Mapping the Cloud with Elixir
summary
Elixir-based probes run across 121 datacenters on AWS, GCP and Azure, collecting about 6 million traceroutes over two months to map datacenter-to-datacenter paths; architecture uses a central command-and-control, probes that check in for IP lists, Terraform/Terragrunt for provisioning, spot instances to cut costs, and burrito-built releases to avoid libc mismatches. Key results: AWS yields near-complete ICMP hop visibility, GCP filters many ICMP responses, Azure has inconsistent VM sizes and tooling, and total cloud cost averages ~$738/month with IPv4 addressing making up a large share.
AI-written summary · about
chapters (16)
- 0:00Introduction
- 1:066M traceroutes visualized
- 2:33How traceroute works
- 5:47ICMP packets explained
- 8:59Elixir’s binary matching
- 11:46Sending packets with Erlang
- 13:31Raw vs datagram sockets
- 16:37Probes in 121 data centers
- 18:05Burrito for portable releases
- 20:33VMs with Terraform/Terragrunt
- 25:30Monitoring with Sentry
- 26:52Cost cuts with spot instances
- 29:21Cloud cost comparisons
- 31:07AWS vs GCP vs Azure
- 34:33Final demo insights
- 36:38Reliability metrics
related talks
29mFinding the Needle in a Haystack: Tenant-Level Network...
35mPCEI at the EDGE
46mLessons learned from monitoring a large production network
25mMeasuring the Internet with Python
14mSkyline: A Cloud Centric Internet Monitoring Engine
47mFast and Furious Multi-Cloud Connectivity
32mActive Fault Finding in Networks
26mTuning Your Cloud: Improve Global Network Performance for Application