simoop: get rid of rw_threads, make defaults saner
simoop was originally meant to replicate what java was doing with
hadoop, so it sprayed new threads all over the place for little to no
reason. This isn't ideal for careful timing, so instead do reads and
writes directly from the worker threads.
Before, read/write latencies covered the whole time it took to
pthread_create(), do 128 IOs, pthread_join(). Now the read/write
latencies cover the time required for a single read or write on a single
file.
The timings do include the fadvise(DONT_NEED) or any sync_file_range()
from write_through mode.
This commit also makes the defaults for think time and memory hogging 0.
The idea is to start benchmarking just the IO.
Signed-off-by: Chris Mason <clm@fb.com>
1 file changed
