Commutative property of the commands

Question

I need to ask this question just to double check the answer.

Does the order of the commands matter? For example:

Is this command

 taskset 0x2 time echo "foo"

equal than

time taskset 0x2 echo "foo"

?

I need to know if all the commands followed by the taskset will have the same CPU affinity or just the command immediately after it.

Eugeniu Rosca · Accepted Answer · 2015-06-16T16:39:16.863

I need to know if all the commands followed by the taskset will have the same CPU affinity or just the command immediately after it.

Here is a little experiment:

Start two BG tasks

$ > taskset 0x2 sleep 50 & sleep 60 &

Get their PIDs

$ > ps
  PID TTY          TIME CMD
18562 pts/81   00:00:00 bash
20750 pts/81   00:00:00 sleep
20751 pts/81   00:00:00 sleep
20752 pts/81   00:00:00 ps

Get the CPU affinity of the known PIDs:

$ > taskset -p 20750
pid 20750's current affinity mask: 2
$ > taskset -p 20751
pid 20751's current affinity mask: 3f

So, it seems like the CPU affinity is set for the first process only.

Update (trying to explain the following behavior):

/usr/bin/time taskset 0x2 sleep 100000=> only sleep gets the affinity mask 2 (somewhat expected!)
taskset 0x2 /usr/bin/time sleep 100000 => both time and sleep get the affinity mask 2 (need to clarify!)

In the second case, let's call ps -f to get the PPID (parent PID) for each process:

$> taskset 0x2 /usr/bin/time sleep 60 &
[1] 5942
$> ps -f
UID        PID  PPID  C STIME TTY          TIME CMD
user      5942  9698  0 18:19 pts/261  00:00:00 /usr/bin/time sleep 60
user      5943  5942  0 18:19 pts/261  00:00:00 sleep 60
user      5944  9698  0 18:19 pts/261  00:00:00 ps -f
user      9509  9508  0 16:19 pts/261  00:00:00 -bash
user      9698  9509  0 16:20 pts/261  00:00:00 bash
$> taskset -p 5942
pid 5942's current affinity mask: 2
$> taskset -p 5943
pid 5943's current affinity mask: 2

What can be seen is that sleep's PPID (5942) corresponds to /usr/bin/time's PID (5942). IOW sleep is a child process of (has been forked from) /usr/bin/time. Because any child process inherits the configuration of the parent process, sleep will have the same CPU affinity with /usr/bin/time.

I did this test: **taskset 0x2 /usr/bin/time sleep 100000** and both had the affinity mask: 2. But, when I did **/usr/bin/time taskset 0x2 sleep 100000** they had different affinity masks. — Carlos Vega, Jun 11 '15 at 14:48
@carlosvega: That's probably because `sleep` is forked as a child process by `time` and inherits its configuration. — Eugeniu Rosca, Jun 11 '15 at 14:55
Yes, I'm not sure, because of what you said about sleep forking the command. So does it matter or not? — Carlos Vega, Jun 16 '15 at 15:06

score 1 · Answer 2 · answered Jun 16 '15 at 18:51

You seem to be asking two different questions here - "are they equivalent?" and "is CPU affinity inherited?".

First, the two commands you list are not equivalent. The first:

taskset 0x2 time echo "foo"

assuming a PATH and similar setup to the host I'm on at the moment, is equivalent to:

/bin/taskset 0x2 /usr/bin/time /bin/echo "foo"

Which produces a process tree like this:

/bin/taskset
    |
    \- /usr/bin/time
            |
            \- /bin/echo

The second:

time taskset 0x2 echo "foo"

which is equivalent to /bin/taskset 0x2 /bin/echo "foo" wrapped by the bash builtin time, produces this process tree:

/bin/taskset
    |
    \- /bin/echo

In this case, there are only two external processes - the time part is handled internally by bash instead of calling /usr/bin/time.

To answer your second question, CPU affinity is inherited in Linux, so your first example would bind both /usr/bin/time and /bin/echo to the specified CPU sets. In the second example, since time is the shell builtin, it would be influenced by any CPU affinity set on bash itself, not by the taskset in the current command line.

Commutative property of the commands

2 Answers2