Hacker News Reader

Comment by wasyl | Hacker News Reader

I had no idea rails/django are sooooo slow

Rails IS slow, but part of the reason it shows so poorly on that tests is a weird glitch in how unicorn and wrk interact. In my testing for the Ruby Web Benchmark Report (http://www.madebymarket.com/blog/dev/ruby-web-benchmark-repo...) I ran into pretty bizarre low performance on unicorn and wrk.

TechEmpower uses Unicorn because it's the fastest MRI based rack server solution.

In JRuby, you have a few more high performance options, and Torquebox 4 is nearly as fast as Go in Hello World.

That being said, the only benchmark that matters is how your app performs on your tech stack. Anyone can write a slow DB query that ruins performance.

...which should tell you that speed really isn't as big a deal in web dev as some would have you believe. :)

It doesn't tell you that at all, since plenty of companies lose users/etc due to latency and other issues.

[citation needed]

I imagine that that issue is highly overrated for most companies.

You really want me to start citing research papers which show that if you improve latency, even if users don't "notice" it, you gain users?

"web latency research user conversion rate" on google will turn up a ton.

Heck, back in 2004, ten years ago, you only got two seconds: http://cba.unl.edu/research/articles/548/download.pdf

Now 1 second http://www.australianscience.com.au/research/google/34439.pd...

Even that causes losses: http://www.tagman.com/mdp-blog/2012/03/just-one-second-delay...

Google sees significant effects at 200ms: http://googleresearch.blogspot.com/2009/06/speed-matters.htm...

blah blah blah.

Feel free to find things that say the opposite, i've never seen a study that says "users will choose the slower service, if it's 'better', they just don't give a crap".

And sure there is a diminishing return, but if your latency times are 1s or greater for a web based service, you are almost 100% guaranteed to be losing users/conversions/etc.

From a psychological standpoint, a server that consistently responds quickly & doesn't break is easier to trust.

There are a ton of ways to optimize web apps without needing the application server to be the fastest. Caching, reverse proxy, CDN, scale-out, etc.

It's nice to see how fast Go is already. It'll probably be a couple years ago, but it should improve once they start working on the optimizing compiler.

People have already demonstrated how Go can be used on the server: http://blog.iron.io/2013/03/how-we-went-from-30-servers-to-2...

No, it doesn't imply anything at all about whether speed is a big deal or not.

At best one can conclude that concerns about performance aren't as important as other considerations which make Rails/Django more attractive. My opinion is that these other considerations are more along the lines of cost of labor and ease of learning than anything else.

Actually, performance is quite important for scalability. A lot of sites written in Ruby / PHP have big scalability pains if they get popular. Twitter had to be re-written from Ruby to Java. Facebook had to completely rewrite PHP's compiler to make it close to a compiled / statically typed language.

Joel Spolsky mentioned how StackOverflow only needed 3 servers to run it for a really long time, because it was coded in a compiled language. [1]

[1] http://youtu.be/pPJf8KrvJXU?t=22m33s

He says that at the 22m mark. And he says that currently he can fit his servers into 1 rack, while being a top 100 site.

That may be true, but the main point that's being ignored by such statements should always be mentioned: Django and the like in scripting languages offer an easy to use and fast to implement way of getting an idea off the ground. Once you really go beyond millions of requests, it should be also possible to hire Java developers then to rewrite the app for scaling beyond the tipping point of scripting language apps.

Reply on news.ycombinator.com