Abstract:
Like many of its practitioners, the discipline of load balancing is settling into a comfortable middle age. Load balancing models and objectives are seemingly well understood. The existing suite of software tools seems to enable high performance for most parallel applications.
Unfortunately, this consensus opinion is far from reality. As is common in middle age, the honest truth is less pretty than we believe it to be. Our familiar models are inexpressive and deeply flawed. Our objectives fail to capture any accurate cost measure. And existing software tools are often poorly designed and can deliver disappointing performance to applications.
This talk will critique the complacent state of the art and offer some suggestions for new research directions. Far from being a settled field, there are a number of exciting open questions and opportunities in load balancing. New perspective and insights are needed. By restoring the field to a state of youthful exuberance, the middle-aged speaker hopes to vicariously recapture his own lost youth.
Download full paper.
powerpoint version of overheads,
HTML version of overheads
from Plenary talk at 3rd DRAMA Steering Workshop, September, 1999.