Benchmark 2 - 6x faster than Hadoop 1.0.1
Benchmark 3 - 10x fewer CPU cores
This is much more representative measure of the true workloads found in the enterprise. This benchmark, the Statistical workload Injector for MapReduce (SWIM), was designed by a team in University of California, Berkeley working in co-operation with Facebook. It takes the Facebook workloads from 2009 and 2010 and uses these real workloads to compare the performance of the Hadoop distributions. This test was conducted by an independent body, STAC research. The results they shared with IBM showed that Platform Symphony was on average 6x faster vs Hadoop 1.0.1.
This was the standard Terasort benchmark. Terasort measures the sort performance for Big Data workloads. For this comparison we were only interested in the comparison to other Hadoop implementations that had executed the test. For this reason, we used the same benchmark that Yahoo had executed in their ground breaking result using Hadoop to establish a new Terasort record. In testing with this benchmark Platform Symphony along with Infosphere BigInsights, was able to run the 100TB data set Terasort in about the slightly less time than Yahoo did but by using 10 times fewer CPU cores. This resulted in a huge infrastructure savings.