[dts] [PATCH] tests: clean and reset the hugepages to fix cryptodev perf drop

Zhao, XinfengX xinfengx.zhao at intel.com
Tue May 19 04:20:19 CEST 2020


There is enough hugepage left, but the perf of scheduler+qat dropped.

and add this action, the issue disappeared.

-----Original Message-----
From: Tu, Lijuan 
Sent: Tuesday, May 19, 2020 9:53 AM
To: Zhao, XinfengX <xinfengx.zhao at intel.com>; dts at dpdk.org
Cc: Zhao, XinfengX <xinfengx.zhao at intel.com>
Subject: RE: [dts] [PATCH] tests: clean and reset the hugepages to fix cryptodev perf drop

Why do you add this? Is there anyone occupied many hugepages and don't release them?

> -----Original Message-----
> From: dts [mailto:dts-bounces at dpdk.org] On Behalf Of xinfengx
> Sent: Friday, May 15, 2020 2:38 AM
> To: dts at dpdk.org
> Cc: Zhao, XinfengX <xinfengx.zhao at intel.com>
> Subject: [dts] [PATCH] tests: clean and reset the hugepages to fix 
> cryptodev perf drop
> 
> Signed-off-by: xinfengx <xinfengx.zhao at intel.com>
> ---
>  tests/TestSuite_crypto_perf_cryptodev_perf.py | 5 +++++
>  1 file changed, 5 insertions(+)
> 
> diff --git a/tests/TestSuite_crypto_perf_cryptodev_perf.py
> b/tests/TestSuite_crypto_perf_cryptodev_perf.py
> index defb7c2..910c81a 100644
> --- a/tests/TestSuite_crypto_perf_cryptodev_perf.py
> +++ b/tests/TestSuite_crypto_perf_cryptodev_perf.py
> @@ -71,6 +71,11 @@ class PerfTestsCryptodev(TestCase):
>          }
>          self._app_path = 
> "./x86_64-native-linuxapp-gcc/build/app/test-crypto-
> perf/dpdk-test-crypto-perf"
> 
> +        page_size = self.dut.send_expect("awk '/Hugepagesize/ {print $2}'
> /proc/meminfo", "# ")
> +        if int(page_size) == 1024 * 1024:
> +            self.dut.send_expect('echo 0 >
> /sys/kernel/mm/hugepages/hugepages-%skB/nr_hugepages' % (page_size), 
> '# ', 5)
> +            self.dut.send_expect('echo 16 >
> /sys/kernel/mm/hugepages/hugepages-%skB/nr_hugepages' % (page_size), 
> '# ', 5)
> +
>          if not cc.is_build_skip(self):
>              cc.build_dpdk_with_cryptodev(self)
> 
> --
> 2.17.1



More information about the dts mailing list