Seminar NUMA-aware Locks
The evolution of locks is tightly coupled with the evolution of computing architectures. Modern architectures feature an increasing number of nodes (or sockets), each comprising of a locally attached memory, a fast local cache and multiple processing units (cores). Accesses by a core to a local memory or local cache are significantly faster than accesses to a remote memory or cache lines residing on another node, a characteristic known as NUMA (Non-Uniform Memory Access).
As a result, researchers have proposed multiple designs for NUMA-aware locks, which try to keep the lock owner-ship within the same socket. This approach decreases remote cache misses and the associated inter-socket communication, as it increases the chance that the lock data, as well as the shared data subsequently accessed in a critical section, will be cached locally to the socket on which a lock holder is running.
Veranstalter
Prof. Dr. Bettina Schnor
Dipl.-Math. Petra Vogel
Modulnummern
- Bachelor Computational Science: 6030
- 555601 - Modulteilprüfung
- 555602 - alternativ
- Master Computational Science: 10020
- 555201 - Modulteilprüfung
- 555202 - alternativ
Ort/Termine
Das Seminar findet Dienstag, den 16.4.2019 und 23.4.2019 von 10:00 Uhr bis 12:00 Uhr im Raum 03.04.2.14 statt. Weitere Termine nach Absprache.