Run-state migration ARM big.LITTLE




1 run-state migration

1.1 clustered switching
1.2 in-kernel switcher (cpu migration)
1.3 heterogeneous multi-processing (global task scheduling)





run-state migration

there 3 ways different processor cores arranged in big.little design, depending on scheduler implemented in kernel.


clustered switching

big.little clustered switching


the clustered model approach first , simplest implementation, arranging processor identically-sized clusters of big or little cores. operating system scheduler can see 1 cluster @ time; when load on whole processor changes between low , high, system transitions other cluster. relevant data passed through common l2 cache, first core cluster powered off , other 1 activated. cache coherent interconnect (cci) used. model has been implemented in samsung exynos 5 octa (5410).



in-kernel switcher (cpu migration)

big.little in-kernel switcher


cpu migration via in-kernel switcher (iks) involves pairing big core little core, possibly many identical pairs in 1 chip. each pair operates 1 virtual core, , 1 real core (fully) powered , running @ time. big core used when demand high , little core employed when demand low. when demand on virtual core changes (between high , low), incoming core powered up, running state transferred, outgoing shut down, , processing continues on new core. switching done via cpufreq framework. complete big.little iks implementation added in linux 3.11. big.little iks improvement of cluster migration, main difference each pair visible scheduler.


the more complex arrangement involves non-symmetric grouping of big , little cores. single chip have 1 or 2 big cores , many more little cores, or vice versa. nvidia created similar low-power companion core in tegra 3 soc.



heterogeneous multi-processing (global task scheduling)

big.little heterogeneous multi-processing


the powerful use model of big.little architecture heterogeneous multi-processing (hmp), enables use of physical cores @ same time. threads high priority or computational intensity can in case allocated big cores while threads less priority or less computational intensity, such background tasks, can performed little cores.


this model has been implemented in samsung exynos starting exynos 5 octa series (5420, 5422, 5430).









Comments

Popular posts from this blog

History Swan Lake

Medical vaporizers Vaporizer (inhalation device)

Proto-Slavic Loanwords in Serbian