On Wed, Jun 6, 2012 at 12:33 AM, Joseph A. Farran <[email protected]> wrote: > Yes you guys are doing a great job and I've done my share of programming > back in the stone age, so I do appreciate how difficult it is to upkeep > something this big and complex. OGE is very nice so far and it seems very > flexible.
Thanks, I know why in some cases people like the more straightforward LSF or Torque interface better... We did our best and reviewed code & docs when Sun was in total control (and Sun listened to community input, which worked quite well at that time). However, Grid Engine is based on very old code (like code developed by research projects - (without googling) I think it's close to 20 years ago when it was first started in DQS), and many things were defined before Sun bought it and opensourced it. Only a few features have our inputs (eg. arch string naming, and dynamic spooling, which allows you to pick classic vs BerkeleyDB spooling), but we tried our best and tried to read all the design codes and tried to make sure that features were straightforward - but for older interfaces we will need more manpower to change the code. For example I did not like adding overloading GIDs to create "pseudo job support" - ie. tagging processes by using an additional GID, and thus we added the cgroups support for Linux in GE 2011.11u1. Note that cgroups is designed for this purpose (the Google guys need job support for Linux running in their own datacenter and thus "invented" cgroups"), so we better take advantage of it than rolling our own :-) > One last question on the parallel environment. Many of our users use NAMD > and now NAMD-CUDA. I Google searching for a "namd" OGE parallel > environment thinking I would find a whole bunch but found very little info > on it. I remember Chris wrote something about NAMD & Grid Engine. And gridengine.info points me to: http://www.ks.uiuc.edu/Research/namd/wiki/index.cgi?NamdOnGridEngine > NAMD requires a special file with "group main" followed by a list of compute > nodes to run on, and since OGE does "tight" integration of parallel jobs, is > there a NAMD PE available to add to OGE? On the UIUC page, you will see the PE definition. Remember, the core idea behind a "tight" PE is that the slaves are under the control of Grid Engine. So to change a PE from non-tight to tight, you mainly need to do 2 things: 1) set control_slaves to TRUE in the PE definition. 2) use qrsh -inherit to start all the slave tasks. Rayson _______________________________________________ users mailing list [email protected] https://gridengine.org/mailman/listinfo/users
