FC & MULTIPATH !? (any hope?)
Hi,
is there any hope of working combination of MULTIPATH with FC !?
At the moment I am using raid option multipath but it's one way street, when one FC connection dies it successfully switches onto another FC connection but when that second dies aswell, mount point is in a limbo, no switching back to first FC connection.
Any other solutions, patches ?!
Signature · 3 lines
Mario Mikočević (Mozgy) mozgy at hinet dot hr My favourite FUBAR ...
Quoted 1 line · Mario Mikocevic
is there any hope of working combination of MULTIPATH with FC !?
Yes. QLogic's newest 2200 HBA can do that. I don't know whether that is a possible solution for your problem though.
Sincerely,
Lars Marowsky-Brée <[email protected]>
Signature · 1 line
Perfection is our goal, excellence will be tolerated. -- J. Yahl
Quoted 9 lines · Lars Marowsky-Bree
Quoted 2 lines · Mario Mikocevic
is there any hope of working combination of MULTIPATH with FC !?
Yes. QLogic's newest 2200 HBA can do that. I don't know whether that
is a possible solution for your problem though.
The real question is when will Linux fully support multipath at the CAM/block layer? I'd really like to be able to throw, say, four HBAs into my system and have it use all of them simultaneously and not have to spend all sorts of time trying to set up all 200+ LUNs that I have available to me by hand. Think 200 LUNs presented * 4 HBAs * 4 paths at the controller end per LUN. That's quite a bit of setup to me.
- Pete
Quoted 11 lines · Mario Mikocevic
Hi,
is there any hope of working combination of MULTIPATH with FC !?
At the moment I am using raid option multipath but it's one way street, when one FC connection dies it successfully switches onto another FC connection but when that second dies aswell, mount point is in a limbo, no switching back to first FC connection.
Any other solutions, patches ?!
After analysing the current code in md/multipath and discussing it with other here who have experience with multipath support I have come up with the following approach.
When a path fails and there are no more good paths, an attempt will be made complete the operation using each previously failed path untill the operation succeds, or all know paths have been tried. If the operation succeds, that path will be marked good.
When a operation is attempted and no good paths exist, the operation will be attempted on each know path until success or all know paths are tried.
Probable enhancements to this would include, provideing a method to mark a path to not attempt this crude form of auto recovery and a way to mark a failed path as good. Finally a device wide flag to disable auto-recovery.
A disadvantage to this approach is that it would potentially, multiply the amount or time it takes to ultimately fail the attempt, by the number of paths. This would seem to be acceptable since the alternative is to fail the operation when a good route might exist.
I would appreciate any thoughts, flames, or suggestions.
Quoted 11 lines · Brian Beattie
Probable enhancements to this would include, provideing a method to mark a path to not attempt this crude form of auto recovery and a way to mark a failed path as good. Finally a device wide flag to disable auto-recovery.
A disadvantage to this approach is that it would potentially, multiply the amount or time it takes to ultimately fail the attempt, by the number of paths. This would seem to be acceptable since the alternative is to fail the operation when a good route might exist.
I would appreciate any thoughts, flames, or suggestions.
Combined with the enhancements this makes a lot of sense.
The enhancements are very much required, especially the way to mark a path as good again manually.
I would also liks easily parseable /proc file to query the status of a multi-path device, including all paths associated with it.
Sincerely,
Lars Marowsky-Brée <[email protected]>
Signature · 1 line
Perfection is our goal, excellence will be tolerated. -- J. Yahl
Quoted 7 lines · Lars Marowsky-Bree
Quoted 4 lines · Brian Beattie
Probable enhancements to this would include, provideing a method to mark a path to not attempt this crude form of auto recovery and a way to mark a failed path as good. Finally a device wide flag to disable auto-recovery.
...
Quoted 10 lines
Combined with the enhancements this makes a lot of sense.
The enhancements are very much required, especially the way to mark a path as good again manually.
I would also liks easily parseable /proc file to query the status of a multi-path device, including all paths associated with it.
I have some stuff that uses sysctl, which shows in /proc/sys/... that I was planning to use.