Partially functional diskless workstation

IRIX and IRIX software discussion including open source and commerical offerings.
Forum rules
Any posts concerning pirated software or offering to buy/sell/trade commercial software are subject to removal.
User avatar
pierocks
Donor
Donor
Posts: 528
Joined: Sat May 20, 2006 4:52 pm
Location: Longbranch, WA
Contact:

Partially functional diskless workstation

Unread postby pierocks » Fri Sep 16, 2016 3:21 pm

So I've been playing around with my new (to me) o2, and I decided to mess around with setting it up as a diskless system...the only "problem" being that I only have one SGI. So I decided to roll up my sleeves, and misapply the Diskless Workstation guide to set up my linux-based nas as the diskless workstation server. Long story short, I had some amount of success. My o2 boots and I am able to log in...kind of. The default shell (csh) seems to crash, but changing things to use sh works A-OK. Running an "ls" in most parts of the root filesystem barfs like this:

Code: Select all

# ls /usr
Exceeded file size limit(coredump)


I know the error message means it can't write a core dump to the current directory, but that's all I get, so I included it. Oddly, running it in an nfs mount works:

Code: Select all

# ls /nas
backups    cygwin     debra.img  foscam     media      virt       www
core       dbassett   diskless   irix       testfile   wap


The funny thing (to me, anyway) is that everything is mounted with NFS...the directory that I can list things in is just mounted in manually after the machine has booted.

Various other things fail as well. For example, I could get inst to start using my local nekoware mirrors as sources, but it dies with a memory error after 12% of reading the product descriptions.

This machine did work normally when it had a disk installed in it. At least, as much as I played with it (which was not much).

I doubt it's meaningful at all, but I am accessing this thing over a serial console with console=d. The monitors I have on my desk currently are only partially SoG compatible (everything is green), so serial was the solution.

Does anybody have any idea what's going on here? I only ever knew the basics about Irix...I'm at a loss as to where to even begin debugging.
:O2: :O2: :O2:
They are multiplying. If you can read this, send help.

surrealdeal
Posts: 153
Joined: Sat Oct 19, 2013 12:27 pm
Location: chicago

Re: Partially functional diskless workstation

Unread postby surrealdeal » Sat Sep 17, 2016 9:09 am

I'm not an expert on NFS, but I recall reading about some flaw in how certain versions of it have file-size or number-of-file-handles-open problems. Maybe you have something over 2gb in /usr.

someone purportedly got around a similar error using CIFS, but I don't know if you can run bootp from it.

User avatar
pierocks
Donor
Donor
Posts: 528
Joined: Sat May 20, 2006 4:52 pm
Location: Longbranch, WA
Contact:

Re: Partially functional diskless workstation

Unread postby pierocks » Mon Sep 19, 2016 11:18 am

It definitely seems to be nfs related, actually. I don't think it's those specific things you mentioned, but on a whim I remounted /var after the system had booted (which is mounted in as a separate nfs mount), and I am now able to run ls in /var. Interestingly, extra nfs options show up after remounting manually. I tried forcing the options in /etc/fstab, and this is what I got:

Code: Select all

# mount
10.0.0.1:/nas/diskless/client/susanne on / type nfs3 (vers=3,rw,proto=tcp)
10.0.0.1:/nas/diskless/client/susanne/sbin on /sbin type nfs3 (vers=3,rw,proto=tcp)
10.0.0.1:/nas/diskless/swap/susanne on /swap type nfs3 (vers=3,rw,proto=tcp)
10.0.0.1:/nas/diskless/share/IRIX_65-32/usr on /usr type nfs3 (vers=3,rw,proto=tcp)
10.0.0.1:/nas/diskless/client/susanne/var/share on /var/share type nfs3 (vers=3,rw,proto=tcp)
10.0.0.1:/nas/diskless/client/susanne/var on /var type nfs (vers=3,rw,proto=tcp,dev=140000)
/hw on /hw type hwgfs (rw)
/proc on /proc type proc (rw)
/dev/fd on /dev/fd type fd (rw)
# ls /var
X11            cron           inst           preserve       sysgen
adm            dhcp           mail           run            tmp
arch           dmedia         netls          share
boot           esp            ns             spool
config         flexlm         opt            sysadmdesktop
# ls /sbin
Exceeded file size limit(coredump)


The only difference between working and non-working nfs mounts seems to be the "dev=something" option. This has universally been present whenever I have been able to list files in NFS shares mounted on this box, and absent on shares that do not work.
:O2: :O2: :O2:
They are multiplying. If you can read this, send help.

