Activity: Talk or presentation types › Invited talk › Scientific
Description
The design and analysis of load-balancing algorithms has been a popular research topic over the past three decades. The basic problem consists of assigning a stream of incoming jobs to servers, so as to equalize the server loads as far as possible. The fundamental challenge consists of taking the best assignment decision upon the arrival of each job despite uncertainty on the length of ongoing and future jobs. However, the growing use of cloud-computing platforms for data-intensive applications has added further constraints on load balancing. For instance, the large scale of the system calls for decentralized algorithms that perform well in the presence of many dispatchers, and data locality may restrict the set of servers to which each job can be assigned.