Any strong views, recommendations, on whether there are specific benefits of disabling Hyperthreading on Coherence Storage Nodes.
We have a grid of 21 IBM iDataPlex Blades, with 36GB RAM, and 12 physical CPU’s (24 HT CPU’s), each hosts eight 3GB heaps. Any benefit of turning HT off, and having more processor cache available to each thread without conflict, or is this all managed away under the hood.
There are ~300 Coherence caches, most around 100 MB, a few low GB’s, and one ~40GB.
Coherence version is 22.214.171.124
Disabling HT from CPU is not an always wise choice unless you are dealing with latency sensitive scenarios or is suffering from non-linear throughput. Is that your case?
If not, I would not recommend that. HT is a feature to increase the parallelism capacity, not necessarily, multi-threading. One thing is to process multiple instructions per time (parallelism) and another thing is to load-balance (also known as O.S preemption) between multiple threads. There are scenarios in Coherence when parallelism is important (serialization, locking) and scenarios when multi-threading is important (proxy-nodes, replication, death detection algorithms) so having multiple cores with HT enabled is a good thing.
You're seem to have a good hardware environment right there, not sure that disabling HT is a good idea.