Re: [dpdk-dev] [PATCH v1] ixgbe_pmd: forbid tx_rs_thresh above 1 for all NICs but 82598

["Ananyev, Konstantin" <konstantin.ananyev@intel.com>]

> -----Original Message-----
> From: dev [mailto:dev-bounces@dpdk.org] On Behalf Of Matthew Hall
> Sent: Friday, September 11, 2015 6:18 PM
> To: Vladislav Zolotarov
> Cc: dev@dpdk.org
> Subject: Re: [dpdk-dev] [PATCH v1] ixgbe_pmd: forbid tx_rs_thresh above 1 for all NICs but 82598
> 
> On Fri, Sep 11, 2015 at 07:18:20PM +0300, Vladislav Zolotarov wrote:
> > We thought about linearization too. It's doable with extra mempool and it
> > may be optional so that those that don't need could compile it out and/or
> > disable it in a runtime...
> 
> High-level question. How realistic is sending a 40-segment frame in the first
> place? This whole thing seems kind of academic to me unless I missed
> something.
> 
> I usually use 2K pktmbufs and I don't think this is an uncommon size. Most
> jumbo frame hardware only does 9.5KB max frame size or so.
> 
> So I am having a hard time imagining how I'd end up with more than 10 segments
> as a worst-case scenario.

As I remember, with freebsd stack when TSO is on it was not unusual to see chains of ~30 segments. 
That's over port with 'normal' mtu (1.5K).
Konstantin

> 
> Matthew.