netdev - Re: RSS mode performance in ixgbe (82599) driver

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [day] [month] [year] [list]

Message-ID: <1312893808.2371.19.camel@edumazet-HP-Compaq-6005-Pro-SFF-PC>
Date:	Tue, 09 Aug 2011 14:43:28 +0200
From:	Eric Dumazet <eric.dumazet@...il.com>
To:	"J.Hwan Kim" <frog1120@...il.com>
Cc:	linux-kernel@...r.kernel.org, netdev <netdev@...r.kernel.org>
Subject: Re: RSS mode performance in ixgbe (82599) driver

Le mardi 09 août 2011 à 21:03 +0900, J.Hwan Kim a écrit :
> Hi, everyone
> 
> I'm testing the RSS mode throughput of Intel ixgbe 82599 driver.
> My server's specification is Xeon 3GHz (4cores) and PCIe speed is 
> 2.5Gx8lanes,
> and DD2 memory is about 600~700MHz.
> 
> When I enable the RSS(receive side scaling) mode with 4 Rx queues,
> the Rx performance is below than when I use a single queue.
> 
> I don't know how the cores operate, whether the cores run in sequence
> or simultaneously at the same time.
> 
> I want to know the mechanism in general how the multi-cores run
> and the reason why the performance of ixgbe driver in rss mode is below
> non-rss mode.

CC netdev, as these kind of stuff is better hosted on netdev

You should make sure your 4 rx queues are serviced by different cpus, or
else, one cpu could be the bottleneck.

cat /proc/interrupts



--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html