Java 8 Library for Building High-Performance, Thread-Per-Core Applications (github.com) 2 points by xedin 10y ago ↗ HN
[–] crudbug 10y ago ↗ How is it different from Netty [1] / Reactor [2] ?[1] http://netty.io[2] http://projectreactor.io [–] xedin 10y ago ↗ It uses Thread-Per-Core architecture where Netty/Reactor use thread-pools to process work and have dedicated threads for selecting from network. [–] crudbug 10y ago ↗ So CPU affinity provides further optimization ? I am thinking less context switches.[1] http://conferences.sigcomm.org/imc/2010/papers/p218.pdf [–] xedin 10y ago ↗ Yes, exactly since it allows for fewer context switches and cache locality not only for the network packets but for disk I/O as well.
[–] xedin 10y ago ↗ It uses Thread-Per-Core architecture where Netty/Reactor use thread-pools to process work and have dedicated threads for selecting from network. [–] crudbug 10y ago ↗ So CPU affinity provides further optimization ? I am thinking less context switches.[1] http://conferences.sigcomm.org/imc/2010/papers/p218.pdf [–] xedin 10y ago ↗ Yes, exactly since it allows for fewer context switches and cache locality not only for the network packets but for disk I/O as well.
[–] crudbug 10y ago ↗ So CPU affinity provides further optimization ? I am thinking less context switches.[1] http://conferences.sigcomm.org/imc/2010/papers/p218.pdf [–] xedin 10y ago ↗ Yes, exactly since it allows for fewer context switches and cache locality not only for the network packets but for disk I/O as well.
[–] xedin 10y ago ↗ Yes, exactly since it allows for fewer context switches and cache locality not only for the network packets but for disk I/O as well.
4 comments
[ 1.6 ms ] story [ 22.9 ms ] thread[1] http://netty.io
[2] http://projectreactor.io
[1] http://conferences.sigcomm.org/imc/2010/papers/p218.pdf