new test tools to check thread-pool based disk i/o

This commit is contained in:
Paul Davis 2015-11-17 10:14:44 -05:00
parent 81103d822a
commit a95f8a8648
2 changed files with 475 additions and 0 deletions

89
tools/run-threadreadtest.sh Executable file
View File

@ -0,0 +1,89 @@
#!/bin/sh
dir=/tmp
filesize=100 # megabytes
numfiles=128
nocache=
interleave=
needfiles=1
write_blocksize=262144
args=
if uname -a | grep --silent arwin ; then
ddmega=m
else
ddmega=M
fi
while [ $# -gt 1 ] ; do
case $1 in
-d) dir=$2; shift; shift ;;
-f) filesize=$2; shift; shift ;;
-n) numfiles=$2; shift; shift ;;
-M) args="$args -M"; shift ;;
-D) args="$args -D"; shift ;;
-R) args="$args -R"; shift ;;
*) break ;;
esac
done
if [ -d $dir -a -f $dir/testfile_1 ] ; then
# dir exists and has a testfile within it - reuse to avoid
# recreating files
echo "# Re-using files in $dir"
needfiles=
else
dir=$dir/readtest_$$
mkdir $dir
if [ $? != 0 ] ; then
echo "Cannot create testfile directory $dir"
exit 1
fi
fi
if [ x$needfiles != x ] ; then
echo "# Building files for test..."
if [ x$interleave = x ] ; then
#
# Create all files sequentially
#
for i in `seq 1 $numfiles` ; do
dd of=$dir/testfile_$i if=/dev/zero bs=1$ddmega count=$filesize >/dev/null 2>&1
done
else
#
# Create files interleaved, adding $write_blocksize to each
# file in turn.
#
size=0
limit=`expr $filesize * 1048576`
while [ $size -lt $limit ] ; do
for i in `seq 1 $numfiles` ; do
dd if=/dev/zero bs=$write_blocksize count=1 >> $dir/testfile_$i 2>/dev/null
done
size=`expr $size + $write_blocksize`
done
fi
fi
for bs in $@ ; do
if uname -a | grep --silent arwin ; then
# clears cache on OS X
sudo purge
elif [ -f /proc/sys/vm/drop_caches ] ; then
# Linux cache clearing
echo 3 | sudo tee /proc/sys/vm/drop_caches >/dev/null
else
# need an alternative for other operating systems
:
fi
echo "# Blocksize $bs"
./thread_readtest $args -b $bs -q $dir/testfile_%d
done

