Forum Discussion
davec_20224
Nimbostratus
NimbostratusNode or member version of load balancing method?
Platform: BIG-IP LTM 6400 Active/Standby HA pair
SW: 10.2.0
I understand the difference between the node and member versions of load balancing methods. The part I'm missing is under what scenarios would you choose the node version and under what scenarios would you choose pool member. In the case of each node only belonging to a single pool, I don't think it would matter. But if a node belongs to multiple pools, I imagine the deciding factor would be whether each pool places similar resource demands on the node (e.g., node belongs to two HTTP pools) or each pool has different demands (e.g., node belongs to a low demanding POP pool & a high demanding IMAP pool). For these different scenarios, would node or member versions of the load balancing methods be the best choice assuming we're optimizing for response time to the user? Are there other factors I should consider when deciding between node and member versions of a load balancing method?
Thanks,
Dave C.
hoolioCirrostratus
Hi Dave,
I think your description is a good one. If a node is a part of multiple pools and the load of one pool's usage would impact other services on the same server, you'd want to consider using node based load balancing algorithms.
Aaron
davec_20224Aaron,
After discussing with a colleague, I think we came to the opposite conclusion. For the case of two services with different resource requirements, you would want to balance at the pool level so you get an even distribution of both services across the servers. For the case of two services with similar resource requirements, using node based balancing would be OK because even if it turns out that each server is handling requests for only one of the services, the load is evenly distributed.
Is there a flaw in this logic?
Thanks,
Dave C.- hoolioHi Dave,
My thinking is that if one service on the node requires a lot of resources, you wouldn't want the node to receive a lot of traffic on the second service. One way to avoid that would be to use a load balancing algorithm that takes into account the nodes membership in multiple pools.
If you were actually monitoring node resource utilization in a dynamic load balancing algorithm, this would be less relevant. But if you're just sending packets for multiple services that have varying resource requirements, I think it makes sense to use a node method instead of a pool member method.
Aaron