Design and evaluation of effective load sharing in distributed real-time systems

Implementation of Decentralized Load Sharing in Networked Workstations Using the Condor Package

“…The transfer-out ratio for the Condor package is higher than that for the proposed LS mechanism, which may serve as another evidence of the possibil- We plan to reconfigure some of the low-level process and memory management functions into a kernel server that resides inside the OS kernel to handle IPC and signal facilities. We also plan to incorporate features we proposed in [19,[21][22][23][24] into the LS mechanism, and equip the LS mechanism with the abilities to deal with large communication latencies, excessive task transfers and task collisions, and component failures.…”

“…However, as discussed in [18], the dominating factor in assessing LS performance lies more in the global communication overhead and aggregate resource management than in (small) delays incurred by kernel calls. Moreover, placing the mechanism outside the kernel facilitates later expansion or generalization of other LS strategies to deal with large communication latency [21], excessive task transfer [22], and node/link failure [19,23,24]. One inherent limitation resulted from placing the LS mechanism outside the OS kernel is that interprocess communication and signal facilities cannot be easily implemented, and are not supported in the current implementation.…”

Load Balancing Using Heterogeneous Processors for Continuum Problems on a Mesh

“…This paper shows the balancing of the computational load can eliminate latency and be modeled more accurately than (4).…”

A least flow-time first load sharing approach for distributed server farm

“…In addition, the high task variation in the heavy-tailed task distribution causes small tasks to be severely delayed by large tasks. Previous work [10,17,18] showed that better performance can be obtained (in a heavy-tailed distribution) by un-balancing the load. They assume that the task size is known in advance by the dispatcher which unbalances the load by assigning tasks based on the task size: larger and smaller tasks are assigned to different queues in such a way that mean waiting time per task is reduced.…”