Finish up the testing in lustre. Included should be

• times for directory creation for each write
• times for interleaved reads and writes. This should emulate a common usage pattern better than, say, simply doing all the writes then all the reads.
• Runs for 10^5 files at 10^6 bytes, 10^4 files at 10^7 bytes and 10^3 files at 10^8 bytes
• Averages for all operations in addition to raw times.
• Try to determine the "sweet spot" for directory depth. This will consist of running sets (10^4 files or so??) with different depths and plotting the times against each others. Conjecture: There is a bell curve with optimal times at depth of 3 or 4.

Final reports on reads and writes should consist of charts of files against time with each run.