User avatar
pierocks
Donor
Donor
Posts: 528
Joined: Sat May 20, 2006 4:52 pm
Location: Longbranch, WA
Contact:

Re: Partially functional diskless workstation

Unread postby pierocks » Mon Sep 19, 2016 5:50 pm

I seem to have gotten things going (though time will truly tell). Turns out I should have heeded this warning (that I probably skimmed right past) in the documentation:

6. Are there NFS export conflicts?

Be sure that the parent directory of your diskless tree is not listed as an NFS export. This condition causes NFS conflicts, which prevents the installation scripts from exporting diskless directories correctly and can also compromise security on the exported files. For example, if the diskless tree is installed in the directory /usr/diskless, the /usr directory should not be exported. To check for export conflicts, examine the server's /etc/exports file. If the parent directory of the diskless tree is listed, adjust the NFS configuration before implementing the diskless server.
:O2: :O2: :O2:
They are multiplying. If you can read this, send help.

Y888099
Posts: 523
Joined: Mon Jul 11, 2016 11:00 am

Re: Partially functional diskless workstation

Unread postby Y888099 » Mon Sep 19, 2016 11:33 pm

why NFS sucks?
well, one of the shortcomings of NFS v2 was its lack of cache consistency
and other major and minor things ...
so, if the worst thing that can happen does happen,
you'd better be prepared for the worse:
coherence loss is even worse than file loss
Head Full of Snow. Lemon Scented You

User avatar
dexter1
Moderator
Moderator
Posts: 2743
Joined: Thu Feb 20, 2003 6:57 am
Location: Zoetermeer, The Netherlands

Re: Partially functional diskless workstation

Unread postby dexter1 » Wed Sep 21, 2016 5:02 am

Nice work, Pierocks!

It depends whether this is actually useful for something else than either having no disk at all or suffer from the whining of one with bad bearings, but if it runs, it runs :) . The O2 has a PROM supported 100mbit network interface which makes this option actually viable. I dread the thought having to work on a diskless Indy with 10 mbit connection :o

I myself use a Synology for /home directories shared via nfs and it works fine with old SGI gear: Having at least a 100 mbit connection to the NAS is preferable if you want to do some work as well.

I found out that performing unittests in perl builds sometimes fail on NFS volumes, due to slow transfers or congestion on the internal network. So i stopped doing that and switched to test builds on a dedicated disk on a fast O2.
:Crimson: :PI: :Indigo: :O2: :Indy: :Indigo2: :Indigo2IMP:

User avatar
pierocks
Donor
Donor
Posts: 528
Joined: Sat May 20, 2006 4:52 pm
Location: Longbranch, WA
Contact:

Re: Partially functional diskless workstation

Unread postby pierocks » Wed Sep 21, 2016 9:18 am

dexter1 wrote:Nice work, Pierocks!
It depends whether this is actually useful for something else than either having no disk at all or suffer from the whining of one with bad bearings, but if it runs, it runs :) . The O2 has a PROM supported 100mbit network interface which makes this option actually viable. I dread the thought having to work on a diskless Indy with 10 mbit connection :o


The main motivations were A) eliminate the power draw and heat of an old, tired SCSI disk, B) Remove some unwanted noise, and C) Just for the challenge.

I have run into some more issues with my NFS server that I'm trying to debug at the moment. I had moved the /diskless export to the root of the filesystem when debugging the original NFS issues. This worked, but unfortunately, the root filesystem of my NAS is on an SD card, which connects to the system internally via USB, so it is sssssssslllllllllllooooooooooowwwwwww. When I tried to move everything back to the much faster SATA storage, everything went sideways again...even after editing the nfs exports to not export any of the parent directories of my diskless tree. I'll see if I can determine a root cause for all of this...

I sure wish the o2 had gigabit...even if it couldn't use all of it due to CPU limitations. I know about the p1000, but who knows if that would work for this setup...
:O2: :O2: :O2:
They are multiplying. If you can read this, send help.

Y888099
Posts: 523
Joined: Mon Jul 11, 2016 11:00 am

Re: Partially functional diskless workstation

Unread postby Y888099 » Thu Oct 06, 2016 8:28 am

any progress ?
Head Full of Snow. Lemon Scented You

User avatar
pierocks
Donor
Donor
Posts: 528
Joined: Sat May 20, 2006 4:52 pm
Location: Longbranch, WA
Contact:

Re: Partially functional diskless workstation

Unread postby pierocks » Thu Oct 06, 2016 8:56 am

Yeah. Been too busy to do a writeup. I have a few things I want to test out first, and then I will go through the whole thing again and document it step by step.
:O2: :O2: :O2:
They are multiplying. If you can read this, send help.

User avatar
jodys
Donor
Donor
Posts: 129
Joined: Tue Feb 06, 2007 1:33 am
Location: Denver, CO

Re: Partially functional diskless workstation

Unread postby jodys » Thu Oct 06, 2016 9:41 am

The P1000 is marginally faster than the onboard ethernet, but at the cost of all of your CPU. I bet that jumbo frames would help, but P1000 doesn't support it, AFAIK. To be fair, I don't think you could ever get 100MB over the O2 UMA, let alone over the PCI bus.

Ian Mapelson wrote:Observe how the STREAM bandwidth figures in the following diagram (MB/sec), for a 200MHz R5000SC O2 running IRIX 6.3, gradually improve (ie. increase) as the display complexity decreases:

Display Complexity Copy Scale Add Triad

1280-1024-32-32-75 69.2 69.1 69.6 70.7
1280-1024-16-16-75 74.1 73.1 73.4 74.9
1280-1024-16-0-75 76.6 75.5 74.3 76.2
640-480-16-16-60 82.7 81.3 82.1 83.0
640-480-16-0-60 83.2 81.2 82.2 83.2
:Octane2: :Indy: :O2: :O2: :O3x0: :Indigo2IMP:

User avatar
pierocks
Donor
Donor
Posts: 528
Joined: Sat May 20, 2006 4:52 pm
Location: Longbranch, WA
Contact:

Re: Partially functional diskless workstation

Unread postby pierocks » Thu Oct 06, 2016 10:45 am

jodys wrote:The P1000 is marginally faster than the onboard ethernet, but at the cost of all of your CPU. I bet that jumbo frames would help, but P1000 doesn't support it, AFAIK. To be fair, I don't think you could ever get 100MB over the O2 UMA, let alone over the PCI bus.

Ian Mapelson wrote:Observe how the STREAM bandwidth figures in the following diagram (MB/sec), for a 200MHz R5000SC O2 running IRIX 6.3, gradually improve (ie. increase) as the display complexity decreases:

Display Complexity Copy Scale Add Triad

1280-1024-32-32-75 69.2 69.1 69.6 70.7
1280-1024-16-16-75 74.1 73.1 73.4 74.9
1280-1024-16-0-75 76.6 75.5 74.3 76.2
640-480-16-16-60 82.7 81.3 82.1 83.0
640-480-16-0-60 83.2 81.2 82.2 83.2


Yeah, sometimes I have to remind myself that this machine was introduced two decades ago. It's not even going to keep up with a modern netbook in a lot of cases. But that's why my o2 will be my office jukebox and my daily driver is a 12 core xeon with 96 GB of RAM and not the other way around...
:O2: :O2: :O2:
They are multiplying. If you can read this, send help.


Return to “IRIX and Software”

Who is online

Users browsing this forum: No registered users and 1 guest