3 ms·
> The day healthcare.gov launched, six people total got through. Six. Nobody even knew when the site was down, except by checking CNN. Two months later, Mikey h
by tzs 12y ago
> The day healthcare.gov launched, six people total got through. Six. Nobody even knew when the site was down, except by checking CNN. Two months later, Mikey had rallied the 20-something contractors into punching through dozens of problems, and had recruited a team of (now-) 30-something engineers from Google, Facebook, YC startups, and other fine places to come in on rotations to keep the momentum going. As a result, 8 million people got health insurance.
Out of curiosity, how much of the site's problems at the start were due to high load and how much would have been there even if it had seen much lower traffic? In particular, how would it have fared with the original design and code if only a handful of states had refused to implement their own exchanges, instead of 27 states refusing?
- brandonb 12y agoHard to know for sure, but most of the problems would have been exposed at even moderate amounts of load--there was no load testing or performance monitoring in place initially.
- dschiptsov 12y agoJava in production, especially with heavy "frameworks" use and "waterfall to subcontracted coders-for-paycheck" process (how else there could be a billion lines of code for a site) can't handle even a moderate load and runs out of memory on handling persistent connections (long sessions) and the fixes were mostly by adding tons of hardware (unlimited government money) and to redesign many components to share nothing (decoupling) and to work in asynchronous way with a short requests (service-oriented architecture + Erlang's process-based model) by Google, FB, etc. guys? That's the story? btw, these are very legitimate questions. The original project was a epic failure which perfectly illustrated all the flaws of past-century practices (waterfall process with tons of specs created by idiots, outsourcing to the cheapest or favorite incompetent, blind faith in big names, etc.) and the fix, it seems, was to bring it close to the state-of-the-art tech which was evolved to use-at-home in [more-or-less] independent-thinking shops, like Google or FB.