high performance network

[changtimwu]

We make use of PF_RING in product and surveyed netmap. comparison

New studies are why-do-we-use-the-linux-kernels-tcp-stack. Then snabb draw my eye.

Intel dual 10G adaptor is about 10000 NTD.

[changtimwu]

Brilliant idea of pflua Pflua can load up the libpcap library and use it to compile a pflang expression to BPF. In any case, whether you start from raw BPF or from a pflang expression, the BPF is compiled directly to Lua source code, which LuaJIT can gnaw on as it pleases.

pflang -> BPF -> Lua source -> native code(by luajit)

Can this approach be applied to solidity ?

related works: https://github.com/ethereum/EIPs/issues/125 https://blog.ethereum.org/2014/02/09/why-not-just-use-x-an-instructive-example-from-bitcoin/

[changtimwu]

https://github.com/snabbco/snabb/blob/master/src/doc/getting-started.md can be a 10G tcp-replay and user space router/switch.

http://snabbco.github.io/#snabblab plenty of docs/apps. can be scapy alternative

[changtimwu]

http://lwn.net/Articles/691887/

[changtimwu]

Facebook's XDP seems a quite clever technique. 跟現有 kernel 相容, 利用 eBPF 過濾出想特別處理的 packet .

quite a lot docs it provides. https://github.com/iovisor/bpf-docs

很實用 ex. packet classifier https://github.com/iovisor/bpf-docs/blob/master/bpf_netdev_conference_2016Feb12_report.pdf

http://events.linuxfoundation.org/sites/events/files/slides/ebpf_on_the_mainframe_lcon_2015.pdf classic BPF (page 5 ~ page 23) 至少可以做到 JIT

http://man7.org/linux/man-pages/man8/tc-bpf.8.html classic BPF(aka) 也可以做到 cls_bpf

http://borkmann.ch/talks/2014_devconf.pdf pure cBPF doc

[changtimwu]

https://events.linuxfoundation.org/sites/events/files/slides/bpf_collabsummit_2015feb20.pdf

The BPF examples starts from kernel V3.19.
It adds SO_ATTACH_BPF series options, which largely extends the classic filter option SO_ATTACH_FILTER.

The helper functions supports load_ only

From kernel V4.1 it supports stores

OpenWRT mvebu bypassed V3.19. Can we manually merge the eBPF commit to V3.18

see bcc network session for more powerful usage.

[changtimwu]

• Highlevel eBPF code is majorly written in bcc
• P4 looks promising and it can be converted into ebpf.

[changtimwu]

great example of how to us xt_bpf(netfilter + bpf) which is applicable to cbpf.

[changtimwu]

http://lxr.free-electrons.com/source/Documentation/networking/packet_mmap.txt?v=3.18 Packet sockets work well together with Linux socket filters, thus you also might want to have a look at Documentation/networking/filter.txt

kernel 4.3 supports PACKET_FANOUT_EBPF and here is the original commit.

valuable fanout example

[changtimwu]

XDP overview https://www.iovisor.org/technology/xdp XDP makes lots of progress recently especially HW offloading. Note the how complicated of XDP forward and all the techniques involved like page per packet http://netdevconf.org/1.2/session.html?herbert-xdp-workshop

appears in kernel 4.8 and mlx4 is the only NIC it supports(Yes, XDP requires driver modification)

please take a look of the example programs xdp1_kern.c xdp1_user.c xdp2_kern.c