Oh, and version 1.6.1 released last week. Less bugs, faster for resource-allocation oriented problems with tons of VMs.
We are trying to porte btrPlace to the SCHIAAS simulator (i.e. we are porting the whole VMPlaceS framework to the SCHIAAS actually)
Starting with BtrPLace
The main idea of SCHIAAS is to simulate OpenSTack like platforms.
Our goal is to enable researchers to evaluate advanced placement strategies within this framework
The first issue we met is that shareResource is expecting int whereas we using double. Actually, we can make some tricks to be able to give int but we are wondering whether there is a strong constraint for having int vs double
so any hint/advice is more than welcome
Thanks
@alebre the point might also be why using doubles ? It requires the implementation to support float numbers and rounding issues which no one like. We are using integers but we do not force any unit which is convenient when you want to scale numbers (up to the overflow of course). What kind of resource would you like to express using double ?
Floatint operatin per second
Let's say I have a resource and I'm currently over provisioning, but I want btrplace to find a placing where I no longer am over provisioning. Using ShareableResource I get something like "Usage of virtual resource cpu on node node#55851 (32) exceeds its capacity (24.0)" when modelling my current system. How would I go about this problem using btrplace?
And the SubstitutedVM Event has an unresolved merge conflict in the example ;)
here, the notion of over-provisioning does not hold, you just state what you observe. Typically, you cannot have a node loaded at 110%
in terms of CPUs for example
35c3 is another congress I'm at this month. I just thought we could meet up would you be there https://en.wikipedia.org/wiki/Chaos_Communication_Congress
I already set the scheduler verbosity to three but no luck.
I'm close to implement some kind of binary search ^^
I did that in the past inside SafePlace but eventually I removed the code.
What I did was delta debugging. Then it is a matter of granularity. For example, delta debugging affinity constraints is fine
but delta-debugging resource oriented constraint was not very informative due to the granularity and the arbitrary splits
If you use BtrPlace inside your company, it would be great if we can advertise for that. As an open-source project, this matters for visibility and to acknowledge the work done ;)
This is currently my under-the-radar-side-project but I'll let you know should I come up with something we actually use to schedule our VMs. If so chances are I'll opensource it anyway ;) We have a custom stack to manage libvirt VMs made up of: https://github.com/innogames/serveradmin < our CMDB and CMDB client lib and https://github.com/innogames/igvm < our tools and lib to manage libvirt VMs in the CMDB and on the hypervisors. Igvm does automatic placement of VMs but in the sense of "What's the least bad hypervisor for this VM right now". https://github.com/innogames/igvm/blob/302d1eaabaac773c59d94a3f018a371352810fdb/igvm/settings.py#L164
We configure many things from serveradmin data directly but some are just informative. The second kind is sometimes, though seldom, not correct resulting in invalid constraints.
Eye, I'll try a bit more.