All of lore.kernel.org
 help / color / mirror / Atom feed
From: Eric Dumazet <eric.dumazet@gmail.com>
To: Alexander Duyck <alexander.h.duyck@intel.com>
Cc: David Miller <davem@davemloft.net>, netdev <netdev@vger.kernel.org>
Subject: Re: Performance regression on kernels 3.10 and newer
Date: Thu, 14 Aug 2014 12:50:31 -0700	[thread overview]
Message-ID: <1408045831.6804.34.camel@edumazet-glaptop2.roam.corp.google.com> (raw)
In-Reply-To: <1408041962.6804.31.camel@edumazet-glaptop2.roam.corp.google.com>

On Thu, 2014-08-14 at 11:46 -0700, Eric Dumazet wrote:

> I believe you answered your own question : prequeue mode does not work
> very well when one host has hundred of active TCP flows to one other.
> 
> In real life, applications do not use prequeue, because nobody wants one
> thread per flow.
> 
> Each socket has its own dst now route cache was removed, but if your
> netperf migrates cpu (and NUMA node), we do not detect the dst should be
> re-created onto a different NUMA node.
> 
> But really, I am not sure we want to care about prequeue, as modern
> applications uses epoll()/poll()/select() instead of blocking on
> recvmsg()

BTW it looks like you do not use RPS/RFS/RSS, so you have different cpus
doing the dst refcount increment and decrement.

With RFS, your workload just works fine.

  reply	other threads:[~2014-08-14 19:50 UTC|newest]

Thread overview: 57+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2014-08-14 18:19 Performance regression on kernels 3.10 and newer Alexander Duyck
2014-08-14 18:46 ` Eric Dumazet
2014-08-14 19:50   ` Eric Dumazet [this message]
2014-08-14 19:59   ` Rick Jones
2014-08-14 20:31     ` Alexander Duyck
2014-08-14 20:51       ` Eric Dumazet
2014-08-14 20:46     ` Eric Dumazet
2014-08-14 23:16   ` Alexander Duyck
2014-08-14 23:20     ` David Miller
2014-08-14 23:25       ` Tom Herbert
2014-08-21 23:24         ` David Miller
2014-09-06 14:45           ` Eric Dumazet
2014-09-06 15:27             ` Eric Dumazet
2014-09-06 15:46               ` Eric Dumazet
2014-09-06 16:38                 ` Eric Dumazet
2014-09-06 18:21                   ` Eric Dumazet
2014-09-07 19:05                     ` [PATCH net] ipv6: refresh rt6i_genid in ip6_pol_route() Eric Dumazet
2014-09-07 22:54                       ` David Miller
2014-09-08  4:18                         ` Eric Dumazet
2014-09-08  4:27                           ` David Miller
2014-09-08  4:43                             ` Eric Dumazet
2014-09-08  4:59                               ` David Miller
2014-09-08  5:07                                 ` Eric Dumazet
2014-09-08  8:11                                   ` Nicolas Dichtel
2014-09-08 10:28                                     ` Eric Dumazet
2014-09-08 12:16                                       ` Nicolas Dichtel
2014-09-08 18:48                                   ` Vlad Yasevich
2014-09-09 12:58                                   ` Hannes Frederic Sowa
2014-09-10  9:31                                     ` [PATCH net-next] ipv6: implement rt_genid_bump_ipv6 with fn_sernum and remove rt6i_genid Hannes Frederic Sowa
2014-09-10 13:26                                       ` Vlad Yasevich
2014-09-10 13:42                                         ` Hannes Frederic Sowa
2014-09-10 20:09                                       ` David Miller
2014-09-11  8:30                                         ` Hannes Frederic Sowa
2014-09-11 12:22                                           ` Vlad Yasevich
2014-09-11 12:40                                             ` Hannes Frederic Sowa
2014-09-11 12:05                                         ` Hannes Frederic Sowa
2014-09-11 14:19                                           ` Vlad Yasevich
2014-09-11 14:32                                             ` Hannes Frederic Sowa
2014-09-11 14:44                                               ` Vlad Yasevich
2014-09-11 14:47                                                 ` Hannes Frederic Sowa
2014-09-08 15:06               ` [PATCH v2 net-next] tcp: remove dst refcount false sharing for prequeue mode Eric Dumazet
2014-09-08 21:21                 ` David Miller
2014-09-08 21:30                   ` Eric Dumazet
2014-09-08 22:41                     ` David Miller
2014-09-09 23:56                     ` David Miller
2014-08-15 17:15       ` Performance regression on kernels 3.10 and newer Alexander Duyck
2014-08-15 17:59         ` Eric Dumazet
2014-08-15 18:49         ` Tom Herbert
2014-08-15 19:10           ` Alexander Duyck
2014-08-15 22:16             ` Tom Herbert
2014-08-15 23:23               ` Alexander Duyck
2014-08-18  9:03                 ` David Laight
2014-08-18 15:22                   ` Alexander Duyck
2014-08-18 15:29                     ` Rick Jones
2014-08-21 23:51         ` David Miller
2014-08-14 23:48     ` Eric Dumazet
2014-08-15  0:33       ` Rick Jones

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=1408045831.6804.34.camel@edumazet-glaptop2.roam.corp.google.com \
    --to=eric.dumazet@gmail.com \
    --cc=alexander.h.duyck@intel.com \
    --cc=davem@davemloft.net \
    --cc=netdev@vger.kernel.org \
    /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.