[W1] The choice of benchmarks, i.e., JOBlight and TPC-H, is a bit weak.
[D1] Several recent works on (learned) query optimization, have shown that both JOBlight and TPC-H are not challenging enough to yield meaningful insights into the benefits of a (learned) query optimization technique. I therefore recommend that the authors re-run their experiments with JOB (full) and DSB. Also, even though JOBlight is used, it is always referred to as JOB, which I found somewhat misleading.
Action 1: Have three datasets: TPC-H, JOB (if possible the full version), and DSB? or some other non-relational benchmark. Three in-distribution workloads and two out-of-distribution workloads
[W1] The choice of benchmarks, i.e., JOBlight and TPC-H, is a bit weak.
[D1] Several recent works on (learned) query optimization, have shown that both JOBlight and TPC-H are not challenging enough to yield meaningful insights into the benefits of a (learned) query optimization technique. I therefore recommend that the authors re-run their experiments with JOB (full) and DSB. Also, even though JOBlight is used, it is always referred to as JOB, which I found somewhat misleading.
Action 1: Have three datasets: TPC-H, JOB (if possible the full version), and DSB? or some other non-relational benchmark. Three in-distribution workloads and two out-of-distribution workloads