WARNING: The following assumption is based on my little understanding of mesos architecture and internals, you should not take it as a definitive answer and may wait for more experimented suggestions.
About my little understanding of the mesos ressource allocation process, I think that on your kind of environnements (ARM / x86 / GPU / FPGA) it will not really allocate them wisely as the default algorithm is not so smart and consider ressources as commodities and not their real speed etc. I read some topics earlier about the necessity to improve this specific part of the "kernel", but It didn't mention your archs and focused deeply on the x86 family. I think that integrate the GPUs and FPGAs would be awesome! One nice feature would be that the master look at the registered slaves deeper regarding their archs and ressources perfomance before offers any ressource to a task. This kind of feature could be implement when the slave try to register to the master as a pre-fly test (Benchmark??). That would then add smartest offers and ressource scheduling. Anyway, long story short, I don't think so, but you should way for more experimented answers. 2015-02-05 0:00 GMT+01:00 Pradeep Kiruvale <[email protected]>: > Hi All, > > I am new to Mesos and I have heard and read lot about it. > > I have few doubts regarding the resource allocation by the mesos, please > help me > to clarify my doubts. > > In a data center, if there are thousands of heterogeneous nodes > (x86,arm,gpu,fpgas) then is the mesos can really allocate a co-located > resources for any incoming application to finish the task faster? > > How these resource constraints are solved? what kind of a constraint > solver it uses? > > Is the policy maker configurable? > > Thanks & Regards, > Pradeep > > >
