No comment on this? Was it just that I posted it during the VM summit?
On Mon, Aug 09, 2010 at 09:30:00AM -0400, Matthew Wilcox wrote:
>
> This testcase shows some odd behaviour from the Linux VM.
>
> It creates a 1TB sparse file, mmaps it, and randomly reads locations
> in it. Due to the file being entirely sparse, the VM allocates new pages
> and zeroes them. Initially, it runs very fast, taking on the order of
> 2.7 to 4us per page fault. Eventually, the VM runs out of free pages,
> and starts doing huge amounts of work trying to figure out which of
> these clean pages to throw away. In my testing with a 6GB machine
> and 2.9GHz CPU, one in every 15,000 page faults takes over a second,
> and one in every 40,000 page faults take over seven seconds!
>
> This test-case demonstrates a problem that occurs with a read-mostly
> mmap of a file on very fast media. I wouldn't like to see a solution
> that special-cases zeroed pages. I think userspace has done its part
> to tell the kernel what's it's doing by calling madvise(MADV_RANDOM).
> This ought to be enough to hint to the kernel that it should be eagerly
> throwing away pages in this VMA.
>
>
> /*
> * Copyright (c) 2010, Intel Corporation
> * All rights reserved.
> *
> * Redistribution and use in source and binary forms, with or without
> * modification, are permitted provided that the following conditions are met:
> *
> * * Redistributions of source code must retain the above copyright notice,
> * this list of conditions and the following disclaimer.
> * * Redistributions in binary form must reproduce the above copyright notice,
> * this list of conditions and the following disclaimer in the documentation
> * and/or other materials provided with the distribution.
> * * Neither the name of Intel Corporation nor the names of its contributors
> * may be used to endorse or promote products derived from this software
> * without specific prior written permission.
> *
> * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
> * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
> * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
> * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
> * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
> * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
> * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
> * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
> * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
> * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
> * POSSIBILITY OF SUCH DAMAGE.
> */
>
> #include <assert.h>
> #include <errno.h>
> #include <fcntl.h>
> #include <math.h>
> #include <pthread.h>
> #include <signal.h>
> #include <stdint.h>
> #include <stdio.h>
> #include <stdlib.h>
> #include <string.h>
> #include <sys/mman.h>
> #include <sys/stat.h>
> #include <sys/time.h>
> #include <sys/types.h>
> #include <unistd.h>
>
> #define rdtscll(val) do { \
> unsigned int __a,__d; \
> asm volatile("rdtsc" : "=a" (__a), "=d" (__d)); \
> (val) = ((unsigned long)__a) | (((unsigned long)__d)<<32); \
> } while(0)
>
>
> #define MAX_FILE_SIZE ((off_t)1024 * 1024 * 1024 * 1024)
> #define MAX_FILE_IOS 16384
> #define MAX_LATENCY 10000000 // usecs
>
> #define NUM_IOS 1024
> #define IO_SIZE 4096
> #define BUFFER_SIZE (1024 * 1024)
>
> pthread_t tid;
> double cpu_clock;
> long long unsigned cpu_start, cpu_stop;
>
> void *mmap_test(void *arg);
> void die ();
>
> static const char usage_cmds[] =
> "usage: %s [options]\n"
> "cmd line options:\n"
> " -f file_name Read from File named 'file_name'\n"
> " -a file_size File of 'file_size' Bytes/thread\n"
> " -b buffer_size Write/Read into/from buffer of 'buffer_size' Bytes/thread\n"
> " -n num_file_ios Process 'num_file_ios' IOs\n"
> " -s io_size IO Size = 'io_size' Bytes\n"
> " -l max_latency Show latency stats based on usecs of max_latency\n"
> ;
>
> void usage(const char *program)
> {
> fprintf(stderr, usage_cmds, program);
> }
>
> off_t file_size = MAX_FILE_SIZE; // -a
> long long unsigned int buffer_size = BUFFER_SIZE; // -b
> char *filename = "sparse-file"; // -f
> int num_file_ios = NUM_IOS; // -n
> int max_latency = MAX_LATENCY; // -l
> int io_size = IO_SIZE; // -s
> long long unsigned int latency_limit;
>
> int main(int argc, char **argv)
> {
> pthread_attr_t attr;
> cpu_set_t mask;
> FILE *proc;
> char buf[256];
> double mhz = 0.0;
>
> while (1) {
> int option = getopt(argc, argv, "a:b:f:h:l:n:p:s:");
