Brace yourselves, vSphere 8 is coming....

by Phil Snowdon

It that time of year, as NZ heads into spring and the northern hemisphere is heading into winter that VMware Explorer comes round again with the expected list of releases.  The primary one being vSphere 8 announcement

There are a few things this year to note

Tanzu/Kubernetes

Dilbert is always appropriate

This release introduces support for highly available, multi-AZ deployments in VMware Tanzu Standard, and subsequent deployments of Kubernetes clusters. This allows Kubernetes clusters, including the Supervisor Cluster, to be deployed across different vSphere clusters in your infrastructure, providing high availability in the event of site failures.

There are significant enhancements to the API to support this, plus new API for  Kebernetes clusters which aligns with ClusterAPI.  This is the first step on their journey to unify all Tanzu Kubernetes offerings on the vSphere platform, as it means you can now use the same tanzu CLI commands to build Kubernetes clusters using the TKG Service on the Supervisor cluster as an endpoint.

There is also a new Harbor Image Registry Service coming in this release, replacing the earlier embedded Harbor Image Registry. This will allow vSphere with Tanzu customers to use more up-to-date Harbor Registry features.

For more details see CormacHogan.

DPU's (Data Processing Units)

This release delivers on Project Monterey which was announced a while back.  vSphere 8 deploys the vSphere Distributed Services Engine which leverages DPUs, or SmartNICs.  These are basically NICs with a lot of offload features - not just the TCP offload engines, or iSCSI specific HBA cards of old.  For more info on SmartNICs see Corey's blog.

These devices enable you to push more packets (acceleration) and provide the ability to run services on the cards.   The potential here is not just moving more data, but also offloading NSX services from the CPU to DPU, we may be able to provide more security isolation between infrastructure services like NSX and VM workloads as it is shifted off the CPU, but also free up some  CPU cycles.  I read one article that said "up to 20%".  That's a bold claim.

One of the DPUs available is the AMD Pensando or NVIDIA Bluefield and this has got me wondering.  New Tech is cool,  but is it worth it?  I mean these accelerators are likely not cheap, is the potential of 20% extra CPU worth the cost? or should I just buy more cores?

We have 'lost' a lot of CPU power over the last couple of years with all the Spectre and Meltdown speculative execution mitigations.  And indeed we had already had to boost cores to cope with this.   We used to have 18core CPU as standard, but have shifted to 24core to be able to effectively utilise the amount of memory we have on each of our ESXi nodes.  That's about 30% more.

Let do a quick check (using prices from Intel Ark).  

  • Intel 6240 is around US$2620
  • Intel 6248R is around US$2890

That's 'only' about US$250 for a 30% uplift in performance.  Let's say US$1000 per host for a dual CPU setup.

Now I may be a bit cynical, and I don't have any pricing for these DPU to hand, so I remain to be convinced whether this will be a useful/cost effective piece of kit.   I'm happy to be proven wrong, but for the company and for our customers we need to ensure that we are making sound value decisions and not just jumping on the next technological bandwagon.