19.
11.
2015
Multicore Load Balancing Based on Behaviour Observations on µ-Kernels
Philipp Eppelt
TU Dresden
Zwischenpräsentation der Diplomarbeit
Congestion on shared resources reduces performance on multicore CPUs.
Different applications compete for the same resource and affect each
other negatively or, at best, not at all. If the multicore CPU supports
SMT, even more resources are shared.
I will give a overview over research, which tries to reduce resource
congestion by observing thread behaviour and separating heavy resource
users from each other. The presented work explores not only cache and
memory congestion, but also cooperation between SMT-threads.
Based on this review, I will present my approach for a load distribution
service on L4Re, running on a partially symmetric multiprocessor. In
addition, it will provide isolation guarantees for security or real-time
tasks.
|
