340
edits
Line 165: | Line 165: | ||
* Max RX queues per socket: This setting specifies the amount of threads dedicated to read and write interactions with the network interface controllers. By increasing this value, network receive bandwidth can be increased before packet drops occur. By decreasing this value, data analysis will improve. The default setting of 2 RX queues is suitable for most configurations as data analysis typically needs much more processing ressources. | * Max RX queues per socket: This setting specifies the amount of threads dedicated to read and write interactions with the network interface controllers. By increasing this value, network receive bandwidth can be increased before packet drops occur. By decreasing this value, data analysis will improve. The default setting of 2 RX queues is suitable for most configurations as data analysis typically needs much more processing ressources. | ||
* Use Hyper-Threading for RX queues: This setting allows enabling or disabling Hyper-Threading for the threads dedicated to read and write interactions with the network interface controllers. By disabling it, network performance can be improved as the RX queues will be distributed to physical CPU cores only. By enabling it, RX queues will also be distributed to virtual Hyper-Threading CPU cores which is not as efficient as physical CPU cores. By using Hyper-Threading, data analysis will improve as there are more CPU cores available for these tasks. Hyper-Threading is used by default. This is suitable for most configurations as data analysis typically needs much more processing ressources. | * Use Hyper-Threading for RX queues: This setting allows enabling or disabling Hyper-Threading for the threads dedicated to read and write interactions with the network interface controllers. By disabling it, network performance can be improved as the RX queues will be distributed to physical CPU cores only. By enabling it, RX queues will also be distributed to virtual Hyper-Threading CPU cores which is not as efficient as physical CPU cores. By using Hyper-Threading, data analysis will improve as there are more CPU cores available for these tasks. Hyper-Threading is used by default. This is suitable for most configurations as data analysis typically needs much more processing ressources. | ||
* Preferred Network interface controller: This setting allows fine tuning of network and data analysis performance for dedicated network controllers. The selected set of network controllers will be preferred over the others. Usually the fastest or the network controller with the most traffic should be preferred. The ‘Auto’ | * Preferred Network interface controller: This setting allows fine tuning of network and data analysis performance for dedicated network controllers. The selected set of network controllers will be preferred over the others. Usually the fastest or the network controller with the most traffic should be preferred. The ‘Auto’ setting is used by default, preferring the fastest network controller. | ||
setting is used by default, preferring the fastest network controller. | |||
You should only change these parameters in discussion with the Allegro Packets support. | You should only change these parameters in discussion with the Allegro Packets support. |
edits