Disclaimer: I recently attended VMworld 2014 – SF.  My flights and accommodation were paid for by myself, however VMware provided me with a free pass to the conference and various bits of swag. There is no requirement for me to blog about any of the content presented and I am not compensated in any way for my time at the event.  Some materials presented were discussed under NDA and don’t form part of my blog posts, but could influence future discussions.

BCO2701 – vSphere HA Best Practices and FT Tech Preview

If you’ve read my other posts from this week it’s probably pretty clear  that I should give up the charade of not providing a transcript of the session. So let’s just call them notes and not talk about it any more, yeah?

This session was presented by:

• GS Khalsa – VMware Senior Technical Marketing Manager
• Manoj Krishnan – VMware Engineer – vSphere HA

Agenda:

• What’s New
• Failure Events
• Best Practices (Network, HA and VSAN, Host Isolation Response and Admission Control)
• Tech Previews

What’s New
In vSphere 5.5

• Protection for VSAN VMs (major)
• AppHA Integration
• VM-VM Anti-affinity rule

What is HA?

• Minimises downtime
• Auto VM recovery in minutes
• Protects against 3 types of failures:

– Infrastructure – host, vm
– Connectivity – host network iso, DS incurs PDL
– App – GuestOS hangs / crashes, app hangs / crashes

• Cluster with a master and multiple slaves
• Heartbeats via network and storage
•  HA network can be management network or VSAN network (if using VSAN)

Failure Events
Once the HA agents are up and running, they go into election mode.

What if a slave dies? Master declares and restarts affected VMs on another host. If the Master fails, slaves go into election mode and restart VMs. Note that vCenter only talks to one master, so if you have a network partition, this can impact where VMs will startup in a failure event.

Host Isolation
The host declares itself isolated if it can’t talk to Master via network. Lets the master know via the datastore heartbeat

Best Practices

Network

• Redundant HA network
• Fewest hops possible
• Consistent portgroup names, network labels
• Route based on originating port ID
• Failback policy = no
• Enable portfast, Edge, etc
•  Same MTU Size
•  Disable host monitoring during NW maint
•  vmknics on separate subnets

Storage

• heartbeats – all hosts should see the same datastores
• choose a heartbeat datastore that is fault isolated from ha network and resilient to failures
• override auto-selected options as required

HA and VSAN

• Heartbeat datastore is not necessary for VSAN cluster
• Add non-VSAN datastore to cluster hosts if VM MAC address collisions on the VM network are a significant concern
• Choose a datastore that is fault isolated from the VSAN network
• Set the isolation address with das.isolationAddressX
• Configure HA to not use the default management network gateway
• If VSAN is non-routable, use a pingable isolation address somewhere else in the network

Host Isolation Response

To delay response in 5.1+, use das.config.fdm.isolationPolicyDelaySec

There are a few options:

• Leave Powered On (default with 5.1+)
• Shutdown (default with 4.x)
• Power Off

Which to use? It depends.

• What type of VM?
• Will host likely retain access to storage?
• Will VM likely retain access to VM?

Admission Control
If you turn admission control off, you’re limiting what HA can do for you.

You can reserve resources in case of host failures

• Ensures resources are available
• No guarantee VMs will be happy after a failure though
• Working on closing this gap

How to do it?

• Select the appropriate policy
• Enable DRS
• Simulate failures to test by using maintenance mode and the Impact Assessment fling

Make adjustments if:

• VMs are not restarted
• Desired performance is not realised

Reducing Impediments

• Maximise utility by enabling DRS in automatic mode
• Maximise hosts a given VM can run on
• Ensure sufficient resources
• Ensure critical VMs get what they need (HA restart priority – this doesn’t guarantee VM restart order)

There are 3 Admission Control Policies
1. Percentage of Cluster Resources

• Often best choice
• Maximises cluster resource use
• Use when reservation varies considerably

2. Number of hosts policy (slot algorithm)

• Maximises chances of restarting VMs with reservations
• Avoids fragmentation
• Often conservative – if this is a concern, use dedicated failover host policy.

3. Dedicated Failover Host(s)

• Best if VMs have large resource reservations
• Impact to VMs on other hosts minimised
• Restart time can be longer
• Failover hosts are idle prior to failure
• If you use this, select largest hosts to be failover hosts.

Tech Previews – FT and HA
FT for zero-RPO, zero-RTO (The version in the beta)

• 64GB RAM per protected VM
• 4 vCPUs
• Uses individual, separate storage (no longer shared) for Primary and Secondary VM
• Complete “re-imagining” of how FT functioned.
• Lockstep was hard to scale, now uses “incredibly fast check-pointing”.

HA – VM Component Protection

• Problem: Host has storage connectivity issue – APD or PDL
• Difficult to manage VMs
• Approach: Move VMs to a healthier host
• You can choose how you want to protect and under what circumstances (sounds a bit like Australia’s border protection policy – sigh).

Admission Control Fling – vRAS

• Coming in the next month or two
• vSphere Resource and Availability Service
• Assess impact of host failures and VM migration on resources using DRS dump files
• Provides sample what-if scenarios

And that’s it. Top session, very informative. 4 stars.