Do you mean you are writing a resource manager (scheduler)?  Why would the architecture matter?  You should worry about the behavior first, and let that drive the architecture.  For instance, unless you're managing VERY large clusters, you can certainly do everything on a single management host.  But if you worry about what happens when that host crashes (and even a good linux server can lose against a storm), then you need to contemplate a replicated or distributed architecture.  You could even go to the extreme: every node participates as a peer in the management of resources - but doing this would be inefficient and complex, and probably not justified.

Scheduling a cluster (Iaas, PaaS, etc) is not difficult.  It's wise to start with a simple, modular design first, and only complicate it if necessary.

Thanks Mark Hahn ,

The idea of architecture is more clear now,

I will choose to start with a simple, modular design first, and only complicate it if necessary.

The Research Papers attached may help you.