First page Back Continue Last page Summary Graphics
Cluster-standalone node hybrid
Combine the previous ideas and link an application into a UML cluster
Kernel code would be partitioned across nodes
Userspace code tied to original node
Kernel able to access all resources of hosts
Application code is standalone node
Kernel code is clustered
The UML library and clustering ideas can be combined to allow normal applications to be clustered fairly transparently. They will be clustered to the extent they use kernel functionality.
Their own code will be tied to the node that they started on, but the kernel code will be perfectly willing to migrate between nodes.
So, if the app has its own threads running on different nodes, any kernel data that they use will be transparently migrated as needed between nodes.
The result would be a hybrid between a cluster and a standalone node (or a set of standalone nodes if the application creates its own threads on those nodes).