I’ll be going through iSLB, explaining what it is, and showing how to configure it. A full template is at the bottom of this post.
Part 1 of the series, “Configuring iSCSI for CCIE DC” can be found here.
What is iSLB?
iSLB is iSCSI Server Load-Balancing. It is iSCSI, so don’t confuse it as some other protocol, think of it as iSCSIv2. iSLB introduces a few new features to iSCSI:
– Load-balancing between MDS’s (or ports on the same MDS)
– Cisco Fabric Services (CFS) Distribution
– iSLB Initiators (with Automatic Zone creation)
Load-balancing
iSLB uses VRRP between two MDS switches for high availability and load-balancing. With VRRP you have a master and backup virtual gateway. Typically all traffic is sent to the master active gateway. So how does load-balancing work? A pair of MDSs will run CFS to keep track of an iSLB VRRP table. This table records the current load for each Initiator-to-MDS pair. When an initiator request comes into the VRRP master switch, the table is checked to see the current load on each MDS. The master will take the initiator and create a session if it’s load is lower than the backup switch. If it’s current load is higher, the master switch sends an ICMP redirect back to the initiator and a new session is built to the direct IP of the backup MDS switch. The default weight (load) for each initiator is 1000. This, of course, can be changed to influence path selection.
Although not visible initially, the master of VRRP starts automatically with more load since it has more responsibility. This means that the first session is always going to be redirected and load-balanced to the backup MDS switch. All sessions afterwards will be load-balanced based on load reported in the table.
As an example, say we have 3 initiators. Initiator 1 has a default metric of 1000, Initiator 2 has a configured metric of 900, and Initiator 3 has a default metric of 1000.
(more…)