While reading a paper, I came across the term multi-socket architecture - an extreme case of vertical scaling. Here’s a quick note about it.
This architecture enables multiple CPU sockets on a single motherboard, which means we can plug multiple physical processors on the same motherboard. This is well-suited for workloads demanding high parallelism, such as scientific computing and in-memory databases.
Each socket has direct access to its local memory and I/O, but inter-socket communication introduces latency. This is where NUMA-aware scheduling and memory allocation help avoid performance bottlenecks.
This setup is adopted when we do not want to manage several physical machines but still need more compute power from a single box.
Fascinating stuff.