SVC to PowerVM NPIV SAN Zoning Best Practices

From Ernie O. and Chuck L of IBM…

Please do not contact the author on this but raise any concerns on this paper through your IBM Storage Support channel.

                                                                                                                                      

I have always considered 8 NPIV paths/zones, each with a single initiator and single target, to be the best configuration for performance and availability.

Most vendors Path Control Modules recommend 8 paths for performance and, properly cabled, 8 paths also allows for a concurrent failure on one of the SAN Fabrics (assuming 2) as well as one of the VIO servers, (assuming 2).

However, the SVC using SDDPCM has 4 paths as it’s optimal number for performance.  If you use 8 paths, and LPM a partition, the SVC has to manage 16 paths during the LPM move and this has resulted in a failed LPM move, with data loss, resulting in a reload of the partition.

 

Click for detailed diagrams .pdf —–>  SVC zoning for PowerVM NPIV

 

Please do not contact the author on this but raise any concerns on this paper through your IBM Storage Support channel.

2 thoughts on “SVC to PowerVM NPIV SAN Zoning Best Practices

  1. Please do not contact the author on this but raise any concerns on this paper through your IBM Storage Support channel.

    It is understandable that many people have questions about this and have probably configured 8 paths back to the SVC, as that is typically best practices in most SAN storage pathing scenarios. This is why the post was published, to raise awareness.

    The data loss was experienced in a customer database LPM move by the SVC specialist that help develop this paper. Since the paper posted, other customers and IBM specialists have replied that they have had the same problem with failures during LPM with 8/16 paths to SVC LUNs. There was one person that said that they did not have a problem. Since the issue is suspected to be processing capacity of the SVC, it is understandable that results may vary based on conditions, size of the data, what is in-flight, number of LPM moves, etc….

    The IBM SVC Support group have been asked to make a support statement on this, but no such statement exists at this time.

Leave a Reply