You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
As one of the first step to try to get FastClick more open and properly maintained by the community, I'd like to fix the Kernel mode.
With the last patch, it is compiling on up to 4.4. However recent kernel crashes, I think it's due to the last changes by Eddie in the scheduling, among them the recent issues mentioned there : kohler/click#407 . @pallas@peterhurley did you experience this kind of problem with a recent merge?
In a more long-term, I think we can bring back Kernel Click in the front stage if, instead of using the rx_handler we used the xdp_handler. One can hijack this one to not use eBPF but eg a Click function. This allows to completely byoass the kernel stack functions, including the heavy sk_buff creation step - which is mostly useless for a Click program - and achieve speed comparable to what DPDK proposes, but in-kernel. Looking at the ixgbe driver, the XDP buffers are safe to keep as the ring bucket will be re-mapped to some new buffer in the DMA zone, so for this kind of driver we do not even need to remap or copy the buffers. If some people are still looking into Kernel Click, this would probably be the biggest performance improvement for Kernel Click in years.
This discussion was converted from issue #90 on February 09, 2021 09:33.
Heading
Bold
Italic
Quote
Code
Link
Numbered list
Unordered list
Task list
Attach files
Mention
Reference
Menu
reacted with thumbs up emoji reacted with thumbs down emoji reacted with laugh emoji reacted with hooray emoji reacted with confused emoji reacted with heart emoji reacted with rocket emoji reacted with eyes emoji
-
As one of the first step to try to get FastClick more open and properly maintained by the community, I'd like to fix the Kernel mode.
With the last patch, it is compiling on up to 4.4. However recent kernel crashes, I think it's due to the last changes by Eddie in the scheduling, among them the recent issues mentioned there : kohler/click#407 . @pallas @peterhurley did you experience this kind of problem with a recent merge?
In a more long-term, I think we can bring back Kernel Click in the front stage if, instead of using the rx_handler we used the xdp_handler. One can hijack this one to not use eBPF but eg a Click function. This allows to completely byoass the kernel stack functions, including the heavy sk_buff creation step - which is mostly useless for a Click program - and achieve speed comparable to what DPDK proposes, but in-kernel. Looking at the ixgbe driver, the XDP buffers are safe to keep as the ring bucket will be re-mapped to some new buffer in the DMA zone, so for this kind of driver we do not even need to remap or copy the buffers. If some people are still looking into Kernel Click, this would probably be the biggest performance improvement for Kernel Click in years.
Beta Was this translation helpful? Give feedback.
All reactions