HCX 4.10 has just dropped and with it comes a gem of a new feature ‘HCX Assisted vMotion’ or HAV.
Check out the Whats New in HCX 4.10 Release Notes – https://docs.vmware.com/en/VMware-HCX/4.10/rn/vmware-hcx-410-release-notes/index.html
So, don’t we already have Replication Assisted vMotion (RAV)? Well yes, RAV is a great tool but is limited by the throughput of the IX appliances.
IX appliances can pass around 1.7 to 2Gbps at a time, this is why some people now scale out and have a multi service mesh model for a single source cluster, please refer to a previous blog on Multi-Mesh.
So what is the difference between RAV and HAV?
HAV cuts out the middleman, HCX can now support line rate assisted vMotions direct from Host to Host rather than Host to IX to IX to Host.
Line rate means as fast as your physical NICs or Network IO Control (NIOC) will allow.
With customers running 10, 25 or even 100Gb NICs this dramatically improves throughput for HCX.
How does HAV differ to Cross vCenter vMotion?
Cross vCenter vMotion is a manual process in vCenter, there is now a UI to select multiple VMs but it lacks the ‘orchestration’ that HCX provides.
HCX acts as the orchestration engine for the HAV migrations, initially allowing for around 50 migrations per HCX manager and adding a touch of magic.
Note – As of the time of writing I do not believe HAV supports a scheduled switchover process, maybe a future enhancement.
How do I enable it?
HAV is another service that is enabled in a Compute Profile:
Note – If you are editing a Compute Profile and adding the HAV service, any Service Meshes using that Profile will also need to also be updated.
This is not picked up with a Resync, instead you have to edit the existing mesh and tick the new available service.
Shown below is the new HAV service icon in an updated service mesh:
RAV vs HAV
OK, so lets kick the tyres with HAV and run a direct comparison with RAV.
Shown below is a new Migration job summary for a test VM using ‘HCX Assisted vMotion’ from the Migration Profile
This will use the exact same settings, VM size etc that we will use for the RAV comparison.
The 2 x TEST VMs are actually clones so identical in all ways except for name.
Here are the RAV results:
As you can see, it took about 12 minutes end to end. The actually Data Sync Process was about 3 minutes in my lab. The other 9 minutes were delta syncs and all those other 60 tasks, now that seems to have increased since version 4.9.
So, drum roll, how did the HAV job do?
Well, off course it was quicker as we are going Host to Host.
Shown below was my first run:
In summary, about 3 minutes end to end, only 11 tasks in HCX Manager so a lot less back and forth. The data sync task was seconds rather than minutes.
Well, thats is impressive, so much so, I ran 2 x HAV jobs in a mobility group to see:
OK, thats 2 x VMs in just under 3 minutes, this would have been about 16 minutes with RAV due to post sync switchover jobs.
Summary
So would I throw RAV out with the bath water? The answer is no.
Each migration profile has its benefits, RAV will allow for more concurrent migrations, HAV is limited by traditional svMotion vCenter limits.
Additionally, some customers may not want to run multiple vMotion jobs full bore as they could impact network or storage resources.
I think you should give it a go, maybe consider it for those bigger VMs that can just migrate and not require a switchover window for testing.
For more detail on HCX Assisted vMotion, please visit the VMware documentation here – https://docs.vmware.com/en/VMware-HCX/4.10/hcx-user-guide/GUID-FAAA06F2-74E0-4606-8CEA-D7040AC8EE44.html