386
tools/thread_readtest.cc Normal file
View File

@ -0,0 +1,386 @@
/* g++ -o thread_readtest thread_readtest.cc `pkg-config --cflags --libs glibmm-2.4` -lm */
#ifndef _WIN32
# define HAVE_MMAP
#endif
#include <stdlib.h>
#include <errno.h>
#include <stdio.h>
#include <unistd.h>
#include <stdint.h>
#include <string.h>
#include <getopt.h>
#include <fcntl.h>
#include <math.h>
#undef HAVE_MMAP
#ifdef HAVE_MMAP
# include <sys/stat.h>
# include <sys/mman.h>
#endif
#include <glibmm.h>
char* data = 0;
void
usage ()
{
fprintf (stderr, "thread_readtest [ -b BLOCKSIZE ] [ -l FILELIMIT] [ -n NTHREADS ] [ -D ] [ -R ] [ -M ] filename-template\n");
}
Glib::Threads::Cond pool_run;
Glib::Threads::Cond pool_done;
Glib::Threads::Mutex pool_lock;
std::list<int> pool_work;
std::vector<Glib::Threads::Thread*> thread_pool;
int pool_errors = 0;
bool thread_pool_lives = true;
struct ThreadData {
int id;
char *data;
size_t block_size;
};
void
thread_pool_work (ThreadData* td)
{
Glib::Threads::Mutex::Lock lm (pool_lock);
while (thread_pool_lives) {
pool_run.wait (pool_lock);
if (!thread_pool_lives) {
return;
}
/* we're awake ... get some work */
while (!pool_work.empty()) {
int file_descriptor = pool_work.front ();
pool_work.pop_front ();
/* release the lock while we do work */
lm.release ();
/* do the work */
int err = 0;
ssize_t nread;
again:
if ((nread = ::read (file_descriptor, td->data, td->block_size)) != td->block_size) {
if (nread != 0 && errno == EAGAIN) {
fprintf (stderr, "read requires retry\n");
goto again;
}
if (nread != 0) {
fprintf (stderr, "thread %d has error = %s\n", td->id, strerror (errno));
}
err++;
}
/* reacquire lock so that we can check the status of
* things and possibly wake the master.
*/
lm.acquire ();
pool_errors += err;
if (pool_work.empty()) {
/* work is finished, tell the master */
pool_done.signal ();
}
}
}
return;
}
void
stop_thread_pool ()
{
{
Glib::Threads::Mutex::Lock lm (pool_lock);
thread_pool_lives = false;
pool_work.clear ();
pool_run.broadcast ();
}
/* XXX wait for each thread to finish */
}
void
build_thread_pool (int nthreads, size_t block_size)
{
for (int n = 0; n < nthreads; ++n) {
ThreadData* td = new ThreadData;
td->data = (char*) malloc (sizeof (char) * block_size);
td->block_size = block_size;
td->id = n;
thread_pool.push_back (Glib::Threads::Thread::create (sigc::bind (sigc::ptr_fun (thread_pool_work), td)));
}
}
int
run_thread_pool (int* files, int nfiles)
{
Glib::Threads::Mutex::Lock lm (pool_lock);
/* Queue up all the files */
for (int n = 0; n < nfiles; ++n) {
pool_work.push_back (files[n]);
}
pool_errors = 0;
/* wake everybody up */
pool_run.broadcast ();
/* wait for everyone to finish */
pool_done.wait (pool_lock);
if (pool_errors) {
return -1;
}
return 0;
}
int
main (int argc, char* argv[])
{
int* files;
char optstring[] = "b:DRMl:q";
uint32_t block_size = 64 * 1024 * 4;
int max_files = -1;
int nthreads = 16;
#ifdef __APPLE__
int direct = 0;
int noreadahead = 0;
#endif
#ifdef HAVE_MMAP
int use_mmap = 0;
void **addr;
size_t *flen;
#endif
const struct option longopts[] = {
{ "blocksize", 1, 0, 'b' },
{ "direct", 0, 0, 'D' },
{ "mmap", 0, 0, 'M' },
{ "noreadahead", 0, 0, 'R' },
{ "limit", 1, 0, 'l' },
{ "nthreads", 16, 0, 'n' },
{ 0, 0, 0, 0 }
};
int option_index = 0;
int c = 0;
char const * name_template = 0;
int flags = O_RDONLY;
int n = 0;
int nfiles = 0;
int quiet = 0;
while (1) {
if ((c = getopt_long (argc, argv, optstring, longopts, &option_index)) == -1) {
break;
}
switch (c) {
case 'b':
block_size = atoi (optarg);
break;
case 'l':
max_files = atoi (optarg);
break;
case 'D':
#ifdef __APPLE__
direct = 1;
#endif
break;
case 'M':
#ifdef HAVE_MMAP
use_mmap = 1;
#endif
break;
case 'R':
#ifdef __APPLE__
noreadahead = 1;
