benchmp, benchmp_getstate, benchmp_interval, start, stop, get_n,
set_n, gettime, settime, get_enough, t_overhead, l_overhead - the lmbench
timing subsystem
#include
typedef u_long iter_t;
typedef (*bench_f)(iter_t iterations, void* cookie);
typedef (*support_f)(iter_t iterations, void* cookie);
void benchmp(support_f initialize, bench_f benchmark, support_f
cleanup, int enough, int parallel, int warmup, int repetitions, void*
cookie);
void* benchmp_getstate();
iter_t benchmp_interval(void* state);
void start(struct timeval *begin);
uint64 stop(struct timeval *begin, struct timeval
*end);
uint64 get_n();
void set_n(uint64 n);
uint64 gettime();
void settime(uint64 u);
uint64 get_enough(uint64 enough);
uint64 t_overhead();
double l_overhead();
The single most important element of a good benchmarking system is
the quality and reliability of its measurement system. lmbench's
timing subsystem manages the experimental timing process to produce accurate
results in the least possible time. lmbench includes methods for
measuring and eliminating several factors that influence the accuracy of
timing measurements, such as the resolution of the system clock.
lmbench gets accurate results by considering clock
resolution, auto-sizing the duration of each benchmark, and conducting
multiple experiments.
- void benchmp(initialize,
benchmark, cleanup, enough, parallel, warmup, repetitions,
cookie)
- measures the performance of benchmark repeatedly and reports the
median result. benchmp creates parallel sub-processes which
run benchmark in parallel. This allows lmbench to measure the
system's ability to scale as the number of client processes increases.
Each sub-process executes initialize before starting the
benchmarking cycle. It will call benchmark several times in order
to collect repetitions results. After all the benchmark results
have been collected, cleanup is called to cleanup any resources
which may have been allocated by initialize or benchmark .
cookie is a void pointer to a hunk of memory that can be used to
store any parameters or state that is needed by the benchmark.
- void
benchmp_getstate()
- returns a void pointer to the lmbench-internal state used during
benchmarking. The state is not to be used or accessed directly by clients,
but rather would be passed into benchmp_interval.
- iter_t
benchmp_interval(void* state)
- returns the number of times the benchmark should execute its benchmark
loop during this timing interval. This is used only for weird benchmarks
which cannot implement the benchmark body in a function which can return,
such as the page fault handler. Please see lat_sig.c for sample
usage.
- void start(struct
timeval *begin)
- starts a timing interval. If begin is non-null, save the start time
in begin .
- uint64 stop(struct
timeval *begin, struct timeval *end)
- stops a timing interval, returning the number of elapsed
micro-seconds.
- uint64
get_n()
- returns the number of times loop_body was executed during the
timing interval.
- void set_n(uint64
n)
- sets the number of times loop_body was executed during the timing
interval.
- uint64
gettime()
- returns the number of micro-seconds in the timing interval.
- void settime(uint64
u)
- sets the number of micro-seconds in the timing interval.
- uint64
get_enough(uint64 enough)
- return the time in micro-seconds needed to accurately measure a timing
interval.
- uint64
t_overhead()
- return the time in micro-seconds needed to measure time.
- double
l_overhead()
- return the time in micro-seconds needed to do a simple loop.
There are three environment variables that can be used to modify
the lmbench timing subsystem: ENOUGH, TIMING_O, and LOOP_O. The
environment variables can be used to directly set the results of
get_enough , t_overhead , and l_overhead . When running
a large number of benchmarks, or repeating the same benchmark many times,
this can save time by eliminating the necessity of recalculating these
values for each run.
Development of lmbench is continuing.
lmbench(8), lmbench(3), reporting(3), results(3).
Carl Staelin and Larry McVoy
Comments, suggestions, and bug reports are always welcome.