Nginx Requests/Second - Raspberry Pi vs. Amazon EC2 | Hacker News Reader

Nginx Requests/Second - Raspberry Pi vs. Amazon EC2

chimerasaurus.com

70 pointsmmastrac15 comments

The follow-up is much more interesting: http://www.chimerasaurus.com/2013/02/raspberry-pi-vs-aws-v2-...

However, the only thing this shows is that AWS micro instances are really slow. In the follow-up the author argues that while a small instance is much faster, its also much more expensive. While thats true i dont get the point, the pi is just not a good choice as a public webserver and if you were to use it that way you would need to co-locate it and costs would probably be comparable.

If on the other hand, you use both enviroments for heavy computing purposes it doesnt make sense to bill the EC2 instance with 24/7 as it probably would not need to be on all the time.

Duplicate: https://news.ycombinator.com/item?id=5083172

I'm really not sure why people keep using micro for benchmarks. Its CPU gets throttled to hell if it gets pretty much any load at all.

You're right. But... if you've ever used a Raspberry Pi as any kind of modern web server, you'll realize just how slow it is. Compile ruby on a modern PC vs a Raspberry, and it really forces you to remember that this is a tiny litte ARM device for fun projects + experiments, and not mean't for a rails stack (as an example).

My point is not to knock the Raspberry, rather, this demonstrates just how pathetic an Amazon Micro instance is. It's abysmal to the point that you have to ask yourself why they provide them at all. It's like they're teasing us, "Here, a free server! Enjoy it!" but in reality you'd rather drink a shot of drano full of nails because it's so painful to work with. Then again, beggars can't be choosers and free is free.

I don't think it's meant for production. I use mine for staging, ssh tunneling, hosting small projects with <1000 users, and to ssh to a Linux box I control and can run emacs on when I don't have my laptop.

I think you're missing the point of the micro instance. It can give you decent performance in short bursts, which is suitable for many light usage cases e.g. a low traffic blog. If you are going to load it via a benchmark you get very heavily throttled way below the performance you would get with light usage.

So one EC2 server might be drasticatlly slower than a raspberry pi, but isn't part of the attractiveness of cloud services like EC2 that there's replication and load balancing with no headaches? now if you build a modular, distributed load balancing replicated network of 10~20 raspberry pis, that all outperform the EC2 servers... that might warrant replacing EC2 with raspberry pis.

That actually sounds like a pretty awesome project... I might do that some day.

It's only drastically slower because the micro instance gets very heavily throttled when you put it under sustained load. If you use it for what it's designed for the performance is much better as you don't get throttled with light usage. The benchmark is of the slowest ec2 instance type while being heavily throttled due to the benchmark generating too much load.

If you need any additional proof that EC2 instances are abysmally slow (at all cost tiers) just look at http://www.techempower.com/benchmarks/#section=data-r4 and compare m1.large versus dedicated hardware.

I don't see how that proves anything. They're using an old first generation instance type and benchmarking it alongside a much more powerful i7 processor. If you want a faster instance get a second generation instance or a larger one.

Sadly the link to the followup (http://www.evilsoapbox.com/wp-admin/post.php?post=1612) fails with a database authentication error.

Here you go:

http://www.chimerasaurus.com/2013/02/raspberry-pi-vs-aws-v2-...

The dynamic link, ie. ?p=1234 has to be served dynamically but the static slug can be served by supercache or similar.

What I find far more fascinating:

    EXT3: 329.83 requests/sec
    EXT4: 484.27 requests/sec

ext4 is 30% faster than ext3? Really? What's going on there...

Strange indeed. Why is the disk accessed in the first place? For a site like this, I would expect all content to be cached in ram all the time.

Your math is a bit off. I think you mean 46% faster. That is surprising if consistently reproducible.

There is no meaningful data to confirm this, to say the least. http://dtrace.org/blogs/brendan/2012/10/23/active-benchmarki...

Reply on news.ycombinator.com