#endif
break;
case 'q':
quiet = 1;
break;
case 'n':
nthreads = atoi (optarg);
break;
default:
usage ();
return 0;
}
}
if (optind < argc) {
name_template = argv[optind];
} else {
usage ();
return 1;
}
while (1) {
char path[PATH_MAX+1];
snprintf (path, sizeof (path), name_template, n+1);
if (access (path, R_OK) != 0) {
break;
}
++n;
if (max_files > 0 && n >= max_files) {
break;
}
}
if (n == 0) {
fprintf (stderr, "No matching files found for %s\n", name_template);
return 1;
}
if (!quiet) {
printf ("# Discovered %d files using %s\n", n, name_template);
}
nfiles = n;
files = (int *) malloc (sizeof (int) * nfiles);
#ifdef HAVE_MMAP
if (use_mmap) {
if (!quiet) {
printf ("# Using mmap().\n");
}
addr = malloc (sizeof (void*) * nfiles);
flen = (size_t*) malloc (sizeof (size_t) * nfiles);
}
#endif
for (n = 0; n < nfiles; ++n) {
char path[PATH_MAX+1];
int fd;
snprintf (path, sizeof (path), name_template, n+1);
if ((fd = open (path, flags, 0644)) < 0) {
fprintf (stderr, "Could not open file #%d @ %s (%s)\n", n, path, strerror (errno));
return 1;
}
#ifdef __APPLE__
if (direct) {
/* Apple man pages say only that it returns "a value other than -1 on success",
which probably means zero, but you just can't be too careful with
those guys.
*/
if (fcntl (fd, F_NOCACHE, 1) == -1) {
fprintf (stderr, "Cannot set F_NOCACHE on file #%d\n", n);
}
}
if (noreadahead) {
if (fcntl (fd, F_RDAHEAD, 0) == -1) {
fprintf (stderr, "Cannot set F_READAHED on file #%d\n", n);
}
}
#endif
files[n] = fd;
#ifdef HAVE_MMAP
if (use_mmap) {
struct stat s;
if (fstat (fd, & s)) {
fprintf (stderr, "Could not stat fd #%d @ %s\n", n, path);
return 1;
}
if (s.st_size < block_size) {
fprintf (stderr, "file is shorter than blocksize #%d @ %s\n", n, path);
return 1;
}
flen[n] = s.st_size;
addr[n] = mmap (0, s.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
if (addr[n] == MAP_FAILED) {
fprintf (stderr, "Could not mmap file #%d @ %s (%s)\n", n, path, strerror (errno));
return 1;
}
}
#endif
}
data = (char*) malloc (sizeof (char) * block_size);
uint64_t _read = 0;
double max_elapsed = 0;
double total_time = 0;
double var_m = 0;
double var_s = 0;
uint64_t cnt = 0;
build_thread_pool (nthreads, block_size);
while (1) {
gint64 before;
before = g_get_monotonic_time();
if (run_thread_pool (files, nfiles)) {
fprintf (stderr, "thread pool error\n");
goto out;
}
_read += block_size;
gint64 elapsed = g_get_monotonic_time() - before;
double bandwidth = ((nfiles * block_size)/1048576.0) / (elapsed/1000000.0);
if (!quiet) {
printf ("# BW @ %lu %.3f seconds bandwidth %.4f MB/sec\n", (long unsigned int)_read, elapsed/1000000.0, bandwidth);
}
total_time += elapsed;
++cnt;
if (max_elapsed == 0) {
var_m = elapsed;
} else {
const double var_m1 = var_m;
var_m = var_m + (elapsed - var_m) / (double)(cnt);
var_s = var_s + (elapsed - var_m) * (elapsed - var_m1);
}
if (elapsed > max_elapsed) {
max_elapsed = elapsed;
}
}
out:
if (max_elapsed > 0 && total_time > 0) {
double stddev = cnt > 1 ? sqrt(var_s / ((double)(cnt-1))) : 0;
double bandwidth = ((nfiles * _read)/1048576.0) / (total_time/1000000.0);
double min_throughput = ((nfiles * block_size)/1048576.0) / (max_elapsed/1000000.0);
printf ("# Min: %.4f MB/sec Avg: %.4f MB/sec || Max: %.3f sec \n", min_throughput, bandwidth, max_elapsed/1000000.0);
printf ("# Max Track count: %d @ 48000SPS\n", (int) floor(1048576.0 * bandwidth / (4 * 48000.)));
printf ("# Sus Track count: %d @ 48000SPS\n", (int) floor(1048576.0 * min_throughput / (4 * 48000.)));
printf ("# seeks: %llu: bytes: %llu total_time: %f\n", cnt * nfiles, (nfiles * _read), total_time/1000000.0);
printf ("%d %.4f %.4f %.4f %.5f\n", block_size, min_throughput, bandwidth, max_elapsed/1000000.0, stddev/1000000.0);
}
return 0;
}