Awarded for safe cloud overbooking without crashes
NEWS
Four researchers at Computing Science, Umeå University, has been awarded Best Paper for an article on techniques to enable data centers to be overbooked without risk for application crashes.
For the last two years, Luis Tomas and Johan Tordsson have worked with overbooking – how to enable data centers to host more applications and services than those allowed by the total available hardware capacity. Through their worked, they have developed intelligent algorithms to predict future workloads. Based on these forecasts, decisions are made whether to accept a new application or not, and on which server(s) to host it.
Avoid slowdown and crasches
During roughly the same period of time, Cristian Klein and Francisco Hernandez-Rodriguez have worked together with researchers from Lund University within the Cloud Control project founded by the Swedish Research Council. They developed Brownout, a paradigm for building more adaptive and robust cloud applications. The term is borrowed from the electrical power industry, where a brownout is a temporary voltage reduction during peak demand to prevent total blackouts. In a similar manner, Brownout-enabled cloud applications can temporary reduce their service quality in order to avoid slowdown or crashes during overload.
These two techniques were combined in the award winning article to enable data centers to be overbooked with reduced risk. In the unlikely event of capacity shortages, Brownout is used to ensure that applications do not crash. An extensive evaluation using web servers with real workload traces demonstrates that the combined approach improves datacenter utilization as well as application performance.
Picture above: Cristian Klein, Johan Tordsson, Francisco Hernandez-Rodriguez and Luis Tomas, The Department for Computing Science, Umeå University, has been awarded at The International Conference on Cloud and Autonomic Computing (CAC 2014).