[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <20240807201857.445f9f95@kernel.org>
Date: Wed, 7 Aug 2024 20:18:57 -0700
From: Jakub Kicinski <kuba@...nel.org>
To: Erni Sri Satya Vennela <ernis@...ux.microsoft.com>
Cc: kys@...rosoft.com, haiyangz@...rosoft.com, wei.liu@...nel.org,
decui@...rosoft.com, davem@...emloft.net, edumazet@...gle.com,
pabeni@...hat.com, linux-hyperv@...r.kernel.org, netdev@...r.kernel.org,
linux-kernel@...r.kernel.org, ernis@...rosoft.com
Subject: Re: [PATCH] net: netvsc: Increase default VMBus channel from 8 to
16
On Mon, 5 Aug 2024 22:55:51 -0700 Erni Sri Satya Vennela wrote:
> Performance tests showed significant improvement in throughput:
> - 0.54% for 16 vCPUs
> - 1.51% for 32 vCPUs
> - 0.72% for 48 vCPUs
> - 5.57% for 64 vCPUs
> - 9.14% for 96 vCPUs
Could you please switch to netif_get_num_default_rss_queues() ?
It used to return hard coded 8, but now it returns #physical cores / 2.
That's based on broad experience with Meta's workloads. Some workloads
need more some needs fewer, but broadly half of physical cores is
a good guess for 90%+
Assuming you have thread siblings in those vCPUs above it should
match what you want, too.
--
pw-bot: cr
Powered by blists - more mailing lists