Along the bottom of the graph are the Pool Size values. The Pool Size for each device ranges from “1GB” to the capacity of the device that was tested. Each Pool Size is marked by circle. If you click the circle a note will pop up displaying the milliseconds required to complete the read samples at that capacity point. There are a maximum of 13 Pool Sizes measured: 1GB, 2GB, 4GB, 8GB, 16GB, 32GB, 64GB, 128GB, 256GB, 512GB, 1TB, 2TB and 4TB. Only the D5 had capacity from 512GB to 4TB.
Along the vertical axis of the graph is the time, in milliseconds, that it took to complete a given Read Sample Size workload, at a given Pool Size.
Lets take the example of running the 128 MB Read Sample Size against the D5. That 128MB test was repeated multiple times at ever increasing Pool Sizes. So you will see several data points in a line very close to the bottom of the graph – THIS IS REALLY GOOD! The D5 shows very constant, low latency I/O performance even as the amount of storage being read increases. This shows how the D5 fully utilizes all of the storage available in parallel. We want to see these workload complete as fast as possible, or in other words, plot points closer to the bottom of the graph. 0 milliseconds (no time at all) would be ideal, and moving up the vertical axis means it’s taking you longer and longer to finish the read test.
In our 128MB Read Sample Size example for the D5, if you were to look at any of the Pool Sizes by clicking on a given circle you will see a nearly constant time to complete the tests!
It gets more interesting if you next select the DRAM data for the same workload. You will see that the DRAM takes a little less time, up to 16 GB (because that is all the DRAM we had to run the test against). DRAM is expected to be faster than the D5, but as you can see it isn’t THAT much faster. D5 random performance is more like DRAM than an SSD!
Turn on the SSD data for the same workload and you will see that the SSD takes 900 milliseconds – or nearly a second, to complete the Read Samples at various pool sizes.