Are We Really Talking About Commodity Hardware When Working With MongoDB?

Last couple of months I have been reading, learning, playing with MongoDB and one thing that I have read or found myself is that its performance depends largely on the amount of RAM in your system. As a general rule larger the RAM better the performance which I can easily understand as you are not hitting disk so you get great performance. When we talk about commodity hardware I think we talk about 4GB or at max 8 GB RAM boxes which means if your application working set can fit in 4GB or 8 GB RAM you are good otherwise your performance will suffer. Then you have two choices either add more RAM or horizontally scale your system i.e Sharding. To me adding more RAM means you are moving away from commodity hardware and moving toward big costly boxes. So we should horizontally scale our system by adding more 4 GB or 8GB RAM boxes. Correct??

I thought companies or people who are using MongoDB would have been following this approach i.e. they are using commodity boxes and scaling their systems. But I was wrong. Most of presentations (from companies like Craiglist and ForeSquare) that I saw are using big 64 GB or more RAM, faster disks. So where are we talking about commodity hardware?