netdev - Re: [RFC PATCH 0/4] inet: add second hash table

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

Message-ID: <CAEnQRZD3o_+fRnnbd74VeFuNvjAVVyq-rE241J96iRXWFDAEPQ@mail.gmail.com>
Date:	Wed, 30 May 2012 15:32:44 +0300
From:	Daniel Baluta <dbaluta@...acom.com>
To:	Eric Dumazet <eric.dumazet@...il.com>
Cc:	Alexandru Copot <alex.mihai.c@...il.com>, davem@...emloft.net,
	gerrit@....abdn.ac.uk, kuznet@....inr.ac.ru, jmorris@...ei.org,
	yoshfuji@...ux-ipv6.org, kaber@...sh.net, netdev@...r.kernel.org,
	Lucian Grijincu <lucian.grijincu@...il.com>
Subject: Re: [RFC PATCH 0/4] inet: add second hash table

On Wed, May 30, 2012 at 10:57 AM, Eric Dumazet <eric.dumazet@...il.com> wrote:
> On Wed, 2012-05-30 at 10:36 +0300, Alexandru Copot wrote:
>> This patchset implements all the operations needed to use a second
>> (port,address) bind hash table for inet. It uses a similar approach
>> as the UDP implementation.
>>
>> The performance improvements for port allocation are very good and
>> detailed in the last message.
>>
>> This is based on a series of patches written by Lucian Grijincu at Ixia.
>>
>> Signed-off-by: Alexandru Copot <alex.mihai.c@...il.com>
>> Cc: Daniel Baluta <dbaluta@...acom.com>
>> Cc: Lucian Grijincu <lucian.grijincu@...il.com>
>> ---
>> Alexandru Copot (4):
>>       inet: add counter to inet_bind_hashbucket
>>       inet: add a second bind hash
>>       inet: add/remove inet buckets in the second bind hash
>>       inet: use second hash in inet_csk_get_port
>>
>>  include/net/inet_hashtables.h    |  140 +++++++++++++++++++++++++++++++--
>>  include/net/inet_timewait_sock.h |    5 +-
>>  net/dccp/proto.c                 |   37 ++++++++-
>>  net/ipv4/inet_connection_sock.c  |   66 ++++++++--------
>>  net/ipv4/inet_hashtables.c       |  158 ++++++++++++++++++++++++++++++++++++--
>>  net/ipv4/inet_timewait_sock.c    |   16 ++--
>>  net/ipv4/tcp.c                   |   17 ++++
>>  net/ipv6/inet6_hashtables.c      |   95 +++++++++++++++++++++++
>>  8 files changed, 477 insertions(+), 57 deletions(-)
>
>
> Its a huge change (with many details to look at), for a yet to be
> understood need.
>
> What sensible workload needs this at all ?

Hi Eric,

Usually our tests use a huge number of virtual interfaces.
Using this patch we get a massive improvement when there are many sockets
bound to the same port, but different addresses for both bind() and
listen() system calls (both call inet_csk_get_port).
	
We provided some data points in the fourth patch:

For 16.000 interfaces each with a distinct IPv4 address, doing bind
and then listen we get:

* Without patch and without SO_REUSEADDR:
   * bind:   1.543 s
   * listen: 3.050 s

* Without patch and with SO_REUSEADDR set:
   * bind:   0.066 s
   * listen: 3.050 s

* With patch and SO_REUSEADDR set / without SO_REUSEADDR:
   * bind:   0.066 s
   * listen: 0.095 s

The source code for tests can be found here [1].
Just run:
* ./prepare_test2.sh
* ./avg_tcp.sh

If I understood it correctly, a similar patch was introduced
for UDP some time ago. [2]

thanks,
Daniel.

[1] http://ixlabs.cs.pub.ro/gitweb/?p=port-allocation.git;a=tree;f=testbind;h=687e4452101e13cb5995b43c1351d76786d98fdd;hb=HEAD
[2] http://www.spinics.net/lists/netdev/msg112056.html
--
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