FYI – the link to the Bw-tree paper isn’t particular visible, so click here for more info: Bw-tree paper from Microsoft Research.

On this one, I’d add that the closer cache levels are becoming increasingly more important in two major ways: NUMA and the emergence of technologies such as Bw-trees.

With both, if we can keep the workload on the same processor group into which the data were loaded originally (e.g. cache pages), we are guaranteed the fastest possible access. Causing different data sets to be moveded in and out of the cache or causing “foreign” data transfers from one CPU group to another all limit us.

The NUMA setup and CPU affinity options for SQL are really key here. Also, with both VMware and Hyper-V, you can now give a VM NUMA awareness and affinity. All important things to consider!

Not sure if CPU-Z gives you the same output somewhere, but coreinfo.exe from sysinternals definitely helps in identifying the NUMA-ness of an architecture, particularly when run from inside a VM.