I'm not sure they're comparable. I did the same evaluation, one of the many criteria that I had is a system that can handle many, many data points from many hosts (at the peak we had 50 production machines) and provide robust flexible queries over them; with almost 0 maintenance/configuration. Metrics are created dynamically, the Graphite database is optimized for this problem (more optimized than RRD - although it seems that recent versions of RRD closed the gap), and it can take a beating in terms of scale. A single Graphite instance does this effortlessly.