I'm using a raspberry pi and I need really fast performance from my CPU for a certain process.
To achieve that, I added isolcpus=3 to my kernel boot parameters, to isolate the core for this process only.
From looking at /proc/interrupts, it seems that this core irqs are also minimal (after isolation).
Now, I'm running this code on the isolated CPU (taskset -p 8 PID):
for (i=0; i<254; i++) {
clock_gettime(CLOCK_REALTIME, &start);
for (rep=0; rep<10000000; rep++) {
}
clock_gettime(CLOCK_REALTIME, &end);
timespec_diff(&start, &end, &diff);
printf("%d\n", diff.tv_nsec);
}
The output is see is:
133562686, 133525447, 133536802, 133525760, 133540134, 133555290, 133540135, 133542218, 133525552, 133524979, 133577791, 133523208, 133525604, 133545916, 87085933, 66719079, 66719339, 66726787, 66719912, 66718870, 66712048, 76724670, 133535917, 133525396, 133528260, 133578416, 133522740, 133525552, 133541177, 133526021, 133553677, 133541906
This is only part of the output. The time is usually consistent on ~133525760, but sometimes it gets faster for a little while, by a multiply of 2.
The tasks running on core 3 are:
PID TID CLS RTPRIO NI PRI PSR %CPU STAT WCHAN COMMAND
22 22 TS - 0 19 3 0.0 S - cpuhp/3
23 23 FF 99 - 139 3 0.0 S - migration/3
24 24 TS - 0 19 3 0.0 S - ksoftirqd/3
25 25 TS - 0 19 3 0.0 S - kworker/3:0
26 26 TS - -20 39 3 0.0 S< - kworker/3:0H
1158 1158 TS - -20 39 3 0.0 S< - kworker/3:1H
1159 1159 TS - 0 19 3 0.0 S - kworker/3:1
5907 5907 TS - 0 19 3 99.1 R - a.out
According to ps, the usage percentage of my process varies between 99 to 100 percent of the CPU (which I also don't understand why it is not consistent on 100%), so the fact that the time is divided by 2 doesn't make sense.
Both speeds are good enough for me, I just need it to be consistent.
Does anyone have an idea why could this happen? Is there any way I can make my loop time consistent?
When I type vmstat -m in command line, it shows:
Cache Num Total Size Pages
fuse_request 0 0 424 9
fuse_inode 0 0 768 5
pid_2 0 0 128 30
nfs_direct_cache 0 0 200 19
nfs_commit_data 0 0 704 11
nfs_write_data 36 36 960 4
nfs_read_data 0 0 896 4
nfs_inode_cache 8224 8265 1048 3
nfs_page 0 0 128 30
fscache_cookie_jar 2 48 80 48
rpc_buffers 8 8 2048 2
rpc_tasks 8 15 256 15
rpc_inode_cache 17 24 832 4
bridge_fdb_cache 14 59 64 59
nf_conntrack_expect 0 0 240 16
For the nfs_write_data line(line 7), why the "pages" is less than "total"?
For some of them, the "total" is always equal to "pages".
Taken from vmstat man page
...
The -m switch displays slabinfo.
...
Field Description For Slab Mode
cache: Cache name
num: Number of currently active objects
total: Total number of available objects
size: Size of each object
pages: Number of pages with at least one active object
totpages: Total number of allocated pages
pslab: Number of pages per slab
Thus, total is the number of slabinfo objects (objects used by the OS as inodes, buffers and so on) and a page can contain more than one object
I wrote a test program which consists of just an infinite loop with some
computations inside, and performs no
I/O operations. I tried starting two instances of the program, one with a high
niceness value, and the other with a low niceness value:
sudo nice -n 19 taskset 1 ./test
sudo nice -n -20 taskset 1 ./test
The taskset command ensures that both programs execute on the same core.
Contrary to my expectation, top reports that both programs get about 50% of the
computation time. Why is that? Does the nice command even have an effect?
The behavior you are seeing is almost certainly because of the autogroup feature that was added in Linux 2.6.38 (in 2010). Presumably when you described running the two commands, they were run in different terminal windows. If you had run them in the same terminal window, then you should have seen the nice value have an effect. The rest of this answer elaborates the story.
The kernel provides a feature known as autogrouping to improve interactive desktop performance in the face of multiprocess, CPU-intensive workloads such as building the Linux kernel with large numbers of parallel build processes (i.e., the make(1) -j flag).
