Shaker performance results after applying the fix, measured instance-to-instance bandwidth in the same L2 domain (instances hosted on different compute nodes): * 1 thread - 3.33 Gb/s * 2 threads - 2 x 2.9 Gb/s = 5.4 Gb/s * 4 threads - 4 x 2.1 Gb/s = 8.4 Gb/s * 6 threads - 6 x 1.35 Gb/s = 8.1 Gb/s (full report attached)
Shaker performance results after applying the fix, measured instance- to-instance bandwidth in the same L2 domain (instances hosted on different compute nodes):
* 1 thread - 3.33 Gb/s
* 2 threads - 2 x 2.9 Gb/s = 5.4 Gb/s
* 4 threads - 4 x 2.1 Gb/s = 8.4 Gb/s
* 6 threads - 6 x 1.35 Gb/s = 8.1 Gb/s
(full report attached)