Closed hbdeshmukh closed 8 years ago
Ah interesting. I didn't consider this... but this is a bug on quickstep's part right? I mean once the HT issues are resolved, we can go back to preloading?
@hbdeshmukh updated via Github commit.
Thanks @cramja
As I mentioned earlier, since we don't preload the buffer pool for other systems, we should follow the suit for Quickstep.
The preload buffer pool flag is set to true in the TPC-H run-benchmark script. I think it should be set to false, because some of the queries require larger hash table allocations. If we preload the buffer pool with unnecessary tables, we may run out of memory soon enough, causing evictions and degrading performance. Besides, we evaluate other systems by not preloading their respective buffer pools.