A new autogroup is created when a new session is created
via setsid(2); this happens, for example, when a new terminal window is started. A new process created by fork(2) inherits its
parent's autogroup membership. Thus, all of the processes in a
session are members of the same autogroup.
When autogrouping is enabled, all of the members of an autogroup
are placed in the same kernel scheduler "task group". The Linux kernel scheduler employs an algorithm that equalizes the distribution of
CPU cycles across task groups. The benefits of this for interactive desktop performance can be described via the following example.
Suppose that there are two autogroups competing for the same CPU
(i.e., presume either a single CPU system or the use of taskset(1)
to confine all the processes to the same CPU on an SMP system).
The first group contains ten CPU-bound processes from a kernel
build started with make -j10. The other contains a single
CPU-bound process: a video player. The effect of autogrouping is that
the two groups will each receive half of the CPU cycles. That is,
the video player will receive 50% of the CPU cycles, rather than
just 9% of the cycles, which would likely lead to degraded video
playback. The situation on an SMP system is more complex, but the
general effect is the same: the scheduler distributes CPU cycles
across task groups such that an autogroup that contains a large
number of CPU-bound processes does not end up hogging CPU cycles
at the expense of the other jobs on the system.
The nice value and group scheduling
When scheduling non-real-time processes (e.g., those scheduled
under the default SCHED_OTHER policy), the
scheduler employs a technique known as "group scheduling", under which threads are scheduled in "task groups".
Task groups are formed in the various circumstances, with the relevant case here being autogrouping.
If autogrouping is enabled, then all of the threads that are
(implicitly) placed in an autogroup (i.e., the same session, as
created by setsid(2)) form a task group. Each new autogroup is
thus a separate task group.
Under group scheduling, a thread's nice value has an effect for
scheduling decisions only relative to other threads in the same
task group. This has some surprising consequences in terms of the
traditional semantics of the nice value on UNIX systems. In particular, if autogrouping is enabled (which is the default in various Linux distributions), then
employing nice(1) on a process has an effect
only for scheduling relative to other processes executed in the
same session (typically: the same terminal window).
Conversely, for two processes that are (for example) the sole
CPU-bound processes in different sessions (e.g., different terminal
windows, each of whose jobs are tied to different autogroups),
modifying the nice value of the process in one of the sessions has
no effect in terms of the scheduler's decisions relative to the
process in the other session. This presumably is the scenario you saw, though you don't explicitly mention using two terminal windows.
If you want to prevent autogrouping interfering with the traditional nice behavior as described here, you can disable the feature
echo 0 > /proc/sys/kernel/sched_autogroup_enabled
Be aware though that this will also have the effect of disabling the benefits for desktop interactivity that the autogroup feature was intended to provide (see above).
The autogroup nice value
A process's autogroup membership can be viewed via
the file /proc/[pid]/autogroup:
$ cat /proc/1/autogroup
/autogroup-1 nice 0
This file can also be used to modify the CPU bandwidth allocated
to an autogroup. This is done by writing a number in the "nice"
range to the file to set the autogroup's nice value. The allowed
range is from +19 (low priority) to -20 (high priority).
The autogroup nice setting has the same meaning as the process
nice value, but applies to distribution of CPU cycles to the
autogroup as a whole, based on the relative nice values of other
autogroups. For a process inside an autogroup, the CPU cycles that it
receives will be a product of the autogroup's nice value (compared
to other autogroups) and the process's nice value (compared to
other processes in the same autogroup).
I put together a test.c that just does:
for(;;)
{
}
And then ran it with your nice's. I didn't run a different sudo for each one, but rather sudo'd an interactive shell and ran them both from there. I used two &'s.
I got one ./test hitting my CPU hard, and one barely touching it.
Naturally, the system still felt quite responsive; it takes a lot of CPU-hogging processes on modern processors to get so much load you can "feel" it.
That stands in contrast to I/O-hogging processes and memory-hogging processes; in these cases a single greedy process can make a system painful to use.
I'd guess either your system has a relatively unique priority-related bug (or subtlety), or there's something up with your methodology.
I ran my test on an Ubuntu 11.04 system.
I'm assuming that there's a & missing at the end of the command line. Otherwise, the second line won't run until the first completes.
