From: Eric Dumazet <eric.dumazet@gmail.com>
To: Tom Herbert <therbert@google.com>
Cc: hadi@cyberus.ca, Stephen Hemminger <shemminger@vyatta.com>,
netdev@vger.kernel.org, robert@herjulf.net,
David Miller <davem@davemloft.net>,
Changli Gao <xiaosuo@gmail.com>, Andi Kleen <andi@firstfloor.org>
Subject: Re: rps perfomance WAS(Re: rps: question
Date: Wed, 14 Apr 2010 22:57:41 +0200 [thread overview]
Message-ID: <1271278661.16881.1761.camel@edumazet-laptop> (raw)
In-Reply-To: <z2y65634d661004141345o9bdcf759sf266866931823baf@mail.gmail.com>
Le mercredi 14 avril 2010 à 13:45 -0700, Tom Herbert a écrit :
> > Only if more than one flow is involved.
> >
> > And if you have many flows, chance they will spread several queues...
> >
>
> But use too many queues and the efficiency of NAPI drops and cost of
> device interrupts becomes dominant, so that the overhead from
> additional hard interrupts can surpass the overhead of doing RPS and
> the IPIs. I believe we are seeing this is in some of our results
> which shows that a combination of multi-queue and RPS can be better
> than just multi-queue (see rps changelog). Again, I'm not claiming
> that is generally true, but there are a lot of factors to consider.
> --
RPS can be tuned (Changli wants a finer tuning...), it would be
intereting to tune multiqueue devices too. I dont know if its possible
right now.
On my Nehalem machine (16 logical cpus), its NetXtreme II BCM57711E
10Gigabit has 16 queues. It might be good to use less queues according
to your results on some workloads, and eventually use RPS on a second
layering.
next prev parent reply other threads:[~2010-04-14 20:57 UTC|newest]
Thread overview: 86+ messages / expand[flat|nested] mbox.gz Atom feed top
2010-02-07 18:42 rps: question jamal
2010-02-08 5:58 ` Tom Herbert
2010-02-08 15:09 ` jamal
2010-04-14 11:53 ` rps perfomance WAS(Re: " jamal
2010-04-14 17:31 ` Tom Herbert
2010-04-14 18:04 ` Eric Dumazet
2010-04-14 18:53 ` jamal
2010-04-14 19:44 ` Stephen Hemminger
2010-04-14 19:58 ` Eric Dumazet
2010-04-15 8:51 ` David Miller
2010-04-14 20:22 ` jamal
2010-04-14 20:27 ` Eric Dumazet
2010-04-14 20:38 ` jamal
2010-04-14 20:45 ` Tom Herbert
2010-04-14 20:57 ` Eric Dumazet [this message]
2010-04-14 22:51 ` Changli Gao
2010-04-14 23:02 ` Stephen Hemminger
2010-04-15 2:40 ` Eric Dumazet
2010-04-15 2:50 ` Changli Gao
2010-04-15 8:57 ` David Miller
2010-04-15 12:10 ` jamal
2010-04-15 12:32 ` Changli Gao
2010-04-15 12:50 ` jamal
2010-04-15 23:51 ` Changli Gao
2010-04-15 8:51 ` David Miller
2010-04-14 20:34 ` Andi Kleen
2010-04-15 8:50 ` David Miller
2010-04-15 8:48 ` David Miller
2010-04-15 11:55 ` jamal
2010-04-15 16:41 ` Rick Jones
2010-04-15 20:16 ` jamal
2010-04-15 20:25 ` Rick Jones
2010-04-15 23:56 ` Changli Gao
2010-04-16 5:18 ` Eric Dumazet
2010-04-16 6:02 ` Changli Gao
2010-04-16 6:28 ` Tom Herbert
2010-04-16 6:32 ` Eric Dumazet
2010-04-16 13:42 ` jamal
2010-04-16 7:15 ` Andi Kleen
2010-04-16 13:27 ` jamal
2010-04-16 13:37 ` Andi Kleen
2010-04-16 13:58 ` jamal
2010-04-16 13:21 ` jamal
2010-04-16 13:34 ` Changli Gao
2010-04-16 13:49 ` jamal
2010-04-16 14:10 ` Changli Gao
2010-04-16 14:43 ` jamal
2010-04-16 14:58 ` Changli Gao
2010-04-19 12:48 ` jamal
2010-04-17 7:35 ` Eric Dumazet
2010-04-17 8:43 ` Tom Herbert
2010-04-17 9:23 ` Eric Dumazet
2010-04-17 14:27 ` Eric Dumazet
2010-04-17 17:26 ` Tom Herbert
2010-04-17 14:17 ` [PATCH net-next-2.6] net: remove time limit in process_backlog() Eric Dumazet
2010-04-18 9:36 ` David Miller
2010-04-17 17:31 ` rps perfomance WAS(Re: rps: question jamal
2010-04-18 9:39 ` Eric Dumazet
2010-04-18 11:34 ` Eric Dumazet
2010-04-19 2:09 ` jamal
2010-04-19 9:37 ` [RFC] rps: shortcut net_rps_action() Eric Dumazet
2010-04-19 9:48 ` Changli Gao
2010-04-19 12:14 ` Eric Dumazet
2010-04-19 12:28 ` Changli Gao
2010-04-19 13:27 ` Eric Dumazet
2010-04-19 14:22 ` Eric Dumazet
2010-04-19 15:07 ` [PATCH net-next-2.6] " Eric Dumazet
2010-04-19 16:02 ` Tom Herbert
2010-04-19 20:21 ` David Miller
2010-04-20 7:17 ` [PATCH net-next-2.6] rps: cleanups Eric Dumazet
2010-04-20 8:18 ` David Miller
2010-04-19 23:56 ` [PATCH net-next-2.6] rps: shortcut net_rps_action() Changli Gao
2010-04-20 0:32 ` Changli Gao
2010-04-20 5:55 ` Eric Dumazet
2010-04-20 12:02 ` rps perfomance WAS(Re: rps: question jamal
2010-04-20 13:13 ` Eric Dumazet
[not found] ` <1271853570.4032.21.camel@bigi>
2010-04-21 19:01 ` Eric Dumazet
2010-04-22 1:27 ` Changli Gao
2010-04-22 12:12 ` jamal
2010-04-25 2:31 ` Changli Gao
2010-04-26 11:35 ` jamal
2010-04-26 13:35 ` Changli Gao
2010-04-21 21:53 ` Rick Jones
2010-04-16 15:57 ` Tom Herbert
2010-04-14 18:53 ` Stephen Hemminger
2010-04-15 8:42 ` David Miller
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=1271278661.16881.1761.camel@edumazet-laptop \
--to=eric.dumazet@gmail.com \
--cc=andi@firstfloor.org \
--cc=davem@davemloft.net \
--cc=hadi@cyberus.ca \
--cc=netdev@vger.kernel.org \
--cc=robert@herjulf.net \
--cc=shemminger@vyatta.com \
--cc=therbert@google.com \
--cc=xiaosuo@gmail.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 an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.