Scenario: Frenquency and size of batch operation
Learn how you can control the frenquency of triggering a new batch operation, the number of agents that should receive the new operation, and resolve the issue of network saturation.
- You have 10000 agents and the operation is to download a 100 MB asset to each agent from EFM. The expectation is that an average agent is able to download that asset in 20-30 sec (3-5MB/s). Assume that EFM is able to deal with 100MB/s network IO for this purpose. So, you can see that roughly 20-30 parallel operations can be handled by EFM. This could easily cause network saturation.
- On the following diagram you can see with the
rollingBatchOperationsFrequencyparameter you can define how often EFM tries to trigger a new batch operation, and the
rollingBatchOperationsSizeparameter defines how many agents should receive the new operation.
- A good starting point could be
rollingBatchOperationsFrequency=5s. In the first 1-4 iterations, the queue becomes full but after it diverges, the operation behaves as expected. The expectation is that roughly 25% of the queue will be free in each iteration with 5s frequency.