WiredTiger recently added support for application defined transaction IDs (timestamps). The timestamp API allows MongoDB to remove some oplog management infrastructure, but there are potentially some performance trade offs, and will likely be work to do to optimize the oplog in WiredTiger to ensure equivalent performance is retained.

We should find or create a workload that can be used to measure oplog throughput, and compare the throughput using different versions of WiredTiger and MongoDB.

The workload should:

• Be able to stress out the oplog capacity.
• Be able to run in both single and multi-threaded configurations
• Probably be capable of adding documents of varying sizes
• Run in either single-node replica set, or multi-node replica set configurations: i.e: be able to measure oplog performance on the primary and on secondaries.