Date: Wed, 21 Jun 2000 19:13:26 +1000 From: Peter Jeremy <peter.jeremy@alcatel.com.au> To: Nate Williams <nate@yogotech.com> Cc: cvs-all@FreeBSD.ORG, cvs-committers@FreeBSD.ORG, Alfred Perlstein <alfred@FreeBSD.ORG> Subject: Re: cvs commit: src/sys/kern uipc_socket.c uipc_socket2.c src/sy Message-ID: <00Jun22.065502est.115212@border.alcanet.com.au> In-Reply-To: <200006160506.XAA28743@nomad.yogotech.com>; from nate@yogotech.com on Fri, Jun 16, 2000 at 03:10:07PM %2B1000 References: <39499E16.9FE731F7@newsguy.com> <XFMail.000616132841.doconnor@gsoft.com.au> <200006160506.XAA28743@nomad.yogotech.com>
next in thread | previous in thread | raw e-mail | index | archive | help
On Fri, Jun 16, 2000 at 03:10:07PM +1000, Nate Williams wrote: >Correct. If we want performance, how about removing the difference >between kernel and userland. Why have the overhead of the kernel >vs. userland change, since it causes us to take a big performance hit. Agreed. If you want absolute maximum performance, you should build the server from the ground up, not build it on top of a general-purpose OS. And I'm not sure that hacking up the kernel is the best approch here. For example, UFS is an excellent general purpose filesystem. If you're only interested in finding the content associated with URLs I suspect you could do a lot better with something that hashed a URL directly to a (starting block, file-length) - rather than searching a directory tree. Other areas: - You don't want your webserver paging or swapping - so most of the VM subsystem can be thrown away. - Since all processes are the server, most of the uid-based security mechanisms can be thrown out. - Dropping the distinction between userspace and kernel space removes the need to support ring-transitions (which are quite expensive). Peter To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe cvs-all" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?00Jun22.065502est.115212>