Skip site navigation (1)Skip section navigation (2)
Date:      Wed, 19 Aug 1998 09:49:43 -0700 (PDT)
From:      Marc Slemko <marcs@znep.com>
To:        Wolfram Schneider <wosch@cs.tu-berlin.de>
Cc:        hackers@FreeBSD.ORG
Subject:   Re: sendfile() API?
Message-ID:  <Pine.GSO.4.02A.9808190942480.13979-100000@redfish>
In-Reply-To: <19980819111809.A22887@caramba.cs.tu-berlin.de>

next in thread | previous in thread | raw e-mail | index | archive | help
On Wed, 19 Aug 1998, Wolfram Schneider wrote:

> On 1998-08-17 18:17:34 +0000, Terry Lambert wrote:
> > I have no doubt that it would be useful for static pages; I am just not
> > convinced that pushing the data over a call boundary is the real overhead
> > here.  For an mmap'ed file, the only overhead is the copy from the VM
> > buffer to the mbuf.  The only way such a call could eliminated this
> > overhead is by passing the VM buffers down as mbuf contents.  This can
> > be done (I did the code for it on the VMS NetWare server), but it's
> > a lot of work.
> 
> Who cares about data copy overhead? If the kernel spend 5% of the CPU
> time in copying, there are still 95% free CPU cycles.
> 
> Freefall.freebsd.org transfers 20KB/s with a CPU load less
> than 0.01. 

Lots of people do and should care about it.

freefall has such a tiny amount of web traffic that you could add an extra
dozen copies to the path without it mattering.

However, as you scale upwards by a few orders of magnitude, a bottleneck
clearly appears on most current systems where you have excessive copying.
Not all of this can be reflected in the CPU use of the process.


To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-hackers" in the body of the message



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?Pine.GSO.4.02A.9808190942480.13979-100000>