Re: --load question

[Ole Tange]

You can guess whether the first implementation included that or
whether it caused an overloaded machine (that had to be rebooted) to
discover that using the 5 min average is a pretty bad idea.

There is, however, no prize for the correct guess :)

/Ole

On Sat, Feb 12, 2022 at 12:46 AM Neal Becker <ndbecker2@gmail.com> wrote:
>
> Thanks for the clarification.  That's what I wanted to be sure of, that the 
> load would immediately include any job just started so the system wouldn't be 
> overloaded
>
> On Fri, Feb 11, 2022, 6:41 PM Ole Tange <ole@tange.dk> wrote:
>>
>> 100% is (as Joe says) computed as it is for --jobs.
>>
>> The load average is computed as:
>>
>>     ps ax -o state,command|grep '^R'| wc -l
>>
>> This is basically 5 min average, but for this second only. This is to
>> make sure that when GNU Parallel starts a job, the load will increase
>> by one.
>>
>> /Ole
>>
>> On Tue, Feb 8, 2022 at 2:42 AM Joe Sapp <sappj@ieee.org> wrote:
>> >
>> > See the man page description for "--use-sockets-instead-of-threads" and 
>> > "--use-cores-instead-of-threads".  I believe GNU Parallel counts the 
>> > number of processes running and uses that information to match what you 
>> > specify.  By default it's the number of hyperthreaded cores available.
>> >
>> > Joe
>> >
>> > On Sat, Feb 5, 2022 at 9:23 AM Neal Becker <ndbecker2@gmail.com> wrote:
>> >>
>> >> Newb here.  I want to schedule a bunch of tasks > #cores.  Let's say I 
>> >> want to run #cores at a time (100% utilization).
>> >>
>> >> If I do
>> >> seq 1 1000 | parallel --load 100% blah blah...
>> >>
>> >> What is the load that is being looked at?  A 5 minute load average?  So 
>> >> at the time I start this 1000 tasks, loadave is 0 (say), will it start 
>> >> all 1000 tasks at once, because loadave is 0 - only to have loadave 
>> >> become 1000?  Or does it do something smarter than that?
>> >>
>> >> Thanks,
>> >> Neal
>> >>
>> >> --
>> >> Those who don't understand recursion are doomed to repeat it