According to the documentation, a management group should have an odd number of managers greater than 1. I have a four node SAN consisting of P4500 G2s. I plan on having two clusters with two nodes each in this management group, i.e.:
-Managent_Group1
-Cluster1
-Node1
-Node2
-Cluster2
-Node3
-Node4
Are there any issues running standard managers on Node1
, Node2
, and Node3
? After reading the documentation, I'm still unclear about whether or not cluster membership matters in quorum consistency, or if they don't matter at all.
HP advised me to have 4 regular managers (1 on each node) and a Failover Manager (FOM) configured for quorum consistency in a management group that has two clusters, with two nodes for each cluster.
For example, let's say you need to update the firmware on Node 3 in Cluster 2. If you have Node 4 running a regular manager, and a FOM in place, you can bring Node 3 down without loosing any quorum consistency. When Node 3 comes back online, it will take approximately 15 minutes for it to re-sync its data with the cluster.
Another alternative is to use a "Virtual Manager". From HP:
A Virtual Manager is added to a management group, as shown in Figure 52 on page 151, but is not started on a storage system until a failure in the system causes a loss of quorum. Unlike the Failover Manager, which is always running, the Virtual Manager must be started manually on a storage system after quorum is lost. It is designed to be used in two-system or two-site system configurations which are at risk for a loss of quorum
As a small side note, you do not want to use a FOM and a Virtual Manager together, its either one or the other.....
If you have an HP Passport, you can go to this link to download a ESX, Hyper-V, or standalone version of the FOM:
https://h20392.www2.hp.com/portal/swdepot/displayProductInfo.do?productNumber=P4000SW
Scroll to the bottom of the page, click "Receive for free" and sign in with your HP Passport.
References:
HP StorageWorks P4000 SAN Solution user guide
http://h20000.www2.hp.com/bc/docs/support/SupportManual/c02008676/c02008676.pdf
After testing and re-reading the HP documentation and consulting the Best Practices Summary of the CMC, it appears that running standard managers on 3 of the 4 nodes is the "correct" way to do this.
Since quorums are per management group and not per cluster, the fact that the four storage nodes are in two separate clusters is not relevant in this case.