Building Large Systems at Google
I watched this video very late last night–way cool. It’s a Google TechTalk about building very large systems.
A few things stand out to me:
- How a large cluster of commodity machines (1000+) can function as a single large fault-tolerant hard drive (GFS)
- How a large cluster of commodity machines can also function as a single large distributed computation grid
I’ve often read people talking about ultra-large scale systems where the big clusters function as a single large computer, but this video brought that point home.
It’s pretty cool to think they have abstracted a cluster as a hard drive, then abstracted on top of that to create a huge virtual machine for solving problems (executing algorithms).