netdev - [RFC PATCH 12/11] netlink: improve flow control and speed up dumps with memory mapped I/O

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: <4E68EF19.8070305@trash.net>
Date:	Thu, 08 Sep 2011 18:36:41 +0200
From:	Patrick McHardy <kaber@...sh.net>
To:	"netfilter-devel@...r.kernel.org" <netfilter-devel@...r.kernel.org>
CC:	NetDev <netdev@...r.kernel.org>
Subject: [RFC PATCH 12/11] netlink: improve flow control and speed up dumps
 with memory mapped I/O

This patch on top of the memory mapped netlink I/O series is
meant to speed up dumps by dumping messages more aggressively.

Messages are dumped until half the ring is used, this is similar
to flow-control used with regular I/O with the difference that
all messages are constructed during a single poll call instead
of multiple recvmsg() calls.

The result is not as good as I expected, dumping 196611
routes using rtnl-route-dump from libmnl show the following
results (averaged over 10 runs):

user-time: -7.9%
system-time: -10.4%
real-time: +0.09%
CPU: -7.45%

Strangely the real time stays almost the same and CPU usage
decreases by a few percent - I would have expected a reduction
of real-time with identical CPU usage (99%). Not sure what
the reason is, will try to investigate further.

View attachment "nldump.diff" of type "text/plain" (1952 bytes)