Achieving 99% Cost Reduction in AWS Data Transfer Expenses
How an unconventional use of S3 turns a cross-availability-zone data-transfer bill into a near-zero one.
AWS data-transfer charges scale unevenly with architecture decisions, and cross-availability-zone transfers are one of the line items that grow fastest. A simple architectural change using S3 as an intermediary cuts the cost by roughly 99% for the right workloads.
The Challenge
AWS data-transfer pricing is one of the more counterintuitive parts of the cloud bill. The rates vary by direction, by region, by service, and most expensively by availability zone — a transfer between two EC2 instances in different AZs of the same region is billed at a per-GB rate that adds up quickly for any workload that moves significant volume across zones.
For workloads that span multiple availability zones for resilience reasons — which most production federal workloads do — those cross-AZ charges become a meaningful fraction of the monthly cloud bill. And because they scale linearly with traffic, they are exactly the kind of cost that gets worse as the workload succeeds.
The Solution
The technique exploits a specific characteristic of Amazon S3 pricing: data transfer between AWS compute and S3 in the same region is free, regardless of which availability zones are involved. S3 acts as a free intra-region transfer fabric in a way that direct compute-to-compute traffic does not.
The architecture is straightforward in principle:
The mechanism
Instead of transferring data directly between resources in different availability zones, the source writes the data to an S3 bucket in the same region, and the destination reads it back from S3. The intermediate write and read are both free of data-transfer charges. The only cost incurred is the brief storage cost while the object exists in S3.
Why the storage cost stays nominal
Objects are deleted immediately after the destination has read them, so the storage footprint at any moment is bounded by the in-flight transfer volume rather than by the cumulative volume. For workloads that run on the order of seconds to minutes per object, the storage cost is rounding error against what the cross-AZ charges would have been.
What changes architecturally
The producer and the consumer of the data stop talking to each other directly. Both talk to S3. This adds a small amount of latency (the round-trip through object storage) and a small amount of complexity (lifecycle management for the intermediate objects), in exchange for nearly eliminating the recurring data-transfer cost.
When This Applies
This pattern is not universally applicable. Two conditions need to hold for it to make sense:
- The workload tolerates the additional latency. S3 round-trips add tens to low-hundreds of milliseconds per object, which is fine for batch and asynchronous processing but not for low-latency synchronous traffic.
- The cross-AZ traffic is large enough that the cost matters. For workloads moving small volumes across zones, the savings do not justify the architectural change. The pattern starts to pay off when monthly cross-AZ charges reach the low thousands of dollars and gets dramatically more attractive above that.
When both conditions hold, the technique generalizes — it works regardless of organization size, AWS account structure, or workload type, and it has been applied across federal data pipelines, analytics workloads, and media-processing systems with consistent results.
Results
In controlled experiments and in production deployments, the technique has produced consistent outcomes:
- Roughly 99% reduction in data-transfer charges for the cross-AZ traffic that was rerouted through S3.
- Negligible incremental storage cost, because intermediate objects are deleted immediately after the destination reads them.
- No change in workload reliability. S3 is a strongly consistent, multi-AZ-replicated service; using it as a transfer fabric does not weaken the durability or availability properties of the workload.
- Architectural portability. The pattern has been applied at workloads of varying scale; it is not a one-off trick.
Ready to discuss your challenges?
Contact One Dynamic to explore how we can help your organization.
CONTACT ONE DYNAMIC