I've been toying around with fio's output for a bit, mostly because Intel's fiovisualizer [0] was the first graphing tool I found for fio, even though gfio exists. The way fiovisualizer works is it reads terse ouptut, parses that, and plots data. gfio on the other hand is a proper built-in client, where communication is exchanged between client/server and the client plots the data. If you end up using scraper for terse data such as fiovisualizer with the prospect to scrape data you will end up with different results if you use a remote server as a backend. While studying fio code I realized that the way fiovisualizer implements graphing is perhaps the wrong / least optimal approach, but also that fio terse output needed a bit of love for client/server relationship to keep fidelity in terms of the expected output. These are just two small enhancements which I hope wil make things clearer. [0] https://github.com/intel/fiovisualizer Luis Chamberlain (2): client: respect terse output on client <--> backend relationship init: add semantics for all types of backends running backend.c | 2 ++ client.c | 16 +++++++++++++--- diskutil.c | 2 +- fio.h | 8 ++++++++ init.c | 1 + stat.c | 4 +++- 6 files changed, 28 insertions(+), 5 deletions(-) -- 2.18.0
