2014 03 23 00 00

For this run I attempted to substitute the c1.xlarge instance with an c3.2xlarge which on paper would be a slightly more performant substitute. Once that job crashed and burned I ran the job again with the old large configuration (12 machines) but I left the reducer count high. The job finished in 1 hour 45 min, which is a big speed-up over the 2 hour, 30 min that the job usually takes. (70% of the time!) and also it got us under an hour increment which saves money.