> if (option == -1) {
> break;
> }
> switch (option) {
> case 'a':
> file_size = strtoul(optarg, NULL, 0);
> printf("a: file_size:%ld Bytes :%ld MB\n", file_size, file_size/(1024*1024));
> break;
> case 'b':
> buffer_size = strtoul(optarg, NULL, 0);
> printf("b: buffer_size:%lld Bytes\n", buffer_size);
> break;
> case 'f':
> filename = optarg;
> printf("f: filename:%s\n", filename);
> break;
> case 'h':
> printf("h: options\n");
> goto help;
> case 'l':
> max_latency = strtoul(optarg, NULL, 0);
> printf("l: latency stats based on max latency:%d\n", max_latency);
> break;
> case 'n':
> num_file_ios = strtoul(optarg, NULL, 0);
> printf("n: num_file_ios:%d\n", num_file_ios);
> if (num_file_ios > MAX_FILE_IOS) {
> printf("-n %d Entered > MAX_FILE_IOS:%d\n", num_file_ios, MAX_FILE_IOS);
> exit(1);
> }
> break;
> case 's':
> io_size = strtoul(optarg, NULL, 0);
> printf("s: io_size:%d Bytes\n", io_size);
> break;
> default:
> help:
> usage(argv[0]);
> printf("default:\n");
> exit(1);
> }
> }
>
> proc = fopen("/proc/cpuinfo", "r");
> if (!proc)
> return 0.0;
>
> while (fgets(buf, sizeof buf, proc)) {
> double cpu;
>
> if (sscanf(buf, "cpu MHz : %lf", &cpu) != 1)
> continue;
> if (mhz == 0.0) {
> mhz = cpu;
> continue;
> }
> if (mhz != cpu) {
> fprintf(stderr,
> "Conflicting CPU frequency values: %lf != %lf\n",
> mhz, cpu);
> return 0.0;
> }
> }
> fclose(proc);
> printf("CPU Clock Freq from /proc/cpuinfo:%.4f\n", mhz);
> //
> // Measure CPU Core Frequnecy over 5 second period
> //
> printf("Measuring CPU Frequency......:");
> rdtscll(cpu_start);
> usleep(5000000);
> rdtscll(cpu_stop);
> cpu_clock = (double)((double)(cpu_stop-cpu_start))/(double)5.0;
> printf("%.3f\n", cpu_clock);
> latency_limit = (long long unsigned int) (cpu_clock*max_latency/1000000);
> printf("latency_limit:%llu cycles or %d usecs\n", latency_limit, max_latency);
>
> pthread_attr_init (&attr);
> pthread_attr_setscope (&attr, PTHREAD_SCOPE_SYSTEM);
> pthread_attr_setstacksize (&attr, (size_t) (1024*1024));
>
> if (pthread_create(&tid, &attr, mmap_test, (void *)(long) 0) != 0) {
> die("Thread create failed!");
> }
>
> CPU_ZERO(&mask);
> CPU_SET(0, &mask);
> if (pthread_setaffinity_np(tid, sizeof(mask), &mask) ) {
> printf("WARNING: could not set CPU Affinity, exit...\n");
> exit(1);
> }
>
> pthread_join(tid, NULL);
> sleep(1);
>
> return 0;
> }
>
>
> void die(char *string)
> {
> fprintf(stderr, "\nmmap_test: %s\n", string);
> exit(1);
> }
>
> void *mmapfile(char *fname, off_t size, int *filed)
> {
> int fd;
> void *file_addr;
> struct stat statbuf;
>
> fd = open(fname, O_RDWR | O_CREAT, S_IRUSR | S_IWUSR);
> *filed = fd;
> if (fd < 0) {
> fprintf(stderr, "unable to open %s to get an FD:%s\n", fname, strerror(errno));
> exit(1);
> }
>
> fstat(fd, &statbuf);
> if (statbuf.st_size < size)
> ftruncate(fd, size);
>
> file_addr = mmap(NULL, size, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
> if (file_addr == MAP_FAILED) {
> fprintf(stderr, "datafile mmap failed: %s\n", strerror(errno));
> exit(1);
> }
>
> madvise(file_addr, size, MADV_RANDOM);
> return file_addr;
> }
>
> void create_offsets(off_t *offset_buf, int threadnum)
> {
> int i, curr_time;
>
> curr_time = time(NULL);
> srandom(curr_time / (threadnum + 1));
>
> for (i = 0; i < num_file_ios; i++) {
> double random1 = ((double)(rand()%(RAND_MAX)) / RAND_MAX);
> offset_buf[i] = file_size * random1;
> offset_buf[i] = offset_buf[i] / io_size * io_size;
> }
> }
>
> void *mmap_test(void *arg)
> {
> int threadnum = (long) arg;
> int fd;
> char *file_ptr, *file_addr;
> char *buf_ptr, *buf_addr = NULL;
> int i, j, ios;
> off_t offset_buf[MAX_FILE_IOS];
> unsigned long long latency_start, latency_stop;
>
> posix_memalign((void *)&buf_addr, 4096, buffer_size);
>
> file_addr = mmapfile(filename, file_size, &fd);
>
> ios = buffer_size/io_size;
>
> create_offsets(offset_buf, threadnum);
>
> for (j = 0; j < num_file_ios; j++) {
> buf_ptr = buf_addr;
> file_ptr = file_addr + offset_buf[j];
>
> for (i = 0; i < ios; i++) {
> rdtscll(latency_start);
> *buf_ptr = *(char *)file_ptr;
> rdtscll(latency_stop);
> printf("%lld\n", latency_stop - latency_start);
> buf_ptr += io_size;
> file_ptr += io_size;
> }
> }
>
> close(fd);
> munmap(file_addr, file_size);
> free(buf_addr);
>
> pthread_exit(NULL);
> return 0;
> }
Hi Matthew,
On Tue, Aug 17, 2010 at 03:50:01PM -0400, Matthew Wilcox wrote:
>
> No comment on this? Was it just that I posted it during the VM summit?
