On 24/03/2022 13:05, Zhang, Qi Z wrote:
-----Original Message-----
From: Kevin Traynor <ktray...@redhat.com>
Sent: Thursday, March 24, 2022 8:18 PM
To: Zhang, Qi Z <qi.z.zh...@intel.com>; Wu, Wenjun1
<wenjun1...@intel.com>; dev@dpdk.org; Yang, Qiming
<qiming.y...@intel.com>
Cc: Van Haaren, Harry <harry.van.haa...@intel.com>; Su, Simei
<simei...@intel.com>; Luca Boccassi <bl...@debian.org>; Christian Ehrhardt
<christian.ehrha...@canonical.com>
Subject: Re: [PATCH v4] net/ice: improve performance of RX timestamp
offload
On 24/03/2022 11:51, Zhang, Qi Z wrote:
-----Original Message-----
From: Kevin Traynor <ktray...@redhat.com>
Sent: Thursday, March 24, 2022 7:17 PM
To: Zhang, Qi Z <qi.z.zh...@intel.com>; Wu, Wenjun1
<wenjun1...@intel.com>; dev@dpdk.org; Yang, Qiming
<qiming.y...@intel.com>
Cc: Van Haaren, Harry <harry.van.haa...@intel.com>; Su, Simei
<simei...@intel.com>; Luca Boccassi <bl...@debian.org>; Christian
Ehrhardt <christian.ehrha...@canonical.com>
Subject: Re: [PATCH v4] net/ice: improve performance of RX timestamp
offload
On 24/03/2022 09:09, Zhang, Qi Z wrote:
-----Original Message-----
From: Wu, Wenjun1 <wenjun1...@intel.com>
Sent: Monday, February 28, 2022 3:36 PM
To: dev@dpdk.org; Zhang, Qi Z <qi.z.zh...@intel.com>; Yang, Qiming
<qiming.y...@intel.com>
Cc: Van Haaren, Harry <harry.van.haa...@intel.com>; Su, Simei
<simei...@intel.com>; Wu, Wenjun1 <wenjun1...@intel.com>
Subject: [PATCH v4] net/ice: improve performance of RX timestamp
offload
Previously, each time a burst of packets is received, SW reads HW
register and assembles it and the timestamp from descriptor
together to get the complete 64 bits timestamp.
This patch optimizes the algorithm. The SW only needs to check the
monotonicity of the low 32bits timestamp to avoid crossing borders.
Each time before SW receives a burst of packets, it should check
the time difference between current time and last update time to
avoid the low 32 bits timestamp cycling twice.
Signed-off-by: Wenjun Wu <wenjun1...@intel.com>
Added cc stable
Hi Qi. The DPDK documentation has guidance about what should be
backported to LTS [0] and distinguishes between fixes and performance
improvements. Please try and stick with this when applying patches or
let LTS maintainers know if there is a debatable case.
Thanks for the comments
Yes, actually this is about a 50% ~ 70% performance improvement, which
maybe critical for some performance sensitive use cases.(e.g. network
forensics) So I'd like to defend with below case
An existing feature in LTS is not usable as intended without it.
If that is the case, then I think the commitlog should be re-written. It just
talks about the code changes, there's nothing about an impact to an existing
use case that was unusuable and is now fixed.
OK, I updated with below commit log in dpdk-next-net-intel.
Kevin , Wenjun, let me know if anything I missed.
Previously, each time a burst of packets is received, SW reads HW
register and assembles it and the timestamp from descriptor together to
get the complete 64 bits timestamp.
This patch optimizes the algorithm. The SW only needs to check the
monotonicity of the low 32bits timestamp to avoid crossing borders.
Each time before SW receives a burst of packets, it should check the
time difference between current time and last update time to avoid
the low 32 bits timestamp cycling twice.
The patch proved a 50% ~ 70% single core performance improvement on a
main stream Xeon server, and it is necessary to be backport to LTS release,
as this fix the performance gap for some use cases.
The comment about backport to LTS can be removed. That is an operational
comment, so not really relevant to a future reader trying to understand
the commit.
thanks,
Kevin.
Thanks
Qi
Thanks
Qi
thanks,
Kevin.
[0]
http://doc.dpdk.org/guides-21.11/contributing/stable.html#what-change
s-
should-be-backported
Acked-by: Qi Zhang <qi.z.zh...@intel.com>
Applied to dpdk-next-net-intel.
Thanks
Qi
