Title
A Low Noise Unikernel for Extrem-Scale Systems.
Abstract
We expect that the size and the complexity of future supercomputers will increase on their path to exascale systems and beyond. Therefore, system software has to adapt to the complexity of these systems to simplify the development of scalable applications. In cloud environments, the activity of a virtual machine on a neighboring core may decrease performance due to issues such as cache contamination (noise neighbor problem). In this paper, we present the unikernel operating system HermitCore coming up with predictable runtimes, which improves the scalability. It extends the multi-kernel approach with unikernel features while providing better programmability and scalability for hierarchical systems. In addition, the same binary can be used to run as unikernel within virtual machines. By using a unikernel, the memory footprint of Virtual Machines (VMs) is decreased, which reduces the pressure on the cache system and improves the overall performance. We prove the predictable runtime of the design via micro benchmarks by taking the example of HermitCore on the upcoming manycore architecture Knights Landing.
Year
Venue
Field
2017
ARCS
System software,Unikernel,Virtual machine,Computer science,Cache,Parallel computing,Message Passing Interface,Memory footprint,Scalability,Cloud computing,Distributed computing
DocType
Citations 
PageRank 
Conference
0
0.34
References 
Authors
11
3
Name
Order
Citations
PageRank
Stefan Lankes115226.39
Simon Pickartz2236.06
Jens Breitbart3355.89