Memory compaction issues

When I've run into problems with compaction on our servers it was painfully obvious. Our workloads would become 100x slower and CPU time would be almost entirely 100% system time. A quick 'perf record -g -a' would point the finger at the compaction code and an "echo never > /sys/kernel/mm/transparent_hugepage/defrag" would solve the problem.

My question is not, how do I know it is happening? Rather once it is happening on my systems, how can I gather useful information that will allow the memory management developers to understand and fix the issue? These issues are admittedly hard because there is very little incentive for us to track down the problem when we can simply turn off defrag.