Exploiting Price and Performance Tradeoffs in Heterogeneous Clouds

Toward a Unified Framework for the Heterogeneous Cloud Utilizing Bytecode Containers

“…The key characteristic in these systems of study is that we wish to leverage the system's heterogeneity to improve some performance characteristic of our applications, such as total execution time, response time, power efficiency, or cost. While such systems have in practice existed for many years, for example in the High Performance Computing (HPC) domain, recently interesting results have become available that show the opportunities for optimization in public cloud infrastructures [31].…”

“…While we have shown that in practice it is possible to benefit from exploiting heterogeneity in cloud systems [31], such systems are often used in a limited capacity in cloud systems; some times limited solely to the use of accelerators such as Graphics Processing Units (GPUs) [31] and rarely do we encounter a system-wide focus on enabling flexibility as well as allowing for optimization in the presence heterogeneity. A unified approach and heterogeneous systems framework for use with container-based application virtualization in the presence of a potentially heterogeneous cloud environment is required.…”

“…Also at the software level, cloud providers may offer instances that are tuned to different performance characteristics such as high memory bandwidth or having a low latency configuration [31].…”

“…As put by Roloff et al, "building a cloud system out of more than one instance type is an area that has been researched less. A system composed of different instance types is considered a heterogeneous cloud" [31]. What this means in practice is when provider(s) make these diverse types of computation resources available to us, perhaps by offering different combinations of hardware/software features to tenants at different prices (for example offering low-power and high performance processor instances), we are in a heterogeneous cloud environment and new optimization opportunities arise.…”

“…In "Exploiting Price and Performance Tradeoffs in Heterogeneous Clouds" Roloff et al explore the use of multiple instance types in the heterogeneous cloud to improve cost efficiency, reducing the price of execution while maintaining a similar application performance [31]. Their "results show that heterogeneous clouds are able to execute parallel applications with a reduced cost, while maintaining a similar performance as homogeneous clouds" [31].…”