Server Placement for VMware Cluster

[Daniel Andryszak]

We are deploying a new 4 node cluster (B200 M4s).

Currently have 2 x 5108 chassis.

What is the downside (if any) if I put 2 nodes in Chassis 1 and 2 nodes in Chassis 2 as opposed to all nodes in Chassis 1 or 2?

I know that communications will have to traverse the fabric interconnects if we split the cluster across different chassis, but it would provide some availability if there was a chassis failure (not very likely, but trying to evaluate all scenarios).

Thanks,

Dan

[Kirk J]

HI Daniel.

One thing to keep in mind is that that IOMs aren't doing 'switching'.  That is done at the FI level.

Traffic from Mac A on Host A still has to go up via IOMs to FIs to get switched to Host B on FI A (or B) regardless of whether there are multiple Chassis involved.  Additional Chassis generally don't involve more latency, where as traffic between two different sets of UCSM domains always requires traffic to traverse northbound switches.

Ideally you want to engineer your vnic arrangement so that your heaviest traffic can be switched on the same FIs if possible.  Hosts with OS IPs associated to vnics on the same side (a|b) and in same vlan, will get switched on same FI.  Traffic heading between different vlans or between vnics pinned to different sides (a|b) will have to take additional hops to northbound switches)

Additional best practice items include using adequate IOM to FI links that use link aggregation/grouping global policy, jumbo frames (requires QOS change in ucsm), OS level Nic Teaming, OS level multipathing for storage, SAN port-channels between FI and storage switches for FC traffic, eth port channels between FIs and VPC enabled northbound switches.

Thanks,

Kirk...