Hitachi Content Platform Scale Testing
by Michael Hay on Jun 6, 2010
We just completed a portion of our scale testing on the Hitachi Content Platform and I’m pleased to say that we were able to fill the system to a billion objects. Some characteristics of the system tested are:
- HCP V3.1
- 35 node HCP300 based system
- DPL2 Configuration
- 5 total clients, with each client driving 8 nodes in the cluster, except for the 5th client which drives 3 nodes
- Each client drives 240 threads across all 8 nodes for an average of 30 threads/node
- Each object is 1KB in size
- ~2200 objects/second
For this characterization we were interested in the overall system behavior during a large scale data load. We weren’t specifically testing for performance, therefore performance observations are a side effect. However, it is important to note that as the system scales we can observe a straight line or linear scaling in performance (see the graph at the end which is a snapshot of the test output).
The most significant take aways are:
- A tested billion object cloud storage system, note we halted the test due to scheduling (meaning there is obviously more headroom in capacity growth)
- With a sustained average object/second behavior over this relatively large object count
