does failover in gateways have anything to do with splitting up responsibilities

[db042190]

Hi, as we move toward fail over gateways, i heard something today that is a bit of a concern.

we are merging the tenant of our purchased company tenant with our own.

today i heard a peer suggest that as we introduce 2 failover machines that perhaps we split the conns between them also. 

Isnt the concept of failovers for the old model that redirects traffic one way if the heartbeat of the other cant be found?   And if traffic is really an issue, wouldnt we really want a separate gateway (perhaps with its own failovers, or instead more horsepower on the 1st)  if we really needed to separate duties down connection lines and there was no need for merges etc across those lines?

[GeraldGEmerick]

@db042190 A gateway cluster can consist of up to 10 gateway members (nodes). The primary gateway in the cluster handles requests by default, but if it becomes unavailable, the system automatically switches to another available gateway in the cluster. This ensures uninterrupted access to on-premises data.

For load balancing, traffic can be distributed across all active gateways in the cluster. By default, the selection of a gateway during load balancing is random, but administrators can configure settings to distribute the load more effectively. For instance, resource throttling can be applied to prevent overloading any single gateway member by setting limits on CPU and memory usage. You can find more information here: Manage on-premises data gateway high-availability clusters and load balancing | Microsoft Learn

[db042190]

thx Gerald.  Are you saying that trying to distribute to / along specific connections isnt going to happen?

[GeraldGEmerick]

@db042190 You could distribute to/along specific connections. That would require creating multiple gateway clusters and then you would only create certain connections on each cluster.

[R1k91 · Accepted solution]

when you setup connections you see gateway clusters not single machine.
therefore you're saying "use that cluster", that can be made up to 10 machines and the one is going to run the actual load is defined by this flag (always one of them and the others are DR or all the machines can get one request randomly).

 

in any case all the requests for that connection will hit the same gateway cluster (pool of machines).
if you want to separate workloads you need to install another gateway cluster (made of other machines) and setup other connections on it.