OVC communication over the federation network

[jlangmead]

Hi

I'm trying to get a better understanding of how, when and if OVCs need to communciate over the federation network. With an environment using a single Federation (single VCSA) and multiple Simplivity clusters do OVCs that are in different clusters still need to communicate together over the federation network - I understand that federation traffic doesn't extend beyond the cluster?? 

If OVCs are in the same federation but can't communicate over the Federation network, am I correct in thinking they then attempt to communicate over the management network?

Therefore, should Federation networks be segragated to only include other OVCs that need to communicate together? eg in the same cluster. If I have four clusters in a single federation  should the federation network be configured (vlan'd) so the cluster OVCs will not have any cross contamination of federation taffic between other cluster's OVCs?

thanks

[pterchan]

the OVCs send and receive VM data replica data via Federation network

and when Fed network is not available, that traffic falls back to Mgmt network

[jlangmead]

Thanks. By ‘replica’ data do you only mean the second copy of the VM within the same cluster?
Therefore federation traffic never passes between different clusters?

Thanks again

[jlangmead]

I've found a really good explanation which answers most of my questions here -

Solved: official definition about "Federation" in SimpliVi... - Hewlett Packard Enterprise Community (hpe.com)

I think I was falling into the cataegory of confusing the SimpliVity 'Federation' with the 'Federation Network'

But it does still raise a couple of questions for me - Below it says that clusters will communicate over the Federation Network if they can but usually clusters talk over the management network. Would it not be better to ensure they can talk over the Federation network (if feasible without needing routing) if doing lots of backup replication etc - or is it still better to have all the Federation traffic VLAN'd to individual clusters regardless?

Secondly it says that Federation and Storage traffic is logically separated although they use the same physical network - is it ever worth physcially separating this traffic if, for example, each node has 4 x 10Gbe ports to use 2 ports for storage and 2 ports for Federation?

thanks

Extract from the link is below:

"The Federation database is replicated across all nodes in the Federation.

It’s easy to confuse the SimpliVity Federation with the Federation Network ( SVT_FedPortGroup), but they are not at all the same thing.

The Federation Network is used within a cluster for Federation-related traffic between nodes in the cluster. It is logically separated from the cluster Storage traffic, even though they run over the same physical network.

Replication of information between clusters is usually (and assumed to be) over the OVC Management Network (however, if separate clusters can communicate over the Federation Network, they will - but it is not required). This includes both Federation-related information (installation of a new node, for example), and Data (replicating a backup).

The SimpliVity OVC is responsible for replicating Federation Information both within a cluster and between clusters."

[gustenar]

Hello @jlangmed, 

Please see this technical article related to HPE Simplivity networking best practices, I think it addresses the questions you have:

https://psnow.ext.hpe.com/doc/a50000783enw?jumpid=in_lit-psnow-red

Hope it helps.

[Moderator edit: Updated the broken link.] 

[Convergo]

@gustenar When running 1 vCenter for 1 Federation, for multiple SimpliVity clusters all SimpliVity nodes in the federation are communicating with each other (at least the leaders per cluster). 

Having 1 cluster within the Federation that is unable to communicate with other nodes in the Federation, what would be the disadvantages for this cluster ?

[gustenar]

The cluster must be able to talk at least to one another cluster (a central hub) if you are trying to implement a partial mesh. Not being able to communicate at all could bring issues managing that cluster, backing up VMs, unable to update transactions, etc. 

[s-]

Hi @jlangmead 

As per Simplivity design, All the nodes first try to communicate its peer over Federation network if that fails it will fail back to Management network. It also apples to Cluster-to-Cluster communication since Federation network is not routable hence it will fail back to Management Network. If both the clusters are at same site, configured with same IP range and connected via L2 switch, in such scenario Cluster-to-Cluster can pass traffic over Federation network .

However, It is best practice keep the Federation traffic within the cluster, Federation & storage traffic should be isolated using VLAN and let the cluster to cluster happen over Management network.

It is ideal configuration(1 federation and 1 vCenter) however you may have multiple clusters under 1 federation and those clusters can be managed by multiple vCenters which are in link mode.

Example:

Prod and DR are two clusters under a Federation. Both are being managed by individual Vcentes which are configured in link mode. If both sites are at same location and connected via L2 switch , both clusters will be able to talk over Federation network. If both sites are connected via WAN/VPN, it will communicate over Management network.

If you have single cluster under federation, all the nodes must talk to each other over Federation network else you may start experiencing issue such as VM HA sync issue, Backup are in queued/failed state, connectivity alarm on Vcenter etc.

Hope it helps.