Cross-cloud application-tier placement and Atlas internet egress
Parent: Migration and Cutover Cost (Live Migration, mongosync, Dual-Running, Egress) · Published reference · snapshot 2026-10-31
↓ Facts as markdownall context files
Source-anchored research for Cross-cloud application-tier placement and Atlas internet egress. Run status: COMPLETED.
These notes link each claim to its source. Some citations identify archived research reports that are no longer hosted on this site. A published reference means the content is available; it does not certify independent review or accuracy.Read the editorial policy and follow the sources before relying on a claim.
How it works
- Private connectivity does not imply zero network cost: Atlas documents private endpoint billing, and AWS documents hourly and data-processing charges for interface endpoints plus charges when VPC peering traffic crosses availability zones. Evaluate the actual route and endpoint costs. (high confidence) (verified as of 2026-10-14) [source]
- When application relocation is impractical, reduce repeated reads and returned fields and enable driver network compression. MongoDB's optimization article also recommends directing eligible secondary reads to nodes in the same provider region; validate CPU impact and read semantics before changing settings. (high confidence) [source]
- A Google Cloud forum billing discussion initially attributed carrier-peering charges to Atlas, but the operator later identified Cloudflare as another possible destination. Use network-flow and billing evidence to attribute charges before assuming that all application-cloud egress comes from Atlas. (low confidence) [source]
- Same-region placement does not guarantee free networking. Google prices external-address traffic between different VPC networks at internet transfer rates, while AWS documents processing and transfer charges for public routes between VPCs, with additional costs when traffic crosses zones. (medium confidence) (verified as of 2026-10-30) [source]
- A VPN does not automatically eliminate internet transfer billing: Google charges internet data-transfer-out rates when a Cloud VPN tunnel connects to a gateway outside Google Cloud. This describes the Google side, not an Atlas billing exemption. (low confidence) (verified as of 2026-10-30) [source]
Problems, failure modes and limitations
- Colocation must preserve the required failure tolerance. MongoDB says regional failure of a single-region deployment requires backup recovery; AWS warns that confining workloads to one availability zone trades reliability for lower network costs. (medium confidence) [source]
Comparisons and alternatives
- Inference for a staged cutover: while applications remain in another cloud, normal database requests and responses continue crossing that boundary. Budget recurring directional traffic during this interval, rather than treating transfer solely as a one-time database copy. (medium confidence) [source]
Facts and statements
- MongoDB recommends placing applications and Atlas resources in the same provider and region to reduce database latency and cross-cloud egress costs; its billing guidance likewise recommends local query origins. (high confidence) [source]
- Google Cloud classifies traffic to a location outside Google, without Cloud Interconnect, as internet data transfer out. Inbound transfer is generally uncharged, but responses to inbound requests count as outbound transfer and processing resources can still incur charges. (low confidence) (verified as of 2026-10-30) [source]
- For AWS-hosted application workloads reaching third-party services through NAT, gateway placement can add cross-AZ transfer costs. AWS recommends considering a NAT gateway per AZ to avoid those transfers, while acknowledging the extra gateway cost; this is an additional application-side budget consideration for Atlas connectivity. (medium confidence) [source]
- For Atlas on AWS or Google Cloud, the billing documentation excludes incoming cluster transfers from charges and separately classifies outbound internet traffic. Free and Flex clusters have an outgoing-transfer exception; dedicated-cluster assumptions should not be applied to those tiers. (low confidence) (verified as of 2026-10-14) [source]
- A 2017–2019 Meteor operator thread reports unexpectedly high transfer after moving databases to Atlas, including DigitalOcean-hosted applications accessing AWS-hosted Atlas. These are historical firsthand reports with unresolved attribution, not controlled measurements establishing a general cross-cloud traffic multiplier. (low confidence) [source]
- AWS private connectivity to Atlas has its own bill: Atlas documents endpoint hosting and capacity charges, plus AWS interface-endpoint and processed-volume charges. Replacing a public route therefore requires a total-cost comparison. (low confidence) (verified as of 2026-10-14) [source]
- Estimate a staged cutover by direction: Atlas-to-application results can generate Atlas egress, while application-to-Atlas traffic can generate charges at the application provider. This is a budgeting inference from Atlas transfer rules and Google outbound/processing rules, not a measured migration result. (medium confidence) [source]
- Google Cloud NAT adds gateway, processed-traffic, and external-address charges alongside applicable outbound transfer. AWS advises evaluating NAT placement: a gateway per availability zone can avoid inter-zone traffic but adds gateway cost. (medium confidence) (verified as of 2026-10-30) [source]
- Atlas recommends projections and client-driver network compression to reduce bytes sent to applications. These can reduce cross-cloud exposure while application placement remains unchanged, but the documentation provides no universal savings percentage. (low confidence) [source]
Children
- No children recorded.