Best practices for scaling Amazon Comprehend custom text classification throughput

Question

We are using the Comprehend asynchronous API for a custom text-only classification model. Per the documentation, asynchronous requests are limited to 10 active jobs. We are finding job completion times ranging from 5-10 minutes, causing a bottleneck for our required sustained throughput of 150 jobs (peaking at 300 jobs) per hour.

Any insights on best practices for increasing Comprehend custom classification throughput would be greatly appreciated. Is there a recommended way to scale out the processing using additional resources?

Answer

I think based on the data you have provided, seems you are doing OK in terms of throughout. Other than making sure to start a new job as soon as one completes (by retrying Start Job Operation on limit exceeded), there's not much you can do.

If you need to increase the active job limit which can help more work done within same time period please cut a support ticket with data and requirement and we can take a look at it.

Best practices for scaling Amazon Comprehend custom text classification throughput

Contenido relevante