Littleboy,

The SDC will use all available paths to the SDS IPs it's configured to access.  The list of those IPs can be printed by running /opt/emc/scaleio/sdc/bin/drv_cfg --query_tgts.

Connectivity to those SDS IPs is determined by the network configuration.

If you have only one NIC on node5 and it's on the 192.168.10.x subnet, then the OS can only send packets to the SDS IPs on the 192.168.10.x subnet.  If you add another NIC and can now send packets to the 192.168.101.x subnet, then the SDC will be able to "talk" to the SDSes on that subnet as well.

Think of the multipathing here as you would a "Symmetrix-Optimized" pathing policy in PowerPath.  The SDC will communicate on all available subnets as they are available.

Caveats:

1. This of course assumes that the network hardware does not provide routes between the subnets, which would be detrimental to performance.
2. OSes generally cannot route between two NICs on the same subnet. (Bonded interfaces being the exception. They are still not preferred, however, as ScaleIO cannot path between them.)  Node5 having two NICs on the 192.168.10.x subnet will still result in only one NIC being used at a time.  While such a configuration would indeed provide redundancy in case of failure, so would connectivity on a second subnet -and with the accompanying performance increase it affords.