Date: Mon, 26 Jan 2009 11:28:56 -0600 From: Alan Cox <alc@cs.rice.edu> To: Jeff Roberson <jroberson@jroberson.net> Cc: Alan Cox <alc@FreeBSD.org>, src-committers@freebsd.org, svn-src-user@freebsd.org Subject: Re: svn commit: r187701 - user/alc/zerocopy/sys/vm Message-ID: <497DF2D8.6040004@cs.rice.edu> In-Reply-To: <20090125172049.R983@desktop> References: <200901260052.n0Q0qAWN031905@svn.freebsd.org> <20090125172049.R983@desktop>
next in thread | previous in thread | raw e-mail | index | archive | help
Jeff Roberson wrote: > On Mon, 26 Jan 2009, Alan Cox wrote: > >> Author: alc >> Date: Mon Jan 26 00:52:09 2009 >> New Revision: 187701 >> URL: http://svn.freebsd.org/changeset/base/187701 >> >> Log: >> Retire the page-based copy-on-write mechanism. > Hey if we're doing this there are some mbuf fields I could axe too. :-) > > This isn't productive for anyone anyway eh? I've done page-flipping > to kernel for nfs and custom protocols but to userland it just doesn't > make sense without an api to support it. It's great for netperf. :-) More seriously, it works well for repeated transmission of the same immutable buffer. Otherwise, the overheads overwhelm the savings. For example, after transmission is complete, it requires a page fault on each page of the buffer to re-enable write access to the buffer. Need I say more ... That said, the real motivation for this work is that the page-based copy-on-write mechanism is still buggy. Specifically, it doesn't play nice with the buffer cache. I just don't see the point in investing any further effort in fixing it. Alan
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?497DF2D8.6040004>