...Because I was not aware of --delay. ;) Thanks.
On Mon, Mar 25, 2013 at 9:19 PM, Ole Tange <[email protected]> wrote: > Can you explain why --load X, --delay Y does not solve this for you? > > /Ole > > On Mon, Mar 25, 2013 at 4:15 PM, Jay Hacker <[email protected]> wrote: > > This is awesome. It works for me (as you said, plus or minus a few > > processes). Thank you. > > > > Would it be possible to use this for the '--slowstart X' feature, that > > starts one job at a time, waits X seconds, and looks at the load before > > starting another? > > > > > > On Wed, Mar 6, 2013 at 8:23 PM, Ole Tange <[email protected]> wrote: > >> > >> I have implemented an instant --load calculator. With this it should > >> be possible to tell GNU Parallel to spawn a process for each CPU that > >> is currently sitting idle. > >> > >> On a 64 core machine this: > >> > >> seq 10000 | parallel --load 20% burnP6 & > >> sleep 2 > >> seq 10000 | parallel --load 100% burnMMX & > >> sleep 4 > >> seq 10000 | parallel --load 100% burnBX & > >> > >> will start 14 burnP6, 52 burnMMX, and 0 burnBX. If the first parallel > >> is killed, the last remaining parallels will spawn around 14 processes > >> it total. > >> > >> It should also count processes that are waiting for local disks as > >> non-idle. So this: > >> > >> ls *iso | parallel --load 1 -j0 -v cat > >> > >> will only spawn 3 cat and not one for every iso file. > >> > >> The biggest problem I have seen so far is that it is not very > >> accurate. So it may shoot a little over or under the target (2-4). But > >> it seems way better than the previous --load. > >> > >> http://alpha.gnu.org/gnu/parallel/parallel-20130307.tar.bz2 > >> > >> Please give it a spin and provide feedback. > >> > >> > >> /Ole > >> > > >
