EPFL has published my doctoral dissertation online through their infoscience repository.
The Challenge
Load imbalance occurs when a few machines have more work and take longer to finish while the others remain idle. This is a critical challenge in distributed computing that significantly impacts system efficiency.
The Scatter Architecture
My thesis proposes the "scatter architecture," which abandons data locality in favor of load balance by distributing data across all machines rather than processing it locally. This approach disaggregates compute and storage resources.
Results
The dissertation describes three implemented cluster applications demonstrating the scatter model's effectiveness. These systems can perform up to an order of magnitude faster than existing systems while handling significant load imbalance gracefully.
The full thesis is available on EPFL Infoscience.