I have not forgotten about it. I just have a hard time reproducing
those extreme stalls you observed.
Running that test on a 2.5GHz machine with 2G of memory gives me
stalls of up to half a second. The patchset I am experimenting with
gets me down to peaks of 70ms, but it needs further work.
Mapped file pages get two rounds on the LRU list, so once the VM
starts scanning, it has to go through all of them twice and can only
reclaim them on the second encounter.
At that point, since we scan without making progress, we start waiting
for IO, which is not happening in this case, so we sit there until a
timeout expires.
This stupid-waiting can be improved, and I am working on that. But
since I can not reproduce your observations, I don't know if this is
the (sole) source of the problem. Can I send you patches?
> On Mon, Aug 09, 2010 at 09:30:00AM -0400, Matthew Wilcox wrote:
> >
> > This testcase shows some odd behaviour from the Linux VM.
> >
> > It creates a 1TB sparse file, mmaps it, and randomly reads locations
> > in it. Due to the file being entirely sparse, the VM allocates new pages
> > and zeroes them. Initially, it runs very fast, taking on the order of
> > 2.7 to 4us per page fault. Eventually, the VM runs out of free pages,
> > and starts doing huge amounts of work trying to figure out which of
> > these clean pages to throw away.
This is similar to one of my test cases for:
6457474 vmscan: detect mapped file pages used only once
31c0569 vmscan: drop page_mapping_inuse()
dfc8d63 vmscan: factor out page reference checks
because the situation was even worse before (see the series
description in dfc8d63). Maybe asking the obvious, but the kernel you
tested on did include those commits, right?
And just to be sure, I sent you a test-patch to disable the used-once
detection on IRC the other day. Did you have time to run it yet?
Here it is again:
diff --git a/mm/vmscan.c b/mm/vmscan.c
index 9c7e57c..c757bba 100644
--- a/mm/vmscan.c
+++ b/mm/vmscan.c
@@ -584,6 +584,7 @@ static enum page_references page_check_references(struct page *page,
return PAGEREF_RECLAIM;
if (referenced_ptes) {
+ return PAGEREF_ACTIVATE;
if (PageAnon(page))
return PAGEREF_ACTIVATE;
/*
> > In my testing with a 6GB machine and 2.9GHz CPU, one in every
> > 15,000 page faults takes over a second, and one in every 40,000
> > page faults take over seven seconds!
> >
> > This test-case demonstrates a problem that occurs with a read-mostly
> > mmap of a file on very fast media. I wouldn't like to see a solution
> > that special-cases zeroed pages. I think userspace has done its part
> > to tell the kernel what's it's doing by calling madvise(MADV_RANDOM).
> > This ought to be enough to hint to the kernel that it should be eagerly
> > throwing away pages in this VMA.
We can probably do something like the following, but I am not sure
this is a good fix, either. How many applications are using
madvise()?
--- a/mm/rmap.c
+++ b/mm/rmap.c
@@ -495,7 +495,7 @@ int page_referenced_one(struct page *pag
* mapping is already gone, the unmap path will have
* set PG_referenced or activated the page.
*/
- if (likely(!VM_SequentialReadHint(vma)))
+ if (likely(!(vma->vm_flags & (VM_SEQ_READ|VM_RAND_READ))))
referenced++;
}
> Mapped file pages get two rounds on the LRU list, so once the VM
> starts scanning, it has to go through all of them twice and can only
> reclaim them on the second encounter.
This can be fixed gracefully based on Rik's refault-distance patch :)
With the distance info we can safely drop the use-once mapped file pages.
Thanks,
Fengguang
> Hi Matthew,
>
> On Tue, Aug 17, 2010 at 03:50:01PM -0400, Matthew Wilcox wrote:
> >
> > No comment on this? Was it just that I posted it during the VM summit?
>
> I have not forgotten about it. I just have a hard time reproducing
> those extreme stalls you observed.
me too.
I never forgot this one, but...
I'll trying this again at next week.
Thanks.