Date: Tue, 01 Dec 2009 16:08:09 -0800 From: Julian Elischer <julian@elischer.org> To: Brett Glass <brett@lariat.net> Cc: net@freebsd.org Subject: Re: Question regarding netgraph and threading Message-ID: <4B15AFE9.5070606@elischer.org> In-Reply-To: <200912012333.QAA16055@lariat.net> References: <200912011952.MAA12927@lariat.net> <4B1580E2.4080006@elischer.org> <200912012333.QAA16055@lariat.net>
next in thread | previous in thread | raw e-mail | index | archive | help
Brett Glass wrote:
> At 01:47 PM 12/1/2009, Julian Elischer wrote:
>
>> well, not all work is done by that thread. It is the
>> backup-doer-of-things, but many netgraph operations are done in the
>> context of a caller such as teh user of a socket.
>
> In the case of a PPTP session, the data (ignoring the control session
> for the moment) flows from the interface (as GRE packets) through PPP
> (also implemented in netgraph) to an "ng" pseudo-interface, where it
> enters the ordinary FreeBSD IP stack. There isn't a user process listening
> on a socket anywhere in that path, so I assume that the netgraph
> kernel thread has to handle all of the work of encryption, decryption,
> handshaking, etc. Am I incorrect about this? I am concerned that the
> performance of a single core will be the bottleneck.
>
in the netgraph code I see:
/* Autoconfigure number of threads. */
if (numthreads <= 0)
numthreads = mp_ncpus;
and the default value of numthreads (it's a tunable) is 0
so you should have one netgraph thread per cpu.
try top -HS
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?4B15AFE9.5070606>