[Cake] perhaps a performance optimization

[Cake] perhaps a performance optimization

[Dave Taht]

thought: since it takes two data packets for tcp to release an ack,
perhaps it is better in some circumstances (notably with offloads
enabled) to peel at quantum * 2,
still delivering packets with a quantum (with a max of a mtu), so that
delivering two unpeeled packets would cost that flow a DRR round, but
save a lot on peeling overhead.

sch_fq has a quantum of 3028.

-- 
Dave Täht
Let's go make home routers and wifi faster! With better software!
http://blog.cerowrt.org

Re: [Cake] perhaps a performance optimization

[Jonathan Morton]

> On 24 Apr, 2016, at 18:44, Dave Taht <dave.taht@gmail.com> wrote:
> 
> thought: since it takes two data packets for tcp to release an ack,
> perhaps it is better in some circumstances (notably with offloads
> enabled) to peel at quantum * 2,
> still delivering packets with a quantum (with a max of a mtu), so that
> delivering two unpeeled packets would cost that flow a DRR round, but
> save a lot on peeling overhead.

If a GSO packet is less than an MTU in total, surely it’ll peel into a single packet anyway?  I don’t see where the gain is.

 - Jonathan Morton