On Fri, 2007-08-03 at 21:12 -0700, Sajid Zia wrote:
> I have a system booted up and running with all the networking
> user/kernel code working fine.
That just means you haven't found the deadlock scenarios in testing yet.
User-space components of IP are mostly involved in failover and recovery
and I'd expect things to appear to work most of the time, but working
most of the time doesn't mean that iSCSI root is done.
> All the code changes are confined to iscsi driver and there are no
> interface/architectural changes.
Have you tried the following?
1) IPMP
enable ipmp on the client.
test an ipmp failover. test it again while the system paging heavily
and some of in.mpathd is swapped out to an iSCSI lun.
2) Put a router between client and server. (iSCSI is scsi over IP, not
scsi-over-ethernet...)
Can you recover from a "route flush" ?
2.A) Router discovery
Configure a client to use router discovery to find its default route.
put a router between client and server. boot the client. crash the
router. bring up a different router. does the client fail over to the
alternate path all the time when the system is under heavy memory
pressure?)
3) VNI-based dynamic routing with OSPF.
(note: I'm being nice. I'm not asking about IPsec/IKE between client
and server).
- Bill
