PhD Thesis: Process migration and load balancing in distributed operating systems
Managing processor allocation in distributed memory computers must be done in a transparent way to virtualize
the global execution resource. This implies, from the operating system view point, to provide functionalities and
policies for global scheduling. I have implemented a process migration facility in the Chorus Distributed operating
system as the base for process placement for multicomputers. Then, I have implemented a load balancing algorithm
in a dedicated server. This new service hides the processor distribution by automatically placing the processes on
lightly loaded nodes and achieves the global scheduling goal.
The document, in french : PhD Thesis
HDR Thesis: Load balancing for objects and components in CORBA
However, developing load balancing tools in operating systems kernel is not possible in every system. The load
balancing policy efficiency depends on the application if performance improvement is targeted. Including this
functionality directly in the operating systems kernel implies a general policy which can not be adapted to each
application. Moreover, process migration is rather costly and must not be used for every process. Just long lived
processes may benefit form migration.
Object migration is less costly as the object size is smaller. We have experimented object migration and load
balancing in CORBA. My work concentrates on load balancing policy adaptation to application and
components constraints. We have show that taking several information into account in the load balancing
decision may lead to important performance improvements.
The document, in french : HDR Thesis