Operating System - HP-UX
1823727 Members
3124 Online
109664 Solutions
New Discussion юеВ

An academic question about scsi controllers and double redundancy

 
SOLVED
Go to solution
Neil Harris
Regular Advisor

An academic question about scsi controllers and double redundancy

We have an L class server, which has two fully loaded FC10's configured as a primary and a mirror. Each FC10 has two FC controller cards a primary path and a secondary path going to each i.e. 4 controllers total. We therefor have double redundancy in the drives. (I wont go into one suppliers suggested configuration for this arrangement when we went out to tender). 11.0 seems to understand this arrangement very well as I befor we took this sytem live as a production machine disconected the fibre optic patch cables without apparent effect on the system other than a warning message on the console. I am about to reconfigure, ok play with:), our aging K220 and make some much needed changes including an upgrade to HP-UX 11.0. The K220 has two FWD SCSI controllers each connected to a Jamaica, again each of which is fully loaded and configured as a mirrored pair. The idea has occured to me can a similar arrangement be achieved with the FWD controllers as has been done above with the FC10's i.e. two controllers on each Jamaica, one at each end of the SCSI drive train. I have had some thoughts ( delusions?) on the idea, such that a different scsi id would be required on each of the FWD cards say 6 and 7, and that each card would have to be configured to see the drives. As I have no desire to be the hapless pioneer with a smoking ruin and a P45 at the end of the week; I am for obvious reasons reluctant to try this in practice without the definite knowledge that it has been successfully achieved elsewhere. Has anyone done this or does anyone know of a white paper that explores this configuration with its benifits and pitfalls. Indeed if it has not been done is it at all feasible. I do know that the concept of a SCSI drive train connected to two controllers on separate machines has been achieved with other operating systems. I perhaps naively can???t see why it can???t be done on one machine. Thus I am seeking comments, ideas and insults about principal. Though maybe just a visit from the men in white coats of whom we have plenty as I work in a hospital is the required solution to my thoughts:)
An inveterate hacker
4 REPLIES 4
A. Clay Stephenson
Acclaimed Contributor

Re: An academic question about scsi controllers and double redundancy

Hi Neil:

Yes that can be done and, in fact, that is very typical in an MC/Service Guard environment. One controller is set to 7 and the other is typically set to 6. In fact, with a 'V' cable, one can actually add a 3rd host and controller set at 5. To make component servicing more feasible, one typically removes the terminators from the controllers themselves and replaces them with 'in-line' terminators. This allows one to replace controllers without killing the SCSI bus.

Regards, Clay
If it ain't broke, I can fix that.
Neil Harris
Regular Advisor

Re: An academic question about scsi controllers and double redundancy

Hi Clay,

Thank you for this. OK so I can see how the hardware is now done. What about getting the operating system to understand that it has two controllers on each jamaica. Or is it a case of having to have MC/Serviceguard to do it which would in this case kill the idea dead. I can get two more FWD controllers for the K class cheaply what I could not get is the MC/serviceguard licence :)
An inveterate hacker
Patrick Wallek
Honored Contributor
Solution

Re: An academic question about scsi controllers and double redundancy

There should not be anything in particular that you need to do in order for the system to see the dual paths to the disks.

If you were to set up a VG, via SAM, on the disks once the dual pathing is set up, SAM should ask if you want to set up the dual path to the disk.

If you set up the VG manually, then once the VG is set up, you do a vgextend with the alternate path and everything should OK.
Tim D Fulford
Honored Contributor

Re: An academic question about scsi controllers and double redundancy

Neil,

as you are going to connect to two SCSI cards, make sure you do not connect to your internal card! This will mean that you will see your root disks down two controllers!

Also take out the internal terminators & use self terminating cables. If you are in a ServiceGuard pair this will mean the SCSI chain will not be broken if the second m/c is down. If you are not in a SG cluster it would be a good idea as if one card goes bang it will not effect the other.

IMHO you do not need this level of redundancy (if I have read your question right). You have a primary & mirror disk on two different controllers, so 2 cards are required. Your method required 4 cards.
That said, if you have a high throughput rate your method will be better for high bandwidth apps.

Tim
-