YouPorn made a big-ish article about their stack, ten years ago or so. The backend in particular was of interest for me. For that time when clusters just starte

YouPorn made a big-ish article about their stack, ten years ago or so. The backend in particular was of interest for me. For that time when clusters just started getting popular, they were rather exceptional in their scale and thus performance requirements: notably they put HAProxy in front of multiple replicas of Redis, for the throughput—while for me Redis was the second-fastest thing around after Memcached. Also apparently they were massively into using syslogd for logging and statistics—with a multi-tier setup where nodes aggregate entries from other ones. Again, that's before logstash and such plus Elasticsearch became widespread.