DPDK patches and discussions
 help / color / mirror / Atom feed
From: Ferruh Yigit <ferruh.yigit@intel.com>
To: Honnappa Nagarahalli <honnappa.nagarahalli@arm.com>,
	dev@dpdk.org, alialnu@mellanox.com, orgerlitz@mellanox.com,
	wenzhuo.lu@intel.com, beilei.xing@intel.com,
	bernard.iremonger@intel.com
Cc: hemant.agrawal@nxp.com, jerinj@marvell.com,
	viacheslavo@mellanox.com, thomas@monjalon.net,
	ruifeng.wang@arm.com, phil.yang@arm.com, nd@arm.com,
	zhihong.wang@intel.com, stable@dpdk.org
Subject: Re: [dpdk-dev] [PATCH v2 1/5] app/testpmd: clock gettime call in throughput calculation
Date: Mon, 6 Jul 2020 16:36:51 +0100	[thread overview]
Message-ID: <6dd1c866-8e41-4952-9792-b1fd492d0908@intel.com> (raw)
In-Reply-To: <20200626220943.22526-1-honnappa.nagarahalli@arm.com>

On 6/26/2020 11:09 PM, Honnappa Nagarahalli wrote:
> The throughput calculation requires a counter that measures
> passing of time. However, the kernel saves and restores the PMU
> state when a thread is unscheduled and scheduled. This ensures
> that the PMU cycles are not counted towards a thread that is
> not scheduled. Hence, when RTE_ARM_EAL_RDTSC_USE_PMU is enabled,
> the PMU cycles do not represent the passing of time.

Does this mean 'rte_rdtsc()' is broken for Arm?
Wouldn't it cause more serious issue than testpmd throughput stats?

> This results in incorrect calculation of throughput numbers.
> Use clock_gettime system call to calculate the time passed since
> last call.
> 
> Bugzilla ID: 450
> Fixes: 0e106980301d ("app/testpmd: show throughput in port stats")
> Cc: zhihong.wang@intel.com
> Cc: stable@dpdk.org
> 
> Signed-off-by: Honnappa Nagarahalli <honnappa.nagarahalli@arm.com>
> Reviewed-by: Phil Yang <phil.yang@arm.com>
> Reviewed-by: Ruifeng Wang <ruifeng.wang@arm.com>
> Tested-by: Phil Yang <phil.yang@arm.com>
> Tested-by: Ali Alnubani <alialnu@mellanox.com>

<...>

> @@ -195,10 +204,17 @@ nic_stats_display(portid_t port_id)
>  		}
>  	}
>  
> -	diff_cycles = prev_cycles[port_id];
> -	prev_cycles[port_id] = rte_rdtsc();
> -	if (diff_cycles > 0)
> -		diff_cycles = prev_cycles[port_id] - diff_cycles;
> +	diff_ns = 0;
> +	if (clock_gettime(CLOCK_TYPE_ID, &cur_time) == 0) {

I guess 'rte_rdtsc()' is faster, but since this is not in the data path, I think
it is OK.

<...>

> @@ -217,10 +233,10 @@ nic_stats_display(portid_t port_id)
>  		(stats.obytes - prev_bytes_tx[port_id]) : 0;
>  	prev_bytes_rx[port_id] = stats.ibytes;
>  	prev_bytes_tx[port_id] = stats.obytes;
> -	mbps_rx = diff_cycles > 0 ?
> -		diff_bytes_rx * rte_get_tsc_hz() / diff_cycles : 0;
> -	mbps_tx = diff_cycles > 0 ?
> -		diff_bytes_tx * rte_get_tsc_hz() / diff_cycles : 0;
> +	mbps_rx = diff_ns > 0 ?
> +		(double)diff_bytes_rx / diff_ns * NS_PER_SEC : 0;
> +	mbps_tx = diff_ns > 0 ?
> +		(double)diff_bytes_tx / diff_ns * NS_PER_SEC : 0;

The calculation also fixes other issue in the stats.
With previous method, if the sampling between two stats is a little long,
"diff_pkts_rx * rte_get_tsc_hz()" can overflow and produce wrong 'bps'.


Reviewed-by: Ferruh Yigit <ferruh.yigit@intel.com>

  parent reply	other threads:[~2020-07-06 15:36 UTC|newest]

Thread overview: 40+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2020-06-17 14:43 [dpdk-dev] [PATCH " Honnappa Nagarahalli
2020-06-17 14:43 ` [dpdk-dev] [PATCH 2/5] app/testpmd: enable burst stats for flowgen mode rx path Honnappa Nagarahalli
2020-06-18 14:57   ` Phil Yang
2020-06-17 14:43 ` [dpdk-dev] [PATCH 3/5] app/testpmd: enable burst stats for noisy vnf mode Honnappa Nagarahalli
2020-06-18  7:16   ` Jens Freimann
2020-06-17 14:43 ` [dpdk-dev] [PATCH 4/5] app/testpmd: fix burst percentage calculation Honnappa Nagarahalli
2020-06-18 15:01   ` Phil Yang
2020-06-23 13:33   ` Ali Alnubani
2020-06-17 14:43 ` [dpdk-dev] [PATCH 5/5] app/testpmd: enable empty polls in burst stats Honnappa Nagarahalli
2020-06-18 15:05   ` Phil Yang
2020-06-23 13:34   ` Ali Alnubani
2020-06-17 15:16 ` [dpdk-dev] [PATCH 1/5] app/testpmd: clock gettime call in throughput calculation Jerin Jacob
2020-06-18  4:03   ` Honnappa Nagarahalli
2020-06-18 10:16     ` Jerin Jacob
2020-06-18 15:08 ` Phil Yang
2020-06-23 13:33 ` Ali Alnubani
2020-06-26 22:09 ` [dpdk-dev] [PATCH v2 " Honnappa Nagarahalli
2020-06-26 22:09   ` [dpdk-dev] [PATCH v2 2/5] app/testpmd: enable burst stats for flowgen mode rx path Honnappa Nagarahalli
2020-07-06 16:02     ` Ferruh Yigit
2020-07-06 17:06       ` Honnappa Nagarahalli
2020-07-06 17:17         ` Ferruh Yigit
2020-07-06 18:46           ` Honnappa Nagarahalli
2020-06-26 22:09   ` [dpdk-dev] [PATCH v2 3/5] app/testpmd: enable burst stats for noisy vnf mode Honnappa Nagarahalli
2020-07-06 16:08     ` Ferruh Yigit
2020-07-06 16:59       ` Honnappa Nagarahalli
2020-07-06 17:11         ` Ferruh Yigit
2020-07-06 17:41           ` Honnappa Nagarahalli
2020-06-26 22:09   ` [dpdk-dev] [PATCH v2 4/5] app/testpmd: fix burst percentage calculation Honnappa Nagarahalli
2020-07-06 16:10     ` Ferruh Yigit
2020-06-26 22:09   ` [dpdk-dev] [PATCH v2 5/5] app/testpmd: enable empty polls in burst stats Honnappa Nagarahalli
2020-07-06 17:05     ` Ferruh Yigit
2020-07-06 19:11       ` Honnappa Nagarahalli
2020-07-07  9:32         ` Ferruh Yigit
2020-07-06 15:36   ` Ferruh Yigit [this message]
2020-07-06 16:53     ` [dpdk-dev] [PATCH v2 1/5] app/testpmd: clock gettime call in throughput calculation Honnappa Nagarahalli
2020-07-06 17:19       ` Ferruh Yigit
2020-07-06 23:32 ` [dpdk-dev] [PATCH v3 1/3] " Honnappa Nagarahalli
2020-07-06 23:32   ` [dpdk-dev] [PATCH v3 2/3] app/testpmd: fix burst percentage calculation Honnappa Nagarahalli
2020-07-06 23:32   ` [dpdk-dev] [PATCH v3 3/3] app/testpmd: enable empty polls in burst stats Honnappa Nagarahalli
2020-07-07 17:47   ` [dpdk-dev] [PATCH v3 1/3] app/testpmd: clock gettime call in throughput calculation Ferruh Yigit

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=6dd1c866-8e41-4952-9792-b1fd492d0908@intel.com \
    --to=ferruh.yigit@intel.com \
    --cc=alialnu@mellanox.com \
    --cc=beilei.xing@intel.com \
    --cc=bernard.iremonger@intel.com \
    --cc=dev@dpdk.org \
    --cc=hemant.agrawal@nxp.com \
    --cc=honnappa.nagarahalli@arm.com \
    --cc=jerinj@marvell.com \
    --cc=nd@arm.com \
    --cc=orgerlitz@mellanox.com \
    --cc=phil.yang@arm.com \
    --cc=ruifeng.wang@arm.com \
    --cc=stable@dpdk.org \
    --cc=thomas@monjalon.net \
    --cc=viacheslavo@mellanox.com \
    --cc=wenzhuo.lu@intel.com \
    --cc=zhihong.wang@intel.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
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).