Hadoop Performance on vSphere 5.1
We’ve just published a third Hadoop performance paper, written by VMware performance expert Jeff Buell, which looks in detail at the relative performance of a bare-metal 32-node Hadoop cluster compared to a range of virtual clusters with up to 128 VMs. The executive summary is that while we saw a 13% performance degradation in a head-to-head comparison of a 32-node physical cluster against a 32-VM virtual cluster (one VM per host) running on the same hardware and running the same tests, virtualized performance can be increased significantly — to the point where virtualized Hadoop actually runs a bit faster than physical — by increasing the number of VMs per host. We’ve seen this effect before with Hadoop and with other resource-intensive HPC applications.
Read the full paper for detailed results and to learn about performance best practices for deploying Hadoop on vSphere.