3 ms·
This article comes across as being really fake. You can clearly tell that the goal is to to drive additional traffic to their v1.0001 instead of actually diving
by jteppinette 7y ago
This article comes across as being really fake. You can clearly tell that the goal is to to drive additional traffic to their v1.0001 instead of actually diving into what they did wrong. Were they really investigating an entire new DB and user interface in the few days after launch? If your initial design is so far off that it requires a complete rewrite on day 1 of launch, then you have serious problems. That is not “avoiding pre-mature optimization”. That is an architectural failure.
- amirathi 7y ago> That is an architectural failure. Let's not jump the guns too soon. Analytic products are notoriously hard from scaling perspective. UI, APIs, Infra that works crisply for 1x data, starts to become sluggish at 10x and really breaks down at 100x. Startups solve the 100x scale problem when they get there (sensibly so). In this case, they got there way too quickly thanks to HN traffic.
- jteppinette 7y agoIMO, they probably failed way before 100x expected usage. It was probably before 10x expected usage. This isn’t crazy. This is a problem that happens for a lot of teams dealing with scale. You engineer for 1x (checking off requirements) and anything over that tears everything to shreds. The difference between developing a solution on your local and hitting your PERF environment is massive. e.g. I can’t run a TB DB on my local, so... if I don’t see queries running in less that 10ms-100ms, I basically know that the queries are never going to complete in PROD. Scale is hard.
- volument 7y agoYes. We really made a bad architectural decision and are now suffering from it. We thought the first version was good enough for the public eye.
- cheez 7y agoAnd... So it was a failure... So they redid the work.
- ykevinator 7y agoRight, it's a good thing not a bad thing.