About Path Selection

Pillar Axiom Path Manager (APM) selects the best paths to access Pillar Axiom LUNs.

Path selection is based on three factors:
  • Path optimization state

  • Path performance

  • Path availability

Path optimization is determined by the Slammer control unit (CU) that the path uses. An optimized path is a path that connects through the Slammer CU on which the LUN is currently resident, or homed. A non-optimized path is a path that connects through the alternate Slammer CU. The LUN can be accessed through the alternate CU, but the LUN is not resident on the alternate CU. Optimized paths are always preferred, but if an optimized path is not available, the non-optimized path can be used temporarily.

Path performance is determined by how quickly and reliably a path transfers I/O traffic to and from a LUN. Generally, FC paths perform better than iSCSI paths, so FC paths are preferred over iSCSI paths.

Note: In this discussion, FC includes FC HBA as well as FCoE CNAs.

Path availability is determined by the ability of the path to transfer I/O traffic. An available path is fully functional but if the path stops working, the path is considered unavailable.

These factors determine how the paths to each Pillar Axiom LUN are divided into groups.

APM groups the paths in the following order of preference:
  • First, FC optimized
  • Next, iSCSI optimized
  • Next, FC non-optimized
  • Finally, iSCSI non-optimized

For each LUN, the currently configured load balancing algorithm is used to select paths from the most preferred group that has paths available. Only paths from a single group are used at any one time.

When an active path fails, I/O traffic is transferred to a different path, and I/O performance is reduced for a short time while the operating system recognizes the failure and makes the path transfer. After the path transfer is complete, I/O performance improves.

If the failing path is optimized and the new path is non-optimized, I/O performance might continue to be reduced after path transfer because a lower-performance path is in use. Within a few minutes of traffic being transferred to a non-optimized path, the Pillar Axiom system attempts move the LUN to the appropriate Slammer CU to make the path optimized. After transfer to an optimized path succeeds, I/O performance improves.