[Bloat] [aqm] big science discovers sch_fq and pacing

[Rick Jones]

On 05/31/2015 10:08 PM, Jonathan Morton wrote:
> My favourite part of that is "disabling TSO leads to a doubling in
> throughput". In precisely the sort of situation that TSO was intended to
> help throughput.

Perhaps my recollection of history is suffering from bitrot, but I 
thought the situation TSO was intended to help was when there wasn't 
enough CPU horsepower on the sender to enable achieving the desired 
transfer rate when going up and down the protocol stack at just 1460ish 
bytes at a time.  One of the reasons it was called (at least by some in 
early days) "Poor Man's Jumbo Frames."

happy benchmarking,

rick jones