Adding nodes to existing P4500 G1

[Jim Silvia]

Right now we have one install with three P4500's each has twelve 300 GB, 15K SAS drives. Each is physically configured with two RAID5 sets. This leaves a reported 2.67 TB of usable space per node.

The P4300 G2, SAS Starter, uses eight 450 GB SAS 15K drives in i believe a single RAID5 set. This should leave about 3 TB of usable space per node.

Can i add the two new nodes to the existing cluster, or, am i shooting myself in the foot due to the difference in spindles and am i better off creating a new cluster for the new P4300 G2's?

Has anyone tried this? If it works it seems more cost effective than adding an additional P4500.

[Uwe Zessin]

You can add the nodes to the existing cluster, but due to how Network RAID works the node with the smallest capacity defines how much capacity is made available by the other nodes.

[Jim Silvia]

They are very close, so, i'm not concerned about losing a small amount of room.

Any thoughts on performance impact though?

[Uwe Zessin]

The existing volumes will be striped across more disks which should give better performance, but it really depends on how much I/O is added by additional hosts/applications/users after the expansion.

[kghammond]

The Lefthand suppot engineers have explained to us that when you cluster different nodes together, the cluster will dumb itself down to the slower and smaller node.

As I understood it, if one node has lower IOPS / throuhput, etc, the other nodes will be limited to the performance characteristics of the slower node.

In general your cluster will have more disks so it will be faster, but whichever nodes are faster, they will slow down to match the performance of the slower nodes.

I am not 100% sure this is all completely accurate or that we completely understood what the engineers were explaining. The Lefthand sales engineers did not address performance when discussing adding nodes to an existing cluster.

From this information, we decided when we got new nodes to put them in their own cluster instead of expanding an existing cluster.

Additionally, at least for larger enviornmnts, we are planning to never run a cluster with less than 3 identical nodes. We are having minor issues when a two node cluster is running very hot, and you need to do maintenance, if you take one node offline, VMware performance degrades very rapidly.