How to handle 1000's of concurrent users on a 360MB VPS | Hacker News Reader

How to handle 1000's of concurrent users on a 360MB VPS

markmaunder.com

8 pointsmmaunder6 comments

Use nginx? That is the sum total of your advice? What a complete waste of time.

This comment bugged me until I read the rest of your comments.

http://news.ycombinator.com/threads?id=evgen

It's simple but non-obvious advice. He also explains the reasoning why.

I had to do the same thing for Delicious, back in the day (except I used Pound[1], nginx was much less mature then, if it existed at all.)

[1] http://www.apsis.ch/pound/

I guess if you have no experience running a large site it might seem non-obvious, but this is not exactly rocket science. Every c10k article will talk about using a reverse proxy and why nginx is the best choice out there right now (varnish can sometimes deliver better throughput but it is finicky and for some reason certain configs will just cause it to explode, it also does not support some compression schemes that you might want to use.) Five years ago a scheme like this would have seemed obscure and a cool discovery, now it is just how things are done.

I am not sure when you last got your hands dirty at the pointy-end of the operations stick, but this is up there with tuning tcp buffers and keeping static content on another server in terms of being par for the course.

A more interesting explanation would have described some of the other options available and how and why you might choose one over the other, or maybe a useful trick or two: for example, if you need to track sessions by IP address using this basic setup you will need to drop the originating ip address into the header as it passes through nginx and then use mod_rpaf on the back-end apache server to reinject it so that your apache code does not even know that a reverse proxy is fronting its requests. I guess something a bit beyond "I used nginx to do what everyone claims it is good for and damn if it didn't work as advertised" would have been nice.

Use proxy_set_header? That's your more-interesting-explanation? You probably don't want to go "tracking sessions by IP address" or you might just think half of AOL is one session.

No, it is not really a more interesting explanation; I really only use it to serve as an additional flag to signal possible voting rings and sock-puppet activity -- either way sometimes the back-end servers really need or want to know the ip address of the original source.

It was mostly random tossed off example of something that might have been mentioned by you in addition to just saying "use nginx as a reverse-proxy." There is nothing wrong with stating this, but I can't see why you think this is something that is not a part of the general knowledge-base of this community. I realize that you want to pimp your blog, but you could have probably chosen something more useful or interesting to the community. At the very least you could have added more content so that the 99% of the community that already knew this would have been able to get something out of this or at least not felt like their time was wasted. At the very least you could have tightened up the headline slightly so that the fact that nginx was the solution to the problem posed was stated in the headline itself.

Reply on news.ycombinator.com