TSO

[Scott Long]

On Feb 26, 2014, at 10:27 AM, Eric van Gyzen <eric at vangyzen.net> wrote:

> On 02/26/2014 11:37, Sami Halabi wrote:
>> Hi,
>> I'm reading (almost) all mailing emails in mailig list...
>> 
>> Almost every / many problem in network performancr / packets loss ended up
>> suggesting disabling TSO.
>> 
>> I wonder why.. Is it a bug in the implementation? Or bybdesign?
>> What are the usecases that TSO is needed? Myabe  it should be disabled bt
>> default?
> 
> In some cases, I have disabled TSO to [successfully] work around a bug
> in a particular NIC's firmware or hardware, usually a low-end "desktop"
> gigabit NIC.
> 

The same thing happened 10-15 years ago with TCP Checksum offload.  Hardware
support appeared, software support appeared, problems cropped up in both, and
it became an iterative process of identifying, disabling, and fixing.  It’s now a solid
feature that works without question (though there was a paper recently claiming
that hardware CSUM offload doesn’t matter anymore for performance; that’s a
different topic of discussion).

Scott