Performance optimizations for LTE User-Plane L2 software
Alapuranen, Sakari (2015-05-04)
Alapuranen, Sakari
S. Alapuranen
04.05.2015
© 2015 Sakari Alapuranen. Tämä Kohde on tekijänoikeuden ja/tai lähioikeuksien suojaama. Voit käyttää Kohdetta käyttöösi sovellettavan tekijänoikeutta ja lähioikeuksia koskevan lainsäädännön sallimilla tavoilla. Muunlaista käyttöä varten tarvitset oikeudenhaltijoiden luvan.
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:oulu-201505061447
https://urn.fi/URN:NBN:fi:oulu-201505061447
Tiivistelmä
Nowadays modern mobile communication networks are expected to be able to compete with wired connections in both latency and speed. This places a lot of pressure on the mobile communication protocols, which are very complex, and much of their implementation depends on the software. The performance of the software directly affects the capacity of the network, which in turn affects the throughput and latency of the network’s users and the number of users the network can support.
This thesis concentrates on identifying software components of LTE User-Plane radio interface protocols for improvements, and exploring the solutions for better performance. This study leans on system component tests and the performance profiler tool perf, which enables tracking the effects of software optimizations from function-level to the whole system-level accuracy. In addition to perf, performance counters provided by the processor are manually observed and they provide the verification on why specific optimizations affect the performance.
Slow memory accesses or cache misses are identified as the most constraining factor in the software’s performance. Also many good practices are found during the optimization work, such as arranging code common path first. Surprisingly, separating hardly executed code from hotspots also has a positive impact on performance, in addition to shrinking the active binary. The optimization work results in the whole software’s load decreasing from 60% to 50% and in some individual functions load decreases of over 70% are achieved.
This thesis concentrates on identifying software components of LTE User-Plane radio interface protocols for improvements, and exploring the solutions for better performance. This study leans on system component tests and the performance profiler tool perf, which enables tracking the effects of software optimizations from function-level to the whole system-level accuracy. In addition to perf, performance counters provided by the processor are manually observed and they provide the verification on why specific optimizations affect the performance.
Slow memory accesses or cache misses are identified as the most constraining factor in the software’s performance. Also many good practices are found during the optimization work, such as arranging code common path first. Surprisingly, separating hardly executed code from hotspots also has a positive impact on performance, in addition to shrinking the active binary. The optimization work results in the whole software’s load decreasing from 60% to 50% and in some individual functions load decreases of over 70% are achieved.
Kokoelmat
- Avoin saatavuus [31657]