Tuning Your Cloud: Improve Global Network Performance for Applications
summary
Explains concrete levers: use CloudFront edges reduce RTT to ~30ms (vs 150ms origin), enable 9K jumbo frames and Nitro-backed EC2 for up to 90 Gbps, 27 µs latency and >5M pps, and benchmark with iperf3 1GB: 150s -> 90s (55→90 Mbps) after Linux TCP buffer tuning but ~1,000 retransmits vs 0 default. Shows TCP startup, loss sensitivity, and kernel knobs to try in controlled tests rather than vague optimization advice.
AI-written summary · about
chapters (14)
- 0:00Intro
- 0:39Topics Understanding Application Performance: Why TCP Matters Choosing the right Cloud Architecture
- 1:43Mice: Short Connections
- 3:10Elephants: Long Connections
- 3:52The Transmission Control Protocol (TCP): Startup
- 5:51The Transmission Control Protocol (TCP): Growth
- 7:02The Transmission Control Protocol (TCP): Loss
- 7:38The Transmission Control Protocol (TCP): Recovery
- 9:29TCP Summary 1 Latency from User to Cloud is important Time to establish connection
- 10:23Solutions: Three things you can influence 1 Latency from your applications to your users: Service Architecture Region selection, use of edge services 2. Throughput from your infrastructure: Infrastructure Design
- 11:29Amazon CloudFront: Improving Latency to Users
- 14:55Why Packet Throughput Matters
- 16:56The Nitro System enables performance improvements
- 23:45Amazon Linux kernel TCP Tuning: 1GB Transfer
related talks
26mTuning Your Cloud: Improve Global Network Performance for Application
31mTuning Linux TCP for data-center networks
21mPerformance Consistency in the Cloud
45mReducing Network Latency: Innovations for a Faster Internet
16m"Performance Testing: ..."
01h 01mWeb Performance Deep Dive
31mA Practical, Research‑Backed Introduction To Cloud‑Native Kernel‑Bypass Networking
25mSilicon and Networking