netdev - Re: [take36 10/10] kevent: Kevent based generic AIO.

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>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:	Mon, 12 Feb 2007 15:19:40 +0300
From:	Evgeniy Polyakov <johnpol@....mipt.ru>
To:	Andi Kleen <ak@...e.de>
Cc:	David Miller <davem@...emloft.net>,
	Ulrich Drepper <drepper@...hat.com>,
	Andrew Morton <akpm@...l.org>, netdev <netdev@...r.kernel.org>,
	Zach Brown <zach.brown@...cle.com>,
	Christoph Hellwig <hch@...radead.org>,
	Chase Venters <chase.venters@...entec.com>,
	Johann Borck <johann.borck@...sedata.com>,
	linux-kernel@...r.kernel.org, Jeff Garzik <jeff@...zik.org>,
	Jamal Hadi Salim <hadi@...erus.ca>,
	Ingo Molnar <mingo@...e.hu>, linux-fsdevel@...r.kernel.org
Subject: Re: [take36 10/10] kevent: Kevent based generic AIO.

On Mon, Feb 12, 2007 at 02:08:10PM +0100, Andi Kleen (ak@...e.de) wrote:
> Evgeniy Polyakov <johnpol@....mipt.ru> writes:
> > 
> > aio_sendfile_path() is essentially aio_sendfile(), except that it takes
> > source filename as parameter, has a pointer to private header
> > and its size (which allows to send header and file's content in one syscall
> > instead of three (open, send, sendfile) and returns opened file descriptor.
> 
> Are you sure this is a useful optimization? Do you have numbers vs open+aio_sendfile+close? 
> 
> Compared to the cost of sending a complete file three system calls should be quite in the noise. 
> And Linux system calls are not that expensive (few hundred cycles normally) 
> 
> Adding such compound system calls would be a worrying precedent because
> I'm sure others would want them then for their favourite system call combo
> too. If they were really useful it might make more sense to have a batch() 
> system call that works for arbitary calls, but I'm not convinced yet
> it's even needed. It would be certainly ugly.

Yes, that call ends up about 10MB/sec faster for 100 1mb file transfers
over 1gbit network (78 MB/s vs 66-72 MB/s over 1 Gb network, sendfile sending server 
is one-way AMD Athlong 64 3500+), but indeed, it can be the case that async IO sending 
was main speed factor.

I added header by request from Suparna Bhattacharya - my main position
is the same about syscall overhead, but I do not that care.

> -Andi

-- 
	Evgeniy Polyakov
-
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html