Originally from the User Slack
@J: How should I think about connections and cpus/ vcpus when working in K8s. 3 per Scylla cpu. In a 3 node cluster each with 16 vcpus (K8s). Should that be 3x16 or 3x8 for the logical core?
@avi: I don’t understand the question. 3x16 what?
@J: Number of connections. Your best practices say number of connections is 1-3 per Scylla cpu per node. But I wanted to check this is per vcpu or per logical core.
>> As a rule of thumb, for Scylla’s best performance, each client needs at least 1-3 connections per Scylla core. For example, in a cluster with three nodes, each node with 16 cores, each client application should open 32 (2x16) connections to each Scylla node.
@avi: In general the drivers we provide take care of this and generate one connection per shard
vcpu == logical core
(physical core == 2vcpu == 2lcore on x86)
@J: Thanks Avi