And it's running as a single process on a single server without a database storing everything on the file system. That it's doing the traffic it is I find quite impressive.
Technically speaking, I find hacker news persistent strategy one of the most interesting things about the implementation.
I use a similar strategy for my blog and just playing with larger datasets I've certainly run into hard limits on what seems to be acceptable.
I am quite interested in the architecture, because I found this approach nice for some sites I was building. They were not very large though.
File system stores aren't that uncommon (aside from the VSAM days, even in RDBMS times+) - it's a common approach for Wiki implementations too.
Why?
I can understand not using a database if this was improving the performance.