I like that the issues of low latency software execution are getting more attention in the kernel community but this is a bit misleading in that no one doing HFT seriously employs kernel's network stack. We just buy Xilinx cards and use OpenOnload to bypass the kernel entirely. You'll see sub-microsecond times from host to host at the min (including an ULL Arista switch along the way). Median would be around 2 micros.
The eliminating jitter part is ok but nothing new as everyone does that these days, especially that most of these techniques are nicely described in kernel docs any way. Good to see the topic on LWN though.