While both processes are running, use something like top and make sure that they each have the nice value that you assigned.
What happens if you launch the processes using only taskset and then adjust their priority with renice after they are running?
Process niceness (priority) setting HAS an effect on Linux! (in practise, but ONLY if you give it enough work to do!)
On my system, as long as all cores are fully loaded, then nice does have an impact. On ubuntu 14.04, processes run with nice -N gets through 0.807 ** N operations compared to processes run without altering the nice value (given you are running one instance per core for each nice level).
In my case I have quad core i7 with hyper threading turned off, so if I run four or less processes, then it doesn't matter what their nice values are - they each get a full core. If I run four processes at nice level 0 and 4 at nice level 12, then the ones at level 12 get through 0.807 ^ 12, ie approx 7% of the work the ones at nice level zero do. The ratio seems to be a reasonable predictor from nice levels 0 through 14, after that it fluctuates (A few runs had nice level 18 processing more than nice 16 for instance) - Running the test for longer may smooth the results out.
(ruby 2.1.2 used)
,cl file:
uptime
nices='-0 -6 -12 -18'
nices='-0 -18'
nices='-0 -2 -4 -6 -8 -10 -12 -14 -16 -18'
rm -f ,n-*
for i in 1 2 3 4
do
for n in $nices
do
nice $n ruby ,count_loops.rb > ,n${n}-$i &
done
done
ps -l
uptime
wait
uptime
ps -l
c=`cat ,n-0-[1234] | total`
last=$c
for n in $nices
do
echo
c2=`cat ,n${n}-[1234] | total`
echo total of `cat ,n${n}-[1234]` is $c2
echo -n "nice $n count $2, percentage: "
echo "3 k $c2 100 * $c / p" | dc
echo -n " percent of last: "
echo "3 k $c2 100 * $last / p" | dc
last=$c2
done
uptime
echo total count: `cat ,n-*-[1234] | total`
,count_loops.rb file
#!/usr/bin/env ruby
limit = Time.new + 70
i=0
while Time.new < limit
i += 1
j = 0
while (j += 1) < 10000
t = j
end
end
puts i
results of sh ,cl - initial diagnostic output:
19:16:25 up 20:55, 2 users, load average: 3.58, 3.59, 2.88
F S UID PID PPID C PRI NI ADDR SZ WCHAN TTY TIME CMD
0 S 1000 4987 4977 0 80 0 - 7297 wait pts/3 00:00:00 bash
0 S 1000 11743 2936 0 80 0 - 2515 wait pts/3 00:00:00 rubymine.sh
0 S 1000 11808 11743 6 80 0 - 834604 futex_ pts/3 00:18:10 java
0 S 1000 11846 11808 0 80 0 - 4061 poll_s pts/3 00:00:02 fsnotifier64
0 S 1000 19613 4987 0 80 0 - 2515 wait pts/3 00:00:00 sh
0 R 1000 19616 19613 0 80 0 - 7582 - pts/3 00:00:00 ruby
0 R 1000 19617 19613 0 82 2 - 7582 - pts/3 00:00:00 ruby
0 R 1000 19618 19613 0 84 4 - 7582 - pts/3 00:00:00 ruby
0 R 1000 19619 19613 0 86 6 - 7416 - pts/3 00:00:00 ruby
0 R 1000 19620 19613 0 88 8 - 6795 - pts/3 00:00:00 ruby
0 R 1000 19621 19613 0 90 10 - 6247 - pts/3 00:00:00 ruby
0 R 1000 19622 19613 0 92 12 - 6247 - pts/3 00:00:00 ruby
0 R 1000 19623 19613 0 94 14 - 6247 - pts/3 00:00:00 ruby
0 R 1000 19624 19613 0 96 16 - 6078 - pts/3 00:00:00 ruby
0 R 1000 19625 19613 0 98 18 - 6012 - pts/3 00:00:00 ruby
0 R 1000 19626 19613 0 80 0 - 7582 - pts/3 00:00:00 ruby
0 R 1000 19627 19613 0 82 2 - 7582 - pts/3 00:00:00 ruby
0 R 1000 19628 19613 0 84 4 - 7582 - pts/3 00:00:00 ruby
0 R 1000 19629 19613 0 86 6 - 7582 - pts/3 00:00:00 ruby
0 R 1000 19630 19613 0 88 8 - 6247 - pts/3 00:00:00 ruby
0 R 1000 19631 19613 0 90 10 - 6247 - pts/3 00:00:00 ruby
0 R 1000 19632 19613 0 92 12 - 6247 - pts/3 00:00:00 ruby
0 R 1000 19633 19613 0 94 14 - 6144 - pts/3 00:00:00 ruby
0 R 1000 19634 19613 0 96 16 - 4971 - pts/3 00:00:00 ruby
0 R 1000 19635 19613 0 98 18 - 4971 - pts/3 00:00:00 ruby
0 R 1000 19636 19613 0 80 0 - 7582 - pts/3 00:00:00 ruby
0 R 1000 19637 19613 0 82 2 - 7449 - pts/3 00:00:00 ruby
0 R 1000 19638 19613 0 84 4 - 7344 - pts/3 00:00:00 ruby
0 R 1000 19639 19613 0 86 6 - 7582 - pts/3 00:00:00 ruby
0 R 1000 19640 19613 0 88 8 - 7416 - pts/3 00:00:00 ruby
0 R 1000 19641 19613 0 90 10 - 6210 - pts/3 00:00:00 ruby
0 R 1000 19642 19613 0 92 12 - 6247 - pts/3 00:00:00 ruby
0 R 1000 19643 19613 0 94 14 - 5976 - pts/3 00:00:00 ruby
0 R 1000 19644 19613 0 96 16 - 6111 - pts/3 00:00:00 ruby
0 R 1000 19645 19613 0 98 18 - 4971 - pts/3 00:00:00 ruby
0 R 1000 19646 19613 0 80 0 - 7582 - pts/3 00:00:00 ruby
0 R 1000 19647 19613 0 82 2 - 7516 - pts/3 00:00:00 ruby
0 R 1000 19648 19613 0 84 4 - 7416 - pts/3 00:00:00 ruby
0 R 1000 19649 19613 0 86 6 - 6247 - pts/3 00:00:00 ruby
0 R 1000 19650 19613 0 88 8 - 6177 - pts/3 00:00:00 ruby
0 R 1000 19651 19613 0 90 10 - 6247 - pts/3 00:00:00 ruby
0 R 1000 19652 19613 0 92 12 - 6078 - pts/3 00:00:00 ruby
0 R 1000 19653 19613 0 94 14 - 6247 - pts/3 00:00:00 ruby
0 R 1000 19654 19613 0 96 16 - 4971 - pts/3 00:00:00 ruby
0 R 1000 19655 19613 0 98 18 - 4971 - pts/3 00:00:00 ruby
0 R 1000 19656 19613 0 80 0 - 3908 - pts/3 00:00:00 ps
19:16:26 up 20:55, 2 users, load average: 3.58, 3.59, 2.88
19:17:37 up 20:56, 3 users, load average: 28.92, 11.25, 5.59
F S UID PID PPID C PRI NI ADDR SZ WCHAN TTY TIME CMD
0 S 1000 4987 4977 0 80 0 - 7297 wait pts/3 00:00:00 bash
0 S 1000 11743 2936 0 80 0 - 2515 wait pts/3 00:00:00 rubymine.sh
0 S 1000 11808 11743 6 80 0 - 834604 futex_ pts/3 00:18:10 java
0 S 1000 11846 11808 0 80 0 - 4061 poll_s pts/3 00:00:02 fsnotifier64
0 S 1000 19613 4987 0 80 0 - 2515 wait pts/3 00:00:00 sh
0 R 1000 19794 19613 0 80 0 - 3908 - pts/3 00:00:00 ps
results of sh ,cl - statistics: (percentage of last is the percentage of this total compares to the count for the last group of processes)
total of 99951 101725 100681 104046 is 406403
nice -0 count , percentage: 100.000
percent of last: 100.000
total of 64554 62971 64006 63462 is 254993
nice -2 count , percentage: 62.743
percent of last: 62.743
total of 42997 43041 43197 42717 is 171952
nice -4 count , percentage: 42.310
percent of last: 67.434
total of 26882 28250 27151 27244 is 109527
nice -6 count , percentage: 26.950
percent of last: 63.696
total of 17228 17189 17427 17769 is 69613
nice -8 count , percentage: 17.129
percent of last: 63.557
total of 10815 10792 11021 11307 is 43935
nice -10 count , percentage: 10.810
percent of last: 63.113
total of 7023 6923 7885 7323 is 29154
nice -12 count , percentage: 7.173
percent of last: 66.357
total of 5005 4881 4938 5159 is 19983
nice -14 count , percentage: 4.917
percent of last: 68.542
total of 3517 5537 3555 4092 is 16701
nice -16 count , percentage: 4.109
percent of last: 83.576
total of 4372 4307 5552 4527 is 18758
nice -18 count , percentage: 4.615
percent of last: 112.316
19:17:37 up 20:56, 3 users, load average: 28.92, 11.25, 5.59
total count: 1141019
( Purists will note I am mixing ruby, shell and dc - they will have to forgive me for old habits from last century showing through ;) )
I run an example program from APUE and nice does have the effect.
