Hi Steven, On Sat, Oct 11, 2014 at 6:05 AM, Steven C Timm <t...@fnal.gov> wrote:
> During most of the time we have been running OpenNebula 2 and 3 we have > been using a > rank based on FREEMEMORY. We are now doing tests using OpenNebula 4.8, in > a use case > where we are filling up an empty cloud. FREEMEMORY still in theory should > be an accurate value > FREEMEMORY hasn't been a Host attribute for a few versions. I can't remember exactly when it was removed, but it is not part of the host now. The predefined variables you can use in the rank or requirements are [1]: NAME MAX_CPU MAX_MEM FREE_MEM FREE_CPU USED_MEM USED_CPU HYPERVISOR By the way, I see this list is buried too deep in the docs, so I'll try to move it to a new section and explain each attribute. > but the problem is that 6-7 VM's are typically being launched on every > SCHED cycle and so a > node that starts out with all of its memory free will end up full of > virtual machines in a single cycle. > > Once we got to a full cloud and steady-state it would be fine but when > you have 8 VM's > starting at once on an old 8-core node, it takes much longer than it > otherwise would. > > Any cheap suggestions to get the default scheduler to do a more > horizontal fill? > > Steve Timm > In /etc/one/sched.conf [2] you can set MAX_HOST to 1. This forces the scheduler to launch one VM to each host per cycle. Combined with the SCHED_INTERVAL, you can configure a slower deployment rate. Best regards, Carlos. [1] http://docs.opennebula.org/4.8/user/references/template.html#requirement-expression-syntax [2] http://docs.opennebula.org/4.8/administration/references/schg.html -- Carlos Martín, MSc Project Engineer OpenNebula - Flexible Enterprise Cloud Made Simple www.OpenNebula.org <http://www.opennebula.org/> | cmar...@opennebula.org | @OpenNebula <http://twitter.com/opennebula> <cmar...@opennebula.org>
_______________________________________________ Users mailing list Users@lists.opennebula.org http://lists.opennebula.org/listinfo.cgi/users-opennebula.org
