Andrius Virbičianskas

It’s been quite a while since I was first interested in Node.js – a server side javascript for scalable network applications. But had no time or excuse to actually try and built something useful with it. Until recently, that is.

Some time ago adclick.lt, an ad delivery network I support, started struggling with available server resources. We had two choices – either get additional server or try to optimize our code. The second choice worked for a while until there was almost no room for improvements. As it turns out, apache + php (php-fpm didn’t help either) can’t handle complex real-time calculations with more than 400 requests per second.

This was the moment where I started looking into alternatives. At first G-WAN was a possible solution. But the lack of documentation, proper JAVA (learning C was out of consideration) support and author’s strange world view quickly made me change my mind. And thus, I began porting adclick.lt engine code to Node.js. Some time later, the engine was ready. At that moment I was not sure if I will gain any performance boost. Javascript isn’t terribly fast language, we all know that. But I hoped, that event driven Node.js web server would help a lot.

CPU. New engine: @19 d. - 50%, @23 d. - 100%

And it did (see graphs, I think they speak for themselves). Initial benchmarks showed that the new engine handled 4 to 5 times more requests per second. I was really satisfied with the result. One may ask, however, how’s that possible when engine relies on complex calculations? Asynchronous I/O operation handling shouldn’t improve that much, should it? Well, it does. In addition to that, persistent in memory data holding completely removes memcached requirement which, according to xdebug profiles, ate up to 30-40% of a request in the old, php based, engine.

There are some gotchas, though. More than on one occasion I found myself solving race conditions. That’s what happens, when you come from synchronous development camp. There’s also the stability issue – Node.js is fairly new (relatively). But so far so good, no problems related to it’s core. This can’t be said about community created packages, however. Because while there’s a huge number of them, most of them are either forgotten or with large lists of issues. Navigating and finding one that suits one’s needs is a separate and a no small task at that.

LOADS. New engine: @19 d. - 50%, @23 d. - 100%

Overall, I can only say, that I just love Node.js. The philosophy behind it hits my way of thinking spot on. It’s just like Ryan Dahl, the author of this creation once said: “The amount of complexity I’m willing to tolerate is proportional to the size of the problem being solved.” I think this is the main reason why coding for Node.js was such a satisfying experience. Add the adclick.lt performance issues being solved and what else could one want?

No related posts.