(Downloads - 0)
For more info about our services contact : help@bestpfe.com
Table of contents
Contents
1 Introduction
2 State of the Art
2.1 Multicore architectures
Architecture details
Related challenges
2.2 Non UniformMemory Access architectures
Architecture details
Linux NUMA policies
Related challenges
2.3 System virtualization
Technical details
Architecture details
Related challenges
2.4 Software settings
2.5 Conclusion
3 TheWell-Known Bottlenecks
3.1 The Virtualized I/O Overhead
Hardware Emulation
The I/OMemoryManagement Unit
Evaluation of IOMMU
3.2 The virtualized IPI overhead
Usage of the IPI
Implementations of IPI
The libactive library
Evaluation of vIPI
3.3 The Xen load balancer
Completely Fair Scheduler
Credit Scheduler
The libpin
Evaluation of pining scheme
3.4 Conclusion
4 The NUMA Bottleneck
4.1 NUMA policies under study
Comparison of NUMA policies effects
The NUMA policy selectionmetric
4.2 Current hypervisor NUMA management
The default round-1G policy
Huge pages and splinterring
Evaluation of the default policy
4.3 Conclusion
5 Virtualization of NUMA Architectures
5.1 Improved hypervisor NUMA management
The Xen implementation of round-4K
The Xen implementation of Carrefour
Limitations of PEBS with virtualization
The Xen implementation of first-touch
Limitations of first-touch with IOMMU
The Xen NUMA policy selection interface
5.2 Evaluation
Evaluation of vNUMA on a singlemachine
Evaluation of vNUMA on severalmachines
5.3 Conclusion
6 Conclusion
6.1 Future works
6.2 Perspectives
Bibliography
