|
Overview
Modern systems architectures are making increasing use of Chip Multithreading (CMT) and multiple cores per socket to increase available computing power using minimal space, power, and cooling. This article discusses the challenges involved in enabling legacy applications to scale on CMT architectures in particular and on multicore architectures in general. This article also discusses possible architectural changes to the application. In conclusion, I propose virtual environments as the quickest path to scalability.
Contents
- Why Do We Need Multiple-Core Architectures?
- CMT Threads
- Thread-Level Parallelism
- Cooltst Tool
- Cooltst Sample Output
- Why My Application Will Not Scale
- Lock Contention
- Solutions
- Virtualization
- In Conclusion
- References
Download the article as a PDF (892 KB).
Author
Rick Weisner is a Senior Engineer working for the Solaris 10 Adoption group in the US Systems Practice. Rick has 18 years of experience at Sun. For the past 12 years he has been doing performance analysis and supporting software development. Prior to joining Sun he had 10 years of experience in research and development, commercial programming, and nuclear engineering.
| 
Print-friendly Version
