From mboxrd@z Thu Jan 1 00:00:00 1970 From: Sam Leffler Subject: Re: Asynchronous crypto layer. Date: Sat, 30 Oct 2004 23:35:28 -0700 Message-ID: <418487B0.7090203@errno.com> References: Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: 7bit Cc: johnpol@2ka.mipt.ru, jamal , netdev@oss.sgi.com, cryptoapi@lists.logix.cz, "David S. Miller" Return-path: To: James Morris In-Reply-To: Sender: netdev-bounce@oss.sgi.com Errors-to: netdev-bounce@oss.sgi.com List-Id: netdev.vger.kernel.org James Morris wrote: > On Fri, 29 Oct 2004, Sam Leffler wrote: > > >>And a paper I wrote explains how I fixed many of the problems with the >