SGI: Hardware

Hi,

I have 4 Altix 350 nodes bought off ebay, one is specced with disks and an IO9 card, the others are barebones just containing CPUs and RAM.

I am using the L2 emulator running on a linux box to provide access to the console of the units.

On its own, the unit with the IO9 card works just fine, but once i connect any of the other units using numalink cables (no router, just cables) i get a system serial mismatch...

It seems that the 3 barebones units have the same serial, while the unit with the IO9 has a different system serial and consequently they won't link up.

Is there anything i can do, or will i have to move the cards, risers and disks from that unit to one of the others and operate the altix with just 3 nodes?

On another note, they also have mismatched firmwares which seem to be fairly old. Is it possible to get the updated firmware anywhere, at least to bring the machines all up to the same revision?

madmax wrote: Hi,

I have 4 Altix 350 nodes bought off ebay, one is specced with disks and an IO9 card, the others are barebones just containing CPUs and RAM.

I am using the L2 emulator running on a linux box to provide access to the console of the units.

On its own, the unit with the IO9 card works just fine, but once i connect any of the other units using numalink cables (no router, just cables) i get a system serial mismatch. ..

It seems that the 3 barebones units have the same serial, while the unit with the IO9 has a different system serial and consequently they won't link up.

Is there anything i can do, or will i have to move the cards, risers and disks from that unit to one of the others and operate the altix with just 3 nodes?

On another note, they also have mismatched firmwares which seem to be fairly old. Is it possible to get the updated firmware anywhere, at least to bring the machines all up to the same revision?

There is a section in the wiki on L2 commands ... http://www.nekochan.net/wiki/L1_and_L2_Commands
The "serial clear" command on the L1 of the problem brick responds with what for you...?

I have never seen an Altix but knowing a little about NUMAlink the if serial clear doesn't work with "system serial mismatch" then it might mean you have to do a little hardware hacking...
The MIPS based Origin 300 is obviosuly diffent from Altix but I removed the DS-1742W-120 from mine and replaced them with new ones..
http://www.nekochan.net/wiki/Use_a_rbri ... Origin_300
Look at the "See also" section for links back to forum topics, particualrily viewtopic.php?f=3&t=16724099&start=0
I think you might be covering new ground here, but if you understnad how the issue is resolved on the MIPS systems then you'll have an idea how investigate the problem on Altix systems.

Bonne chance..!

R.

Serial clear says:

INFO: command not supported on bricks that enforce security.

madmax wrote: Serial clear says:

INFO: command not supported on bricks that enforce security.

Then you need to investigate further...

Open the machine, take photos of the motherboard and post them here or on the wiki...
See if you or other members here can identify where/if the system serial number is stored, buy a new chip and replace it.. It should see the new chip as blank and you follow the steps in the o300 r-brick topic if you can.. It will set stuff from the L2 and you'll probably have to set the module iID.

I don't own an Altix but someone else who knows Altix hardware might step in and help you further.

I think this is new ground or it is at least to me.

R.

madmax wrote: It seems that the 3 barebones units have the same serial, while the unit with the IO9 has a different system serial and consequently they won't link up.

In theory it should be possible to get the serial numbers matched. But I haven't done it yet since my two bricks came with matching serial numbers.
First step is to get the PROM firmware up to date and matched on all bricks; maybe the serial number issue will go away. But that's a bit too optimistic I think. Anyway it's more than suboptimal to run a numalinked system with different firmware on the bricks.
I wrote detailed instructions flashing the PROM some time ago, you can find it here:
http://forums.nekochan.net/viewtopic.php?t=16725651#p7342874

Is there anything i can do, or will i have to move the cards, risers and disks from that unit to one of the others and operate the altix with just 3 nodes?

It may be annoying, but for flashing you'll have to move the IO9 card to the brick which has to be flashed. With your setup I would do the following procedure:
- First, flash the single brick with the IO9 card to the most recent PROM.
- Then, install the card in one of the three bricks with matching serial numbers, numalink all together and flash the remaining bricks.

If you have really old firmware chances are good that the L1 firmware is also old. I think it will be a good idea to also flash the L1 to a more recent one. That's possible from the L2 emulator. I'll have to do it on my bricks since on shutdown the OS complains about not being able to communicate with the L1 for storing the current time and date... but unfortunately I'didnt have the time for setting up the L2 emulator until now. The last days I began setting up the L2 the last days, so it's possible that I figure out how to do it in the next days (weeks?)...

[EDIT]
It would be nice to have more details on your hardware. Can you post PROM bootup messages from the differnt bricks?
L1 output is also interesting, especially from 'serial all'.

Some progress...

Got the firmware updated to 1.62.1 on all bricks..

Stepped through the firmware image in a hex editor and found some undocumented commands, one of which is "let the carnage begin"... executing this command on l1 turned the "security" feature off and reset the system serial to match the other bricks.

madmax wrote: Some progress...

Got the firmware updated to 1.62.1 on all bricks..

Stepped through the firmware image in a hex editor and found some undocumented commands, one of which is " let the carnage begin "... executing this command on l1 turned the "security" feature off and reset the system serial to match the other bricks.

Well done. There is an "undocumented" wiki topic and topics on L1 and L2 commands if you so feel inclined.

R.

madmax wrote: one of which is "let the carnage begin"

Someone that worked for SGI actually told me about that command some time ago, but I (or he) couldn't recall the exact verbiage so it fell by the wayside. Nice find!

madmax wrote: Stepped through the firmware image in a hex editor and found some undocumented commands, one of which is "let the carnage begin"... executing this command on l1 turned the "security" feature off and reset the system serial to match the other bricks.

This inspired me to fire up half of my O300 just to see what it would do:

Verrrry interesting... Might come in handy as I try to integrate the L2 and NUMALink Router I got via ePay (I believe from a parted-out Prism).