Ask HN: How to deal with the C10K problem with a simple script? | Hacker News Reader

Ask HN: How to deal with the C10K problem with a simple script?

9 pointsjbm3 comments

I was asked by a client to setup a survey site for Japanese mobile phones that would be displayed at a major event that would be taking place at the Tokyo Dome. The problem is that I'll be dealing with 10k users connecting to the site at the same time, and that the site will only be up for a day.

The client says that he wants no connections dropped, and to prepare the infrastructure so that it can handle it. (My initial idea was to try to find a web service to handle it, but even the best could only deal with something like 150-200 simultaneous connections.)

My idea is to setup an Amazon EC2 instance with nginx or Lighttpd and to make a simple Ruby or PHP script to handle the form that is submitted, and store the submissions to the Amazon SimpleDB.

Unfortunately, I've never dealt with scaling and load balancing for something like this, and I can't find any information about the kind of load I am looking at. Can anyone offer any advice on how to handle scaling with EC2 or any other cloud service?

You're overthinking this.

The client doesn't mean that 10k people are going to hit the site at the same microsecond. The client means that there will be 10,000 eventgoers accessing the site within a particular period. Many of the clients in this industry, being cough fundamentally non-technical people, believe that the period of reference is "one day".

As you've probably noticed, a $20 a month VPS plus Apache can easily chew through 10k visitors in a day (I wouldn't use Apache personally, but if I were competent at configuring it, yeah, no problem whatsoever). Adjust upwards if your client anticipates a particular instantaneous deluge (think "Steve Jobs stands up at the keynote and says 'My robot minions, all of you need to hit this URL right now!').

P.S. Your client is wildly overestimating the audience that will actually come to the site. Don't tell him I said that.

This isn't the C10k problem. The C10k problem is "how do you handle 10k mostly idle connections" -- the situation where per-concurrent-connection costs matter (e.g., you really don't want to have 10k processes, each handling a single connection).

Your problem is simply "how do I handle huge amounts of traffic". The first half of the answer, as you guessed, is to use EC2; the second half is to throw traffic at it and make sure that you get the response (whether it's in being able to handle the necessary number of requests or scaling automatically) that you want.

The middle half is to make your code as efficient as possible; I would not recommend storing directly to SimpleDB. Just log form submissions locally and then process them as a batch.

I have to agree with this. If you have to process 10k submissions in a very short space of time, why have overhead of storing it in the db right then and there when you can asynchronously process it later.

If you want something interesting to work on, you could have nginx for the front end (faster than lighttpd), accept a form submission which gets written to disk and maybe pop an id on a key-value store like redis, and another poller app in ruby which pops id's off the queue and processes the files on disk.

Reply on news.ycombinator.com