Throughput Limitations for Dedicated Clusters

Hello everyone, I want to know if there are any throughput limitations for dedicated clusters (M10+) cluster similar to the ones applied to free and shared clusters.

I am load testing my REST APIs which use MongoDB, and at a certain RPS (1000), the API response times begin to go up. Interestingly, the CPU of the cluster that I am using (M40) to in this setup stays at 40-50% range even if I increase my RPS, so I assumed maybe the bottleneck is with the compute servers but even if I double the server count it still remains the same which makes me wonder if there is anything else from the database side causing the issue, like an operational throughput limitation. Please note that, Atlas is the only dependency for the APIs which only do perform some read queries (all of which use indexed fields) and return the response. The amount of data stored is only 600 MB, so I would think the Disk IOPS is probably not the reason.

As mentioned in the documentation linked above, the free and shared clusters have operational limits beyond which Atlas throttles them or begins a cool down period. I wonder if there are similar limits to dedicated clusters too.

Moreover, M10 & M20 clusters are recommended for low traffic applications & M30+ is suggested for production deployments. Could this be because of the operational limits?

I would greatly appreciate a detailed response as this is crucial for our capacity planning. Please share your thoughts on what else (related to Atlas) could be causing the high response times.

@Jason_Tran @Aasawari Apologies for the direct ping. Can I please get some help on this?

Hi @Raahel and welcome to the community forum.

As mentioned in our documentation for Atlas Service Limits, there are certain limitations for each of the clusters in terms of number of connections, collections, and the database privileges etc.
My recommendation would be going through the documentation to understand if your cluster fails to satisfy any of these conditions.

For further assistance, I would also recommend you to reach out to the Atlas Chat support who would help you by looking into the cluster and suggest a possible solution.

Best Regards
Aasawari