You guys gave me some great thoughts in my last posting in which I discussed my AX4/iSCSI highly available architecture.  In this posting, I will continue the thread and give you a look at what the Westminster College architecture will look like in a few weeks.  Some of this information is based on ideas provided in your comments.  Although I’ve had the basic architectural diagram in mind for quite some time, your comments have helped to refine it. 

Let’s start with a look at how VMware ESX will fit into our architecture.

This diagram is very similar to the one from the previous posting with one change.  At the bottom of the diagram, I show an ESX cluster, fully VMotion-enabled.  Each ESX server has multiple connections to the iSCSI storage network as well as to the primary network the users use to connect to the ESX servers.  Under this scenario, we will achieve a high level of service availability for all of the servers running on the individual ESX servers.  We’ll get to a highly available architecture with our SQL servers — and well as some other non-ESX services — through clustering, which will also entail a setup like the one above.

The next scenario expands on the scenario shown in the previous discussion.

I mentioned in that posting that, for simplicity’s sake, I wouldn’t show the connections to our core switch — an HP Procurve 5412zl.  One of the comments on the previous posting recommended that we use the HP 5412zl for our primary iSCSI VLAN rather than our Dell blade-based M6220 switch.  Under this scenario, we would bond together the four uplink ports on the M6220 to the 5412zl.  The only downside to this scenario is that all iSCSI traffic from our blade chassis will have to traverse both the M6220 and the 5412zl.  An alternative would be to use one uplink port on each of the M6220’s to connect to the AX4 and connect the other pair of iSCSI ports on the AX4 to the 5412zl.  Doing this, we would have only two ports available to bond together from the M6220s to the 5412zl.  We will test both scenarios, but I suspect that we will go with the alternative scenario I just described as it provides a higher level of redundancy.

I look very forward to your comments and suggestions.