* [dpdk-dev] rte_eth_rx_queue_count accuracy
@ 2015-09-18 10:33 Alejandro Lucero
2015-09-21 22:41 ` Stephen Hemminger
0 siblings, 1 reply; 3+ messages in thread
From: Alejandro Lucero @ 2015-09-18 10:33 UTC (permalink / raw)
To: dev
I have seen the API definition says nothing about accuracy but some PMD
implementations sacrifice accuracy for the sake of performance. If I'm not
understanding the code wrongly i40e and ixgbe check DD bit just for the
first descriptor in a group of 4, and they take all of them as used if the
first descriptor is used.
By other hand, they do a "heavy" calculation when the descriptor ring wraps
which does not make sense (to me) if same performance goal is used.
There are PMDs not supporting this option and I can not see any app or
example using it so I do not know how important is this function, its
accuracy and its performance impact. Can someone comment on this?
Thanks
^ permalink raw reply [flat|nested] 3+ messages in thread
* Re: [dpdk-dev] rte_eth_rx_queue_count accuracy
2015-09-18 10:33 [dpdk-dev] rte_eth_rx_queue_count accuracy Alejandro Lucero
@ 2015-09-21 22:41 ` Stephen Hemminger
2015-09-22 10:40 ` Alejandro Lucero
0 siblings, 1 reply; 3+ messages in thread
From: Stephen Hemminger @ 2015-09-21 22:41 UTC (permalink / raw)
To: Alejandro Lucero; +Cc: dev
On Fri, 18 Sep 2015 11:33:36 +0100
Alejandro Lucero <alejandro.lucero@netronome.com> wrote:
> I have seen the API definition says nothing about accuracy but some PMD
> implementations sacrifice accuracy for the sake of performance. If I'm not
> understanding the code wrongly i40e and ixgbe check DD bit just for the
> first descriptor in a group of 4, and they take all of them as used if the
> first descriptor is used.
>
> By other hand, they do a "heavy" calculation when the descriptor ring wraps
> which does not make sense (to me) if same performance goal is used.
>
> There are PMDs not supporting this option and I can not see any app or
> example using it so I do not know how important is this function, its
> accuracy and its performance impact. Can someone comment on this?
>
> Thanks
I have version of this for virtio/vmxnet3
It is useful when using the interrupt control mode.
^ permalink raw reply [flat|nested] 3+ messages in thread
* Re: [dpdk-dev] rte_eth_rx_queue_count accuracy
2015-09-21 22:41 ` Stephen Hemminger
@ 2015-09-22 10:40 ` Alejandro Lucero
0 siblings, 0 replies; 3+ messages in thread
From: Alejandro Lucero @ 2015-09-22 10:40 UTC (permalink / raw)
To: Stephen Hemminger; +Cc: dev
I can not see that code. Can you point out where is it?
Thanks
On Mon, Sep 21, 2015 at 11:41 PM, Stephen Hemminger <
stephen@networkplumber.org> wrote:
> On Fri, 18 Sep 2015 11:33:36 +0100
> Alejandro Lucero <alejandro.lucero@netronome.com> wrote:
>
> > I have seen the API definition says nothing about accuracy but some PMD
> > implementations sacrifice accuracy for the sake of performance. If I'm
> not
> > understanding the code wrongly i40e and ixgbe check DD bit just for the
> > first descriptor in a group of 4, and they take all of them as used if
> the
> > first descriptor is used.
> >
> > By other hand, they do a "heavy" calculation when the descriptor ring
> wraps
> > which does not make sense (to me) if same performance goal is used.
> >
> > There are PMDs not supporting this option and I can not see any app or
> > example using it so I do not know how important is this function, its
> > accuracy and its performance impact. Can someone comment on this?
> >
> > Thanks
>
> I have version of this for virtio/vmxnet3
> It is useful when using the interrupt control mode.
>
^ permalink raw reply [flat|nested] 3+ messages in thread
end of thread, other threads:[~2015-09-22 10:40 UTC | newest]
Thread overview: 3+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2015-09-18 10:33 [dpdk-dev] rte_eth_rx_queue_count accuracy Alejandro Lucero
2015-09-21 22:41 ` Stephen Hemminger
2015-09-22 10:40 ` Alejandro Lucero
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).