Logging I/O Commands

I/O command logging can be toggled via the proc filesystem.
Logging is disabled by default. To enable logging, write any value to the debug entry and then run your benchmark:

# echo 1 > /proc/nvmev/profile
# ./run_your_benchmark.sh

After the benchmark completes, disable logging to retrieve the collected logs. When logging is disabled, NVMeVirt prints all buffered log entries to the kernel log.

# ./trace.sh

(in another terminal)
# echo 1 > /proc/nvmev/profile

Profiling

We currently provide two plotting scripts for analyzing I/O behavior.

• plot_access_pattern.py visualizes the access pattern of I/O commands.

  • X-axis: time
  • Y-axis: starting LBA

• plot_latency_distribution.py plots the latency distribution of read commands.

  • X-axis: latency ranges
  • Y-axis: number of commands in each range
  • For ease of analysis, the overall latency range is divided into 20 bins.

To run the profiling scripts:

$ python3 plot_access_pattern.py
$ python3 plot_latency_distribution.py