The Post-SaaS Architecture: How a $14,000/month AWS bill for 35 req/sec cured our team of cloud delusions
A few months ago, I was brought in to audit the infrastructure of a post-Series A fintech company that was bleeding cash. The leadership team couldn't understand why their monthly Amazon Web Services bill was hovering around $14,200 while their actual traffic—measured at the edge—peaked at roughly 35 requests per second during business hours. Thirty-five. That is not a typo. That’s about 2,100…
In a recent case study, a post-Series A fintech company had been struggling with an astronomical AWS bill of $14,200 per month. Despite having traffic peaking at around 35 requests per second, the founders were convinced that a cloud architecture consisting of fourteen microservices across numerous EKS pods, an Aurora PostgreSQL cluster, NAT Gateways, managed MSK, and Datadog monitoring was the epitome of modern, resilient architecture.
This high-level approach was born out of reading various blog posts from successful tech companies like Netflix and Uber, and hiring developers who were only familiar with AWS tools.
Upon audit, it became clear that the excessive cloud infrastructure was not only draining their finances but also impacting performance. The team identified that they were incurring costs for unnecessary services and network hops, which could be eliminated to achieve significant cost savings and improved performance.
The team embarked on a four-week project to rebuild their infrastructure. They transitioned from a modular monolith to a multi-layered microservices architecture, removed Kafka, replaced Datadog with structured logging, and opted for a single pair of dedicated bare-metal servers with advanced hardware components. This shift resulted in a monthly cost reduction from over $14,000 to around $720, including offsite backup storage and DNS routing.
Most notably, the team experienced a dramatic improvement in performance. Their average API p99 latency dropped from 145 milliseconds to just 11 milliseconds. This was attributed to the elimination of multiple internal network serialization hops, software load balancers, and TLS handshakes that were adversely affecting their system's efficiency. They realized that physical hardware could execute a vast number of operations per second with predictable latency, something virtualized cloud instances could not match.
The key takeaway from this case study was the industry's shift from understanding hardware fundamentals to relying on managed cloud APIs. It highlighted the importance of learning your operating system, understanding memory hierarchy, and writing clean queries. The story serves as a reminder that cloud infrastructure should primarily be utilized to transform capital expenditures into predictable operational expenditures, not to monetize architectural inefficiencies.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
