By the numbers

Workload mix

For each workload in the mix

1. Allocation/De-allocation and long-vs-short-lived behaviors

2. Access (and co-access) patterns in space and time

3. Page-level hotness/coldness

4. Page classification attributes

5. Data structure tags if any

[Pankaj] Our modeling work begins by learning from an all time great, David Bailey of NASA Ames and DOE LBL NERSC. His 1998 paper Challenges of Future High-End Computing models a system of compute and memory using the simplest rule of queueing theory, Little's Law, and presents beautiful and timeless conclusions about the relationship between memory latency and the parallelism needed to exploit that memory at a certain bandwidth. We start to understand where the price will be paid for the extra latency CXL is adding to the memory path.