1

I'm considering enabling load balancing on our data gateway clusters for Power BI, where we have two servers in a cluster, and the secondary is configured as a failover for the primary (only operational when the primary is offline). I was wondering how likely in the real world it is that using load balancing to push traffic through both servers would end up reducing high availability compared to only using the one, considering the following scenario:

  • Assume a constant stream of traffic generated by application
  • While there is no load balancing, one server does all the work and defines the capacity limits for the operation. If it fails, the other server picks up the work and because of identical specs, can handle the traffic as expected. Monitoring is configured so that an alert is triggered when either server hits 80% capacity, so scaling can take place before bottlenecks are hit.
  • Load balancing is switched on, both servers are now processing traffic, and traffic increases to >50% capacity on both servers. One fails, and now the remaining server has to process >100% of its capacity, resulting in request delays or failures. At no point with either server were monitoring alerts triggered.

I suppose I could rephrase my question as 'does using load balancing to handle increased traffic instead of scaling the servers up reduce the redundancy required for high-availability?'

Improve this question

1 Answer 1

Reset to default
0

Load balancing doesn't decrease availability; failing to measure load does. In all cases, you should have load monitoring set to thresholds that will allow for a single server failure.

You have good monitoring in your first scenario but for some reason you removed it in your second scenario. Rather, you're not monitoring the whole system. Measure load at the load balancer.

Improve this answer
2
  • Thanks, that confirms my suspicions, that with load balancing enabled I would want to build in resilience for the eventuality we're down to one server. So in the second scenario, we'd want alerts to trigger at 40% capacity
    – JTDotNet
    Jun 10, 2022 at 16:11
  • Well kind of. Monitor the whole system, including your LB not just the servers.
    – mfinni
    Jun 10, 2022 at 17:35

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .