From: Eliezer Tamir <[email protected]>
Date: Wed, 27 Feb 2013 09:55:49 -0800
> This patchset adds the ability for the socket layer code to poll directly
> on an Ethernet device's RX queue. This eliminates the cost of the interrupt
> and context switch and with proper tuning allows us to get very close
> to the HW latency.
>
> This is a follow up to Jesse Brandeburg's Kernel Plumbers talk from last year
> http://www.linuxplumbersconf.org/2012/wp-content/uploads/2012/09/2012-lpc-Low-Latency-Sockets-slides-brandeburg.pdf
I just wanted to say that I like this work a lot.
It proves a lot of things I try to impress upon people who talk about
how RDMA is such a "must have" because the software stack can't
possibly give anything near HW latency.
Obviously such arguments are complete bullshit as is shown by these
changes.
This is exactly the kind of approach that makes sense rather than
trying to put entire TCP stacks in the network card firmware.
Thanks again for doing this work and I look forward to applying
this stuff once all the kinks are worked out. The folks in the
Intel NIC group continue to impress me.
> This is exactly the kind of approach that makes sense rather than
> trying to put entire TCP stacks in the network card firmware.
>
And should also obviate the need to put a full TCP stack in user space!
> Thanks again for doing this work and I look forward to applying
> this stuff once all the kinks are worked out. The folks in the
> Intel NIC group continue to impress me.
>
> --
> To unsubscribe from this list: send the line "unsubscribe netdev" in
> the body of a message to [email protected]
> More majordomo info at http://vger.kernel.org/majordomo-info.html
On 01/03/2013 23:24, David Miller wrote:
> From: Eliezer Tamir <[email protected]>
> Date: Wed, 27 Feb 2013 09:55:49 -0800
>
>> This patchset adds the ability for the socket layer code to poll directly
>> on an Ethernet device's RX queue. This eliminates the cost of the interrupt
>> and context switch and with proper tuning allows us to get very close
>> to the HW latency.
>>
>> This is a follow up to Jesse Brandeburg's Kernel Plumbers talk from last year
>> http://www.linuxplumbersconf.org/2012/wp-content/uploads/2012/09/2012-lpc-Low-Latency-Sockets-slides-brandeburg.pdf
>
> I just wanted to say that I like this work a lot.
>
> It proves a lot of things I try to impress upon people who talk about
> how RDMA is such a "must have" because the software stack can't
> possibly give anything near HW latency.
>
> Obviously such arguments are complete bullshit as is shown by these
> changes.
>
> This is exactly the kind of approach that makes sense rather than
> trying to put entire TCP stacks in the network card firmware.
>
> Thanks again for doing this work and I look forward to applying
> this stuff once all the kinks are worked out. The folks in the
> Intel NIC group continue to impress me.
It would really help us to hear opinions on the open issues we listed.
Thanks,
Eliezer