Notes from the team building C3X. Cost estimation for Terraform, the economics of cloud infrastructure, and how to ship FinOps tooling without a SaaS gate.
AWS Secrets Manager charges per secret per month plus API calls; Azure Key Vault and GCP Secret Manager charge mostly per operation with tiny per-secret fees. The cheapest depends on secret count vs access frequency. This compares them.
AWS API Gateway, Azure API Management, and GCP's API Gateway/Cloud Endpoints price API management differently, per-request vs tiered subscription vs per-call. The cheapest depends on volume and feature needs. This compares them.
Partitioning splits a large table into segments by a key (often date), so queries with a matching filter scan only relevant partitions instead of the whole table. On per-scan and IO-billed systems, that directly cuts cost. Here is the tradeoff.
Rate limiting caps how many requests a client can make, protecting your backend from abuse, runaway clients, and traffic spikes that would drive up compute and downstream cost. It is a cheap control that prevents expensive incidents. Here is the case.
ExpressRoute bills a circuit port fee by bandwidth, in a metered plan (per-GB egress) or unlimited plan (flat, higher fee), plus provider connectivity charges. For large, steady traffic to Azure it can beat VPN. Here is the model.
Private Link bills per private endpoint per hour plus per GB of data processed, keeping traffic to Azure services on the backbone privately. It adds cost but can reduce egress and improve security. Here is the model and the tradeoff.
Cloud Interconnect bills per VLAN attachment (by capacity) plus discounted egress over the connection, in Dedicated (your own 10G/100G ports) or Partner (via a provider) flavors. For large steady traffic it beats VPN. Here is the model.
Azure NetApp Files bills provisioned capacity pool size per hour by service level (Standard, Premium, Ultra), which set throughput per TB. You pay for the pool whether or not it is full, so right-sizing capacity and tier is the lever.
Azure Files bills by tier, transaction-optimized, hot, and cool (pay per GB stored plus transactions), or premium (provisioned capacity with bundled performance). Matching tier to access pattern and using premium only when needed are the levers.
Hyperdisk is GCP's newer block storage that decouples capacity, IOPS, and throughput, so you provision and pay for each independently. This flexibility avoids over-buying, but over-provisioning any dimension wastes money. Here is the model.
Stream Analytics bills per streaming unit-hour, the provisioned capacity that runs your streaming query, whether or not data flows. Right-sizing streaming units to real throughput and stopping idle jobs are the levers. Here is the model.
Dataplex bills per data compute unit (DCU) hour for its processing tasks (discovery, quality, tiering) plus shuffle storage, and its catalog metadata. Running unnecessary tasks and over-provisioning DCUs drive the bill. Here is the model.
gRPC's binary encoding and HTTP/2 multiplexing send fewer bytes and use connections more efficiently than JSON over HTTP/1.1, which can cut egress and compute at high volume. But REST is simpler and universal. Here is the cost tradeoff.
Images are often the largest bytes a site serves, so optimizing them, right-sizing dimensions, modern formats like WebP and AVIF, and compression, cuts egress and storage substantially. Here is where image optimization saves cloud cost.
AWS (ALB/NLB), Azure (Load Balancer, Application Gateway), and GCP (Cloud Load Balancing) price load balancing on hourly plus capacity/data axes. Layer-4 is cheaper than layer-7. This compares them and the shared choice of layer to your needs.
Terraform remote state on S3, GCS, or Azure Blob costs almost nothing, tiny storage plus a few requests per run. The only real cost is state locking (a DynamoDB table on AWS) and, at scale, request volume. Here is the tiny model.
Translate bills per character of input text (about $15 per million), with a free tier and higher rates for custom terminology and Active Custom Translation. Translating more text than needed drives the bill. Here is the per-character model.
Polly bills per character of text synthesized to speech, with neural voices costing several times more than standard, plus a free tier. Synthesizing the same text repeatedly, or using neural where standard suffices, drives the bill. Here is the model.
Kendra bills per index-hour by edition (Developer and Enterprise), running around the clock, plus connector scans and extra query or document capacity. The always-on index fee is a significant baseline. Here is when Kendra's intelligent search pays off.
Personalize bills data ingestion, training hours to build recommendation models, and real-time inference by provisioned throughput (TPS-hours). Over-provisioned recommendation capacity and frequent retraining drive the bill. Here is the model.
Amazon Connect bills per minute of end-customer contact usage plus telephony (per-minute inbound/outbound and phone number charges), with optional add-ons for features. Call volume and duration drive the bill. Here is the pay-as-you-go model.
Communication Services bills per usage across channels, per minute for calling, per SMS, per email, per chat message, with pay-as-you-go rates and no fixed platform fee. Channel volume drives the bill. Here is the per-channel model.
Cloud Vision bills per image per feature (label, face, text, object, safe-search, and more), so requesting several features on one image bills for each, with a free tier and volume discounts. Analyzing more features than needed drives the bill.
CloudFront, Azure Front Door, and Google Cloud CDN price content delivery similarly, egress to users (tiered by region) plus requests, with free origin-to-CDN transfer. The cheapest depends on traffic geography and volume, and a CDN usually lowers total cost.