILES, LETTING dms DO IT ALL. (3) All three loaders (primary, secondary, tertiary) loaded, according to messages in /usr/spool/mqueue/syslog. "whatever" began booting. We saw Ultrix come up, say how much memory we have (12MB, with only about 8MB free), and then the list of devices. Then it said something like "/ mounted on /dlclient0/whatever.root, /usr on /dlenv/root.mips, swap on /dlclient0/whatever.root/dev/swap". [NOTE THE ERROR -- WE WANT TO SWAP ON LOCAL DISK rz3c.] Then it told us the swap size: 32K blocks. Then it hung. (4) So, after investigating, we discovered that /etc/dms did not compile the secondary loader, /dlclient0/whatever.root/etc/netblk.o, although it did update netblk.c with the current information. So, we compiled netblk.c by hand: # cc -c netblk.c and booted the client again. (Just to be safe, we totally removed the client directory and remade it first.) This time, it did not hang when it got to the swap line. Instead, after the device list, "whatever" flashed a very fast message across the display [too fast to see] saying something about "panic", and "whatever" reset itself (countdown from 7..0). We are stuck. Here are some answers to some common questions. o Yes, our disk is an RZ23 at SCSI address 3. o Yes, I ran nfssetup. NFS software running is: 4 nfsd 4 biod portmap mountd Question: Do we need lockd statd, and/or rwalld? o /etc/exports looks like: /dlenv0/root0.mips/usr -o -r=0 whatever /dlclient0/whatever.root -r=0 whatever This is exactly what is created by /etc/dms, and it matches the 4.0 release notes. The 3.1D line: /dlclient0 -n -r=0 nobody is no longer need in 4.0. (But we did try putting it anyway.) o /dlclient0/whatever.root/etc/fstab looks like: /dlenv0/root0.mips/usr@server:/usr:ro:0:0:nfs:: This is exactly what is created by /etc/dms. o We did try having the machine swap on the server. We did add the line: /dlclient0/whatever.root/dev/swap -r=0 whatever to /etc/exports in that case. Made no difference -- the client hung as usual. o It is not a hardware problem. We have tried several DS3100's, all with the same results. (Now maybe they're ALL broken...!) o I believe it is not an ethernet hardware problem. My VS2000 is on the same connection right now, and I've never experienced any problems. o I have noticed that there is no kernel config file called: /dlenv0/root.mips/usr/sys/conf/mips/WHATEVER Should there be? Who makes it? o Is anything missing from /etc/ttys, maybe? An entry for the console itself, instead of "/dev/console", maybe? We have a VS262 monitor. Any help is appreciated. I am totally clueless, grasping at straws. Dan //////////////////////////////////////\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\ | Dan Barrett - Systems Administrator, Computer Science Department | | The Johns Hopkins University, 34th and Charles Sts., Baltimore, MD 21218 | | INTERNET: barrett@cs.jhu.edu | | | COMPUSERVE: >internet:barrett@cs.jhu.edu | UUCP: barrett@jhunix.UUCP | \\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\/////////////////////////////////////
hagan@scotty.dccs.upenn.edu (John Dotts Hagan) (07/20/90)
I have had it with dms myself! Mark Litwack and I spent DAYS with dms trying to get it to be useful. However, no luck. Anyway, dms sucks because you HAVE to have an extra whole /usr environment loaded. In our case, we wanted to share the same /usr the server has with the clients (like SUN would let you do). So here is what we did to get our systems running, and save about 250 MEG. Just load a diskfull system (like the server) and copy his root partition to a file. Then, use standalone on the "diskless" systems with local swap disks, and rsh over the dd'ed root partition into the a partition of the local swap disk. Fix up some files, like the /etc/hosts and /etc/rc.local so the client system has its own personality (there are other files to fix up, too - if interested ask me and I will send you a list of the files we edit), then make /etc/fstab mount the server's /usr (make sure every system has its own /var somehow clever with symbolic links). That works for us. We take 16 MEG from our 104 MEG swap to be a root file system, and poof - no dms crap used at all! We can even use clients that are seperated by an IP router from their server (couldn't before - needed MOP_MOM to boot)! I have asked Digital REPEATEDLY to tell me how/if I can boot DECstation 3100's with TFTP, but all the people I asked NEVER UNDERSTOOD WHAT I WAS ASKING. THEY SENT ME SEVERAL WILD ATTEMPS TO ANSWER MY QUESTION THAT CLEARLY SHOWED THEY DID NOT EVER CATCH ON TO WHAT I WAS ASKING. I believe this is more or less what MIT Athena does to support workstations. I now see why... --Kid.
steved@longs.LANCE.ColoState.Edu (Steve Dempsey) (07/21/90)
In article <5876@jhunix.HCF.JHU.EDU>, barrett@jhunix.HCF.JHU.EDU (Dan Barrett) writes: > > I am having severe problems attempting to boot a DS3100 as a > diskless node, using a DECsystem 5400 as the file server. > [...stuff deleted...] > If you have successfully done a diskless boot using "/etc/dms" > under either Ultrix 3.1D or Ultrix 4.0, please tell me what we have done > wrong!! We have not done it with risc machines yet but have 100 diskless VS2000s. > [much detail deleted] > Then it said something like "/ mounted on /dlclient0/whatever.root, > /usr on /dlenv/root.mips, swap on /dlclient0/whatever.root/dev/swap". > [NOTE THE ERROR -- WE WANT TO SWAP ON LOCAL DISK rz3c.] You must edit the config file (/sys/conf/mips/WHATEVER?) and change the config line to something like: config vmunix swap on rz3c You may also have to add the controller & disk devices (copy from disked system config file) because the standard diskless config expects no disks (imagine that :-). > Then it hung. Hanging at this point indicates that nfsd may not be running on the server; check with nfssetup AGAIN and run ps and rpcinfo to verify that it's really alive. > (4) So, after investigating, we discovered that /etc/dms did not > compile the secondary loader, /dlclient0/whatever.root/etc/netblk.o, > although it did update netblk.c with the current information. So, > we compiled netblk.c by hand: Probably unnecessary since dms didn't bother with it. > Question: Do we need lockd statd, and/or rwalld? Useful, but not required. > ... [there is no config file] > > /dlenv0/root.mips/usr/sys/conf/mips/WHATEVER > > Should there be? Who makes it? Gotta be around somewhere. You can make your own in /sys/conf/mips and build the kernel that way. > o Is anything missing from /etc/ttys, maybe? An entry for the > console itself, instead of "/dev/console", maybe? We have a > VS262 monitor. Doubtful. You should see the `automatic reboot in progress...' and other output from rc & rc.local. Finally, as hagan@scotty.dccs.upenn.edu (John Dotts Hagan) writes in article <27278@netnews.upenn.edu>, you should consider sacrificing 15Mb for an `a' partition and mount /usr, et. al. from elsewhere. If I had a pile of RD53's, I'd put one in every diskless VS2000 and be done with it. Complete diskless booting still has its advantages, especially when you get a trashed boot block, or whatever. I've had occasion to boot a '3600 from a '2000 to fix broken file systems, etc. Much neater than loading the standalone from tape, especially on a machine with no tape drive :-). Hope this helps. Steve Dempsey, Center for Computer Assisted Engineering Colorado State University, Fort Collins, CO 80523 +1 303 491 0630 INET: steved@longs.LANCE.ColoState.Edu, dempsey@handel.CS.ColoState.Edu UUCP: boulder!ccncsu!longs.LANCE.ColoState.Edu!steved, ...!ncar!handel!dempsey