---
title: "How OrderServ achieved up to 64% cost savings with multi-tenant ECS"
description: "OrderServ, a food-ordering platform, faced linear cost growth and slow scaling as it grew from one to multiple QSR brands, each running on dedicated ECS clusters with 21 microservices and 315 sidecar containers. AWS Professional Services and Cognizant transformed it into a multi-tenant SaaS by sharing a single ECS cluster with a tenancy library that enforces tenant isolation at the application level using request-scoped context. They replaced AWS Fargate with EC2 warm pools running Bottlerocket, reducing scaling time from 5-7 minutes to 30-60 seconds and achieving 47-64% compute cost reduction. A crash observability pipeline using FireLens and S3 provided zero-loss logging with full stack traces at $10/month per service, and passkeys via Amazon Cognito cut authentication costs by 80% while improving login latency. The combined annual savings exceeded 60% of previous infrastructure costs, with tenant onboarding reduced from days to minutes."
---

# How OrderServ achieved up to 64% cost savings with multi-tenant ECS

[AWS](https://yomu.fyi/company/aws) · Chetan Yewale · Nov 6, 2026

**Type:** Problem & solution

## Summary

OrderServ, a food-ordering platform, faced linear cost growth and slow scaling as it grew from one to multiple QSR brands, each running on dedicated ECS clusters with 21 microservices and 315 sidecar containers. AWS Professional Services and Cognizant transformed it into a multi-tenant SaaS by sharing a single ECS cluster with a tenancy library that enforces tenant isolation at the application level using request-scoped context. They replaced AWS Fargate with EC2 warm pools running Bottlerocket, reducing scaling time from 5-7 minutes to 30-60 seconds and achieving 47-64% compute cost reduction. A crash observability pipeline using FireLens and S3 provided zero-loss logging with full stack traces at $10/month per service, and passkeys via Amazon Cognito cut authentication costs by 80% while improving login latency. The combined annual savings exceeded 60% of previous infrastructure costs, with tenant onboarding reduced from days to minutes.

## Context

OrderServ, an omnichannel food-ordering solution, initially ran each brand on a dedicated ECS cluster with 21 microservices on AWS Fargate. As it grew from one to five brands, costs scaled linearly, idle capacity was wasted, and scaling during traffic surges took 5-7 minutes, risking lost sales.

## Approach / What changed

The solution involved four architectural changes: (1) consolidating dedicated ECS clusters into a single shared cluster with a .NET Core tenancy library for application-level isolation; (2) replacing AWS Fargate with EC2 warm pools running Bottlerocket for sub-minute scaling; (3) implementing a crash observability pipeline using FireLens, S3, and Athena for zero-loss logging; (4) integrating passkeys via Amazon Cognito to replace SMS OTP authentication.

## Takeaways

- Application-level isolation via a tenancy library can replace per-tenant infrastructure, with IAM and resource policies as a backstop to contain defects.
- EC2 warm pools with Bottlerocket reduce scaling time from minutes to seconds compared to AWS Fargate, enabling 47-64% compute cost savings.
- Passkeys eliminate SMS OTP costs and improve login latency from 15-30 seconds to ~2 seconds, with phishing resistance.

## FAQ

### How does the tenancy library enforce isolation?

The library uses request-scoped storage (AsyncLocal<T>) to resolve tenant context per request, routes all data access with tenant prefixes, and refuses to construct resource IDs without a tenant context. It also acquires tenant-scoped credentials via sts:AssumeRole with session tags, and Amazon ElastiCache for Valkey enforces per-tenant RBAC as an additional layer.

### What is the crash observability pipeline and its cost?

It uses FireLens to send logs to S3, with Athena for querying. It provides zero-loss logging and full stack traces at approximately $10/month per service, with 90-day forensics.

### How were scaling times improved?

By moving from AWS Fargate to EC2 warm pools with Bottlerocket, scaling time reduced from 5-7 minutes to 30-60 seconds. Warm pools keep instances initialized and ready, and Bottlerocket reduces boot time.

**Tags:** [Amazon ECS](https://yomu.fyi/topic/amazon-ecs), [AWS](https://yomu.fyi/topic/aws), [Cost Optimization](https://yomu.fyi/topic/cost-optimization), [Multi-tenant](https://yomu.fyi/topic/multi-tenant), [SaaS](https://yomu.fyi/topic/saas)

- Source: [AWS](https://aws.amazon.com/blogs/architecture/how-orderserv-achieved-up-to-64-cost-savings-with-multi-tenant-ecs)
- Source URL: https://aws.amazon.com/blogs/architecture/how-orderserv-achieved-up-to-64-cost-savings-with-multi-tenant-ecs
- Ingested by Yomu: 2026-11-06T03:00:13.790Z

[Read original post](https://aws.amazon.com/blogs/architecture/how-orderserv-achieved-up-to-64-cost-savings-with-multi-tenant-ecs)