The example program mainly fork a child and both the parent and child execute a i++ increment operation for given time(10s). By giving the child different nice value, the result shows if nice makes a difference.
The book warns that I should run the program with a uniprocessor PC, fisrt I tried with my own PC, i5-7500 CPU # 3.40GHz × 4 (4 cores), giving different nice value, almost no difference.
Then I log into my remote server, 1 processor 1 GB, and get the expected difference.
1 core processor 1 GB Test result:
./a.out
NZERO = 20
current nice value in parent is 0
current nice value in child is 0, adjusting by 0
now child nice value is 0
parent count = 13347219
child count = 13357561
./a.out 20 //child nice set to 20
NZERO = 20
current nice value in parent is 0
current nice value in child is 0, adjusting by 20
now child nice value is 19
parent count = 29770491
ubuntu#VM-0-2-ubuntu:~$ child count = 441330
Test program(I made a little modification), from Section 8.16, APUE:
apue.h is merely a header wrapper
err_sys() is also a error handler wrapper, you can use printf temporarily.
#include "apue.h"
#include <errno.h>
#include <sys/time.h>
#if defined(MACOS)
#include <sys/syslimits.h>
#elif defined(SOLARIS)
#include <limits.h>
#elif defined(BSD)
#include <sys/param.h>
#endif
unsigned long long count;
struct timeval end;
void
checktime(char *str)
{
struct timeval tv;
gettimeofday(&tv, NULL);
if (tv.tv_sec >= end.tv_sec && tv.tv_usec >= end.tv_usec) {
printf("%s count = %lld\n", str, count);
exit(0);
}
}
int
main(int argc, char *argv[])
{
pid_t pid;
char *s;
int nzero, ret;
int adj = 0;
setbuf(stdout, NULL);
#if defined(NZERO)
nzero = NZERO;
#elif defined(_SC_NZERO)
nzero = sysconf(_SC_NZERO);
#else
#error NZERO undefined
#endif
printf("NZERO = %d\n", nzero);
if (argc == 2)
adj = strtol(argv[1], NULL, 10);
gettimeofday(&end, NULL);
end.tv_sec += 10; /* run for 10 seconds */
if ((pid = fork()) < 0) {
err_sys("fork failed");
} else if (pid == 0) { /* child */
s = "child";
printf("current nice value in child is %d, adjusting by %d\n",
nice(0), adj);
errno = 0;
if ((ret = nice(adj)) == -1 && errno != 0)
err_sys("child set scheduling priority");
printf("now child nice value is %d\n", ret);
} else { /* parent */
s = "parent";
printf("current nice value in parent is %d\n", nice(0));
}
for(;;) {
if (++count == 0)
err_quit("%s counter wrap", s);
checktime(s);
}
}
Complete source code link: https://wandbox.org/permlink/8iryAZ48sIbaq27y
I have a multi-threaded application(C++ and pthread library) and I will like to know how much resources(CPU and memory) each thread uses.
Is there a way to find out these details on Solaris and Linux both or on either of them.
You could use the ps command with some option:
ps -eLo pid,ppid,lwp,nlwp,osz,rss,ruser,pcpu,stime,etime,args | more
PID PPID LWP NLWP SZ RSS RUSER %CPU STIME ELAPSED COMMAND
0 0 1 1 0 0 root 0.0 Oct_02 4-02:13:37 sched
1 0 1 1 298 528 root 0.0 Oct_02 4-02:13:36 /sbin/init
2 0 1 1 0 0 root 0.0 Oct_02 4-02:13:36 pageout
Have a look at the ps man's page to get some information (LWP (light weight process))