The role of the Perceus server

There are several ways to run Perceus to make it scale for the specific cluster size that it will manage. You can always redefine any of these methods later, but it will require more hackery.

Single Perceus master, local state directory

Running Perceus like this is the most straight forward and the default out of the box capable solution. It basically configures Perceus to utilize a local state directory.

In this configuration, the Perceus master will need to either export out this directory to the nodes via NFS (default), or you can utilize Apache to distribute the VNFS content (not default and less clean, but due to religion and specific needs this option exists).

How many nodes this method supports is very dependent of the capabilities of the hardware configuration of the Perceus master. Theoretically a decent hardware configuration can support about 32 simultaneously booting nodes (this is not the total number of supported nodes, rather how many nodes can be provisioned in parallel). Using a lightweight VNFS capsule, this makes provisioning a 512 node cluster possible in about 4 minutes. As long as you stagger the boots, this model will scale into the thousands of nodes.

This method is subject to variances in addition to hardware configuration as well. NFS implementation, specific NFS and kernel tuning, etc..

Single Perceus master, non-local state directory

Similar to the above, the primary difference here is that another system is used for NFS. This could be either another Unix/Linux system or something like a NAS device or even a parallel file system. This can drastically increase the number of nodes that can be provisioned simultaneously with ranges highly dependent on the NAS storage system used.

Multiple Perceus master, non-local state directory

Again similar to the above example, it is possible to run multiple Perceus masters all mounting the non-local state directory thus sharing all of the state and provisioning load.