---
title: High-Concurrency Server I/O: epoll, Syscalls, and Zero-Copy on AWS EC2
description: C10k is solved until syscall overhead and context switches eat your Graviton cores. epoll, sendfile, and SO_REUSEPORT behaviors on EC2, and why Lambda caps concurrency differently.
url: https://www.joinhorizon.co/blog/high-concurrency-server-io-epoll-syscalls-zero-copy-aws/
datePublished: 2026-06-12T00:00:00.000Z
dateModified: 2026-10-28T00:00:00.000Z
author: palaniappan-p
category: Cloud Architecture
tags: engineering-guide, performance, ec2, aws
---

# High-Concurrency Server I/O: epoll, Syscalls, and Zero-Copy on AWS EC2

> C10k is solved until syscall overhead and context switches eat your Graviton cores. epoll, sendfile, and SO_REUSEPORT behaviors on EC2, and why Lambda caps concurrency differently.

As of June 2026, Linux on EC2 serves most self-hosted APIs. Event loops use epoll (edge-triggered) to watch thousands of sockets per thread. BSD and macOS dev laptops use kqueue instead, so production parity tests matter.

## Symptoms, mechanisms, and AWS controls

| Production symptom         | Mechanism                                      | AWS control                                             |
| -------------------------- | ---------------------------------------------- | ------------------------------------------------------- |
| CPU pegged at moderate QPS | Excessive syscall churn (read/write per chunk) | sendfile/splice zero-copy, larger socket buffers        |
| Connection accept backlog  | Single-threaded accept bottleneck              | SO_REUSEPORT multi-listener, ALB pre-connection scaling |
| epoll thundering herd      | All workers wake on single event               | EPOLLEXCLUSIVE or SO_REUSEPORT + per-core accept threads |

My recommendation: profile syscalls before rewriting in Rust. On AWS, ALB plus ENA instances with epoll edge-triggering handles most concurrency without kernel bypass.

> **Benchmark pattern (hypothetical workload):** c6gn.2xlarge (ENA enhanced) running epoll edge-triggered server, 48K concurrent connections, 0.8ms p99 accept-to-response, sendfile zero-copy reduces CPU 22% vs buffered I/O at 12K req/sec.

## Syscall and context switch tax

Each `read`/`write` crossing the user/kernel boundary costs CPU. Zero-copy `sendfile()` moves a file to a socket without user-space buffers, which suits static assets behind NGINX on EC2 (often superseded by CloudFront).

| Pattern                  | Benefit on AWS                                  |
| ------------------------ | ----------------------------------------------- |
| epoll + non-blocking I/O | Few threads, many connections on `c7g` Graviton |
| SO_REUSEPORT             | Accept queue spread across workers              |
| sendfile                 | Static file serving from EBS                    |

Lambda avoids your epoll tuning: concurrency is execution-environment reuse with platform limits, and CPU-bound work still belongs on EC2/ECS.

## AWS services map

| Need                       | Service                                 | Skip when                                    |
| -------------------------- | --------------------------------------- | -------------------------------------------- |
| High connection fan-in     | ALB (L7) or NLB (L4)                    | <1K connections on single instance           |
| Enhanced networking        | ENA-enabled instance types (c6gn, m6in) | Bursty Lambda with no persistent connections |
| Kernel bypass alternatives | Consider Nitro + DPDK only at 10Gbps+   | Standard web API under 1Gbps                 |

## When this advice breaks

- **TLS termination on instance:** crypto dominates, so offload to ALB/CloudFront first.
- **Tiny payloads:** syscall cost is noise next to business logic.

## What to do this week

1. Compare `nginx`/`envoy` worker count to CPU cores on origin ASG.
2. Profile with `perf top` during a load test and look for `syscall` hotspots.
3. Move static assets to S3+CloudFront; keep dynamic on epoll stack.

## More in this track

Part of the **Engineering Guides** library (June 2026).

- Previous: [Part 2](https://www.joinhorizon.co/blog/tls-13-handshake-internals-aws-alb-cloudfront-acm/)
- Next: [Part 4](https://www.joinhorizon.co/blog/cpu-memory-model-cache-coherence-false-sharing-cloud/)
- Browse tracks: [Engineering Guides hub](https://www.joinhorizon.co/resources/engineering-guides/)

## What this guide doesn't cover

CPU cache false sharing is covered in part 4 of this track.

## FAQ

### Should I rewrite the origin in Rust to fix syscall overhead?
Profile first. On AWS, ALB plus ENA instances with epoll edge-triggering covers most fan-in. sendfile/splice and larger socket buffers beat a language rewrite when CPU is spent in read/write per chunk.

### Does Lambda need epoll tuning?
No. Lambda concurrency is execution-environment reuse with platform limits. CPU-bound work still belongs on EC2 or ECS. Do not port epoll knobs into a function.

### When is kernel bypass (DPDK) the wrong next step?
Skip DPDK under about 10 Gbps. Standard web APIs should stay on epoll. TLS terminated on the instance also dominates, so offload to ALB or CloudFront first.

### What could go wrong with a single accept thread?
Accept backlog and thundering herd. Use SO_REUSEPORT multi-listeners (or EPOLLEXCLUSIVE on a shared listen socket), per-core accept threads, and let ALB pre-scale connections. Tiny payloads make syscall noise look like a business-logic bug.

---

*Source: https://www.joinhorizon.co/blog/high-concurrency-server-io-epoll-syscalls-zero-copy-aws/*