distinguish the memory access behaviors of benchmarks when they run concurrently

[bigplanewangyue]

When running two benchmarks (e.g., Benchmark A and B) concurrently in a runner, how can I distinguish their distinct memory access behaviors—such as which addresses each benchmark accesses and how much time they spend on memory operations? I’ve attempted to do it but haven’t succeeded. Could you provide guidance?

[syifan]

Could you explain slightly more about what you want to achieve? For example, at which level (L1 cache, TLB, DRAM) do you want to differentiate the source of the transaction? Also, for the time that is spent on each workload, how do you consider overlapped transactions?