0

I have ubuntu server 14.04, and I'm wondering if there is any easy way to achieve this task.

For example, I can already do this for single processes with taskset -pac 2 29435, but is there any way to assign it via username instead of process id?

Thank you in advance.

user3794271
  • 3
  • 1
  • 2
  • 1
    Yeah, it's possible. Can you give some detail on why you're doing this? – ewwhite Jan 08 '16 at 16:43
  • I intend to divide my dedicated server into 4 parts (or 4 users). Each user would have a left 4 dead 2 gameserver running on them. Since srcds does not support multicore, the most efficient way to make use of my machine is to create 4 gameservers running in different ports. However I also wish to have one core for each user. – user3794271 Jan 08 '16 at 17:02
  • You may want to use LXC containers instead. – ewwhite Jan 08 '16 at 17:03
  • Taskset would be easiest approach for me at this time, as my total knowledge is fairly limited. Is it really impossible to complete with taskset? Also, thank you for the continued assistance – user3794271 Jan 08 '16 at 17:06

1 Answers1

2

Do nothing. Really, do nothing.

If your purpose here is to spread your resources as efficiently as possible, the proper thing to do is leave the operating system to move the processes to relevant CPUs on demand and as necessary.

  • If all the processes are demanding CPU at the same time, the operating system will migrate them in such a way they will run on different processors anyway.
  • If one of the processes spend most of their time idle, its probably better to leave it sharing a CPU with other processes.

For the case of performance optimization, limiting CPU resources or usage is never a good idea -- you'll only make performance worse. You only want to restrict CPU resources or CPUs that can be used where you are deliberately attempting to cripple the process.

Times where you might want to cripple a process would be:

  • You are a hosting provider and are offering a minimum/maximum banding to what resources are available for a resource consumer.
  • The process is very badly written/misbehaves and will improperly consume all resources on a given system bringing down other processes. Doing this would make your load shoot up but 'save' the system from unnecessarily consuming CPU that could be used for other things. Youd generally seek to actually fix the program in this case as a proper fix.

So - do nothing and let the operating system sort it out. After a while of your instances running (and if they really are CPU heavy) you can run the command ps -Lo psr,pid,tid $(pgrep <processname>) and you'll see that each resource is being divvied up correctly on the CPU.

If you want to determine if each process is getting its fair share, and how much you really are utilizing in each program you do the following replacing the process name and get the following results:

$ ps -Lo psr,pid,tid,etime,cputime,comm $(pgrep firefox)
PSR   PID   TID     ELAPSED     TIME COMMAND
  2  3400  3400  1-07:16:10 01:22:29 firefox
  2  3400  3425  1-07:16:10 00:00:00 gdbus
  2  3400  3426  1-07:16:09 00:00:00 Gecko_IOThread
  3  3400  3427  1-07:16:09 00:00:00 Link Monitor
  1  3400  3428  1-07:16:09 00:02:50 Socket Thread
  1  3400  3429  1-07:16:09 00:00:00 firefox
  0  3400  3430  1-07:16:09 00:00:25 JS Helper
  3  3400  3431  1-07:16:09 00:00:26 JS Helper
  3  3400  3432  1-07:16:09 00:00:25 JS Helper
  1  3400  3433  1-07:16:09 00:00:25 JS Helper
  1  3400  3434  1-07:16:09 00:00:26 JS Helper
  3  3400  3435  1-07:16:09 00:00:25 JS Helper
  0  3400  3436  1-07:16:09 00:00:25 JS Helper
  0  3400  3437  1-07:16:09 00:00:26 JS Helper
  2  3400  3438  1-07:16:09 00:00:02 JS Watchdog
  2  3400  3439  1-07:16:09 00:00:00 Hang Monitor
  1  3400  3440  1-07:16:09 00:00:00 BgHangManager
  3  3400  3441  1-07:16:09 00:00:32 Cache2 I/O
  0  3400  3442  1-07:16:09 00:02:41 Timer
  3  3400  3444  1-07:16:09 00:00:00 GMPThread
  2  3400  3447  1-07:16:09 00:07:24 Compositor
  0  3400  3448  1-07:16:09 00:01:08 ImageBridgeChil
  3  3400  3449  1-07:16:09 00:00:31 ImgDecoder #1
  1  3400  3450  1-07:16:09 00:00:32 ImgDecoder #2
  3  3400  3451  1-07:16:09 00:00:31 ImgDecoder #3
  2  3400  3452  1-07:16:09 00:00:00 ImageIO
  2  3400  3453  1-07:16:09 00:04:07 SoftwareVsyncTh
  0  3400  3454  1-07:16:08 00:00:00 firefox
  2  3400  3455  1-07:16:08 00:00:00 Cert Verify
  2  3400  3456  1-07:16:08 00:00:00 IPDL Background
  0  3400  3457  1-07:16:08 00:00:37 DOM Worker
  2  3400  3458  1-07:16:08 00:00:03 HTML5 Parser
  2  3400  3462  1-07:16:07 00:00:01 mozStorage #1
  1  3400  3463  1-07:16:07 00:00:00 Proxy R~olution
  1  3400  3464  1-07:16:07 00:00:49 URL Classifier
  2  3400  3466  1-07:16:07 00:00:02 mozStorage #2
  0  3400  3467  1-07:16:07 00:00:00 gmain
  3  3400  3468  1-07:16:07 00:00:00 Cache I/O
  3  3400  3471  1-07:16:07 00:00:00 mozStorage #3
  2  3400  3477  1-07:16:07 00:00:35 DOM Worker
  2  3400  3479  1-07:16:07 00:00:00 mozStorage #4
  0  3400  3482  1-07:16:07 00:00:00 localStorage DB
  2  3400  3483  1-07:16:07 00:00:03 mozStorage #5
  1  3400  3519  1-07:15:57 00:00:00 mozStorage #6
  2  3400  3537  1-07:14:09 00:00:31 DOM Worker
  0  3400  3562  1-07:08:35 00:00:00 mozStorage #7
  0  3400  3587  1-06:59:39 00:00:00 threaded-ml
  2  3400  3597  1-06:49:40 00:00:00 mozStorage #8
  2  3400  7594  1-01:36:55 00:00:34 threaded-ml
  3  3400 11679    10:48:07 00:00:00 firefox
  2  3400 11684    10:48:07 00:00:00 typefind:sink
  2  3400 11687    10:48:07 00:00:00 typefind:sink
  1  3400 11689    10:48:07 00:00:00 typefind:sink
  0  3400 11690    10:48:07 00:00:00 mpegaudioparse0
  1  3400 11691    10:48:07 00:00:00 mpegaudioparse1
  2  3400 11692    10:48:07 00:00:00 mpegaudioparse2
  0  3400 11693    10:48:07 00:00:00 aqueue:src
  1  3400 11694    10:48:07 00:00:00 aqueue:src
  1  3400 11695    10:48:07 00:00:00 aqueue:src
  2  3400 22770    05:38:46 00:00:00 firefox
  3  3400 29803       10:17 00:00:00 DNS Res~er #226
  3  3400 30018       01:28 00:00:00 DNS Res~er #228

In this example I've used firefox on my machine but you can change the process name to suit your needs.

Here, I'm requesting each thread that lives in the process. The columns mean the following:

  • PSR is the processor number assigned to that task.
  • PID is the process id.
  • TID is the thread id. (the main processes tid equals its pid)
  • ELAPSED provides the total amount of time the process has been runnable for. Basically, the amount of time its been started for.
  • TIME is the total amount amount of time the process has actually ran on a CPU.
  • COMMAND is the command name as declared by the process. Here you can see each actual thread is given a particular name, presumably being used to describe its purpose.

To determine a proceses utilization during its lifetime as a percentage you could perform the following calculation (there I'm using firefox):

TIME     /    ELAPSED * 100  = UTIL
112570   /    4949    * 100  = 4.40

Note: The actual process ID (the main starting thread whose tid == pid) acts as a 'container' for the cumalative total of all threads (existing or no longer existing) CPU times, so gives you a reasonably accurate depiction of a processes entire usage.

To explain, if a processes lifetime equals its cpu time it means that for all the time the process has ever lived, its demanded and recieved a CPU to run on. This would equal 100% of cpu utilization.

I'm almost certain in reality you're going to find that your processes are going to be using hardly any CPU.

So - to reiterate, to perform as efficiently as possible do nothing as your kernel knows how to best prioritize CPU resources to best utilize your system. Anything you could possibly add is in most cases is reducing your overall effectiveness.

Unless its your plan to actually cripple the processes in some way (and there are circumstances where you may actually intend to do that) you dont want to use taskset, control groups or LXC/Docker to get the best performance possible.

Matthew Ife
  • 23,357
  • 3
  • 55
  • 72