Moving to Google Cloud managed services, from a FinOps point of view

1

Google Cloud Pub/Sub

Pub/Sub, the GCP managed service for message queuing, has two levels of services on standard and one Lite. Standard is the high availability version of it and Lite could be a zonal or regional service with infrastructure managed by the client. Obviously, the model pricing will be very different with a x10 between Standard and zonal Lite. However, the model pricing is the same is based on throughput for message publishing, message storage costs and egress for message distribution. Here, we totally break the similarity with a VM model (except on storage). Everything is drived on volumetry and performance of inbound and outbound messages.

#Stream Processing #Data Integration #Web Service Automation 15 social mentions

2

Google Cloud Memorystore

Memorystore, the GCP managed service for cache, is not a service by itself, you need to choice the backend behind with Redis or memcached. These two kinds of configurations for Memorystore do not have the same model pricing. Memorystore for memcached is mostly based on Compute Engine model with pricing based on the number of nodes and vCPU + RAM per node. Even if the model pricing is nearly the same, the prices themselves are different to allocate a VM with a flavor. However, the egress traffic is not billed within the same region on Memorystore for memcached contrary to VM inter-zone egress. What can we conclude from that? Well, the node itself will be more expensive than a VM but the inter-zonal egress is not to take into account. The migration choice should be balanced on both model pricing differences.

#API Tools #API Management #APIs 6 social mentions

3

Google BigQuery

BigQuery has a pricing model close to Pub/Sub : you pay for what you insert on the database (in streaming) and the storage of these data. The main difference is on what you can do with these data. BigQuery is not a message queuing service, this is a data warehouse service. It proposes a query service to exploit these data and you pay for these queries. Actually, not on the query itself but on the quantity of data manipulated for producing the results of the query. This means that you do not directly pay for a power capacity on query but on data transfer to produce the result which is very different from a none managed database perspective such SQL databases where the main model pricing is the node size to store and query data.

#Data Management #Data Warehousing #Data Dashboard 35 social mentions