Sunday, 20 September 2015

Why commodity machines in a distributed cluster?



Cost comparisons of using server class machines and commodity machines have been worked out in the paper:

https://cs.uwaterloo.ca/research/tr/2009/CS-2009-09.pdf 

Basically for google-like usecase we calculate how many hard disks and compute power we need.
and we calculate the cost with both for server class and commodity machines.

STEP 1:

Calculating the size of the web and the size of search index, and estimate how many hard disks we need :




STEP 2:

Compare the costs for CPU and disks with server class machine and commodity machine:



No comments:

Post a Comment