ErlCee++

todd <>
Tue Jun 8 16:34:08 CEST 2004


Thread scheduling overhead and latency is very OS dependent, especially 
as the
number of threads increase, mutexes are used,  different priorities are 
used,
and load increases.

As a real-time guy i've been interested in how the 0(1) scheduler is 
performing.
The previous incarnations of linux were not adequate for many realtime apps.

Luke Gorrie wrote:

>Luke Gorrie <> writes:
>
>  
>
>>  Linux 2.6 does one thousand context-switches per second, spends
>>  absolutely bugger-all CPU to do it, and AFAIK the scheduler is O(1).
>>    
>>
>
>Let me clarify what I meant:
>
>In 2.6 the default time-slice for a process is 1ms. So if you have two
>or more CPU-busy processes then you will have at least one thousand
>context switches per second, and the overhead of this is very small -
>i.e. Linux context switching is "pretty darn fast."
>
>Specifically I did not mean to say that 1000 switches per second was a
>maximum (that a switch takes a whole millisecond), nor to imply that
>other Unices are slower (I've no idea.)
>
>Happy now, Per? :-)
>
>-Luke
>
>P.S., actually 'vmstat' says my laptop is making about 2000 switches
>per second while it's (from my point of view) idle.
>
>
>  
>



More information about the erlang-questions mailing list