DRAFT: Branch for policy iteration with rocksdb. - #282
Conversation
Runs it with 30 random seeds first filling the database with random data and then running the mixgraph workload. We modify the common files to create a way to parse the output of a seeded run. At some point other large benchmarks should be added with the same behavior and the scripts will need to be modified. However, for the time being we just write the script in jq. Perhaps a python parser is warranted for later. This does generate the data in a nice comparable format.
97bddad to
e78dd48
Compare
|
I don't think you actually wanted to request a review. There is a "draft PR" state which avoids sending automatic review requests. I will note however, that I'm confused why this branch is needed. Iteration should be happening locally, if we don't have a good way to run the benchmark. We need to fix that. I think there should be a script already. |
Only choose the same two seeds so that the run takes ~10 mins.
Add t-test infra in order to check if things are different. We will then add commit hashes to the changes to get testing to work.
I messed up when creating the branch my bad.
I wanted to show the code to Gaurav to see what parts he needed to peel off for the benchmark suite. I just am pushing to this in case there is some stuff he can use. If preffered I can close this. |
2 participants
No description provided.