I'm seeking clarification on the NUMA architecture in SQL Server. Is it accurate that the default configuration utilizes round robin to select the next NUMA node for new connections to functioning ports, rather than the next core or scheduler? With connection pooling in our application, do connections get reused? If the load is low, leading to predominantly serial access, could this result in all traffic being directed to NUMA node 0 until there is increased parallel usage? Could this lead to much of the buffer pool residing in Numa node 0?

Understanding NUMA in SQL Server

Henrik Alstad 0

Apr 16, 2025, 5:57 PM

I'm seeking clarification on the NUMA architecture in SQL Server.

Is it accurate that the default configuration utilizes round robin to select the next NUMA node for new connections to functioning ports, rather than the next core or scheduler?
With connection pooling in our application, do connections get reused? If the load is low, leading to predominantly serial access, could this result in all traffic being directed to NUMA node 0 until there is increased parallel usage? Could this lead to much of the buffer pool residing in Numa node 0?