Cloud providers offer many services in managed form: the provider operates the underlying servers, and the customer is billed by broker, node, or database instance. Amazon MSK (managed Kafka), Amazon MQ, ElastiCache, and DocumentDB are examples. Usage and billing files describe these lines by service rather than by server, so the instance hardware running the service does not appear in the billing data in the usual way. Tailpipe identifies the instance family that runs each broker or node and applies the same per-component hardware model used for compute instances.
Identifying the Underlying Instance
For MSK, Amazon MQ, and ElastiCache, the usage type of each billing line records the instance shape that runs the broker or node, for example Kafka.m5.large, Single-AZUsage:mq.t3.micro, or NodeUsage:cache.r6g.large. Tailpipe reads the instance shape from the usage type and applies the hardware profile of the equivalent instance family, so an MSK broker on kafka.m5.large is modeled as an m5.large. Each component of the standard model then applies as described in the preceding sections: CPU power draw from the TDP curve, memory, motherboard, power supply losses, and PUE, together with the embodied emissions of manufacture.
DocumentDB billing lines already carry a database instance shape, such as db.r5.large, so Tailpipe calculates DocumentDB emissions directly using the database instance profiles it already uses for RDS.
Utilization
Tailpipe does not currently collect the per-broker and per-node CPU metrics that these services publish, so managed service nodes are calculated at Tailpipe’s standard default of 15% utilization. This is the same default applied to any instance without collected metrics, and is at the low end of the measured power curve described in the Processors section.
Storage
Managed service brokers and nodes store their data on network volumes, which the provider bills as separate storage lines. Tailpipe calculates the emissions of those lines using the Network Storage method, with per-GB power coefficients by drive class. To avoid counting the same storage twice, the compute line of a managed service carries no local-disk component; storage emissions are attributed only to the storage lines. MSK broker storage does not yet have a published coefficient in Tailpipe’s model, so it is currently left unattributed. This understates MSK emissions slightly and will be resolved as coverage extends.
Lines Without an Instance Shape
Some managed service lines carry no instance shape, such as serverless variants like MSK Serverless, or request-based charges. Where the underlying hardware cannot be established from the billing data, the line retains its cost but is not attributed modeled emissions. The same applies to services whose energy use cannot credibly be derived from billing data, such as GuardDuty and CloudWatch ingestion. The services Tailpipe covers are listed in “Which Cloud Services does Tailpipe cover?”.