Closed simpx closed 6 years ago
and, why we need rte_eth_promiscuous_enable
?
simpx,
Please see this message board: https://community.hpe.com/t5/Switches-Hubs-and-Modems/When-to-use-Flow-Control/td-p/4337588. Although the content focuses on switches, the same logic can also be applied to TCP endpoints. Flow control in TCP endpoints can easily be handled in the transport layer. You can re-enable NIC flow control if you want, but my intention is to squeeze more efficiency from disabling this feature.
Enabling promiscuous mode was added as an after-thought (and was related to performance metrics). It's purpose was to show how many packets-per-second can the link layer handle (in the best case) irrespective of the labelled destination MAC address.
@ajamshed thx! 👍
I run mTCP and rdma in the same node, rdma rely on NIC flow control to ensure a lossless ethernet
so I re-enable this feature is my case
I will remove promiscuous mode in my next commit. Enabling promiscuous mode makes more sense for projects like the mOS networking stack.
https://github.com/eunyoung14/mtcp/blob/master/mtcp/src/dpdk_module.c#L715
I review the commit which add this code, but I don't know why should mTCP disabled flow control?