- 01
Market
Saudi Arabia
- 02
Written for
CTO, Head of Digital
- 03
Decision
Find where the transfer charges come from and re-architect the paths that bleed
- 04
Languages
EN · AR
The hiding places
Four places transfer charges accumulate without being named
None of these paths is a mistake. Each is correct engineering that happens to cross a priced boundary — which is exactly why the charges survive every architecture review.
- 01
Services in different zones holding a conversation
Spreading services across availability zones is what resilience looks like — and traffic between zones is metered, often in both directions. A request that fans out into a dozen internal calls can cross that boundary a dozen times before the user sees a byte.
- 02
The NAT gateway metering every byte it forwards
Private subnets reach the internet through a NAT gateway, and the gateway charges for the bytes it processes — on top of the egress those bytes already incur. An estate that pulls container images, packages and updates through NAT all day pays twice for every download.
- 03
Replication doing its job across a priced border
Cross-region replicas, disaster-recovery copies and backups shipped to another geography are exactly what a resilient estate should have. Every byte of them crosses the most expensive kind of boundary, continuously, and the charge scales with how much data the business accumulates.
- 04
API responses heavier than anyone budgeted
Everything your APIs return to the internet is egress. Payloads accumulate fields release by release, mobile apps poll on schedules, and nobody owns the weight of a response — so the line grows in proportion to the product’s success, which is why it never looks like a problem.
The architecture
Chatty services turn resilience zones into a metering boundary
The pattern arrives honestly. A monolith is split into services; the services are spread across zones so no single failure takes the system down. Both decisions are right. But a call that used to be a function invocation inside one process is now a network request, and when caller and callee sit in different zones, it is a network request with a price on it. One user action that fans out into dozens of internal calls multiplies that price on every request, invisibly — no single service sees more than its own small share, and the total appears only on the bill.
The fixes are architectural, and they start with measurement rather than principle. Zone-aware routing keeps a request’s fan-out local where health allows, crossing the boundary only when resilience requires it. Batching collapses chains of small calls into fewer, fuller ones. And sometimes the honest finding is about the boundaries themselves: two services that exchange this much data on every request may be one service wearing two deployment costumes, and merging them removes the metered path entirely. Which fix pays depends on which paths carry the spend — which is what the flow logs are for.
The cost tools
Every hop a response avoids is a hop nobody pays for
Caching and CDNs are usually sold as performance tools. They are also cost tools, for the same reason: the cheapest byte to move is the one that stops travelling.
A flow: serve repeats from the CDN edge, absorb reads in a cache, keep fan-out inside the zone, shrink what still travels — with a branch for measuring the paths first so each tool is justified.
- 01
Serve the repeat from the edge
A cacheable response served by the CDN near the user never reaches the origin at all — the estate’s egress simply does not carry it, and the request costs whatever the edge charges instead.
- 02
Absorb the read before the backend
A cache in front of the services keeps repeated answers inside one zone, so the same question stops re-crossing the metered boundary to be answered identically each time.
- 03
Keep the fan-out inside the zone
Zone-aware routing prefers local replicas of the services a request needs, crossing to another zone only when the local copy is unhealthy — resilience kept, chatter unmetered.
- 04
Shrink what must still travel
Compression, pagination and payloads trimmed to the fields the client actually uses make the remaining traffic lighter — unglamorous work that discounts every path at once.
Measure the paths first
Flow logs decide which of these tools pays: a CDN in front of traffic that never repeats, or a cache for answers that change every second, adds a bill without removing one.
The exit dimension
Egress is also the exit fee, so price the door while you are happy
Egress pricing says something plainly that deserves to be heard plainly: data flows into the cloud free and is charged on the way out. The consequence reaches beyond the monthly bill. The cost of moving your data to another provider, or back on-premise, is metered by the byte — so the more data the estate accumulates, the more expensive its own exit becomes, without anyone signing a lock-in clause. That is not an accusation of bad faith; it is the published price structure, and it deserves to be read as one.
The sensible response is not refusing to commit; it is knowing the number. Estimate what moving the estate’s data would cost today, and revisit the estimate as the data grows. Keep the heaviest flows behind interfaces that could be repointed, so the biggest datasets are movable in principle. And weigh egress in architecture reviews the way latency is weighed — as a property of every arrow drawn. Lock-in is not avoided by hovering at the door; it is managed by knowing the exit price before the day it matters.
Our role
What we measure on the wire and what your architects decide
- 01
The paths, attributed
We turn the transfer line into named routes with flow-log analysis — which services chat across zones, what moves through NAT, what replication and responses carry — so the conversation is about paths, not a total.
- 02
The fixes, ranked by what they remove
CDN offload, caching, zone-aware routing and payload work are proposed in the order the measurements justify, each applied as an observed, reversible change rather than a belief about best practice.
- 03
The exit price, on paper
The review includes an estimate of what moving the estate’s data would cost today, so the lock-in dimension of egress enters your planning as a number that gets revisited, not a worry that gets repeated.
How much locality to trade against resilience, and how much data gravity to accept, are architecture calls that stay with your team; we make sure they are made looking at the priced paths rather than the unpriced diagram.
Common questions
What buyers ask
Why did splitting the monolith into microservices raise our transfer bill?+
Because calls that used to happen inside one process became network requests, and many of them now cross availability-zone boundaries that are metered per gigabyte. The multiplication is what hurts: one user action fanning out into dozens of internal calls pays the crossing many times per request, and no single team sees more than its own slice. The fix is rarely fewer services. It is zone-aware routing so a request’s fan-out stays local when health allows, batched calls where chains of small ones grew up, and an honest look at any pair of services that exchange large volumes on every request — that pattern sometimes means the boundary between them was drawn in the wrong place.
Will a CDN cut our transfer costs, or just add a second bill?+
It depends entirely on how much of your traffic repeats. A CDN earns its keep when many users request the same responses — static assets, media, cacheable API answers — because every edge hit is a request the origin never serves and the estate’s egress never carries. It adds little when responses are unique per user, personalised or changing every second, because those requests pass through to the origin anyway, and now two providers are involved in delivering them. The deciding measurement is cacheability: what share of responses could legitimately be served from a cache, and for how long. Measure that first, and the CDN decision makes itself — in either direction.
Are egress charges really a form of vendor lock-in?+
They function as one, whatever the intention behind them. Bringing data in is free; taking it out is metered; therefore the cost of leaving grows automatically with everything you store. No contract clause is involved — the price list does the work. The productive response has three parts. Know your exit number: estimate what moving the estate’s data would cost today, and update it as data accumulates. Keep the heaviest data flows behind interfaces you could repoint, so the largest datasets remain movable in principle. And treat egress as a criterion in architecture reviews, the way latency and resilience already are. None of that means avoiding the cloud; it means committing with the door priced.
In this cluster
Map your transfer paths with us
A transfer-path review that attributes the egress line to named routes — zone crossings, NAT, replication, responses — and ranks which fix pays first.
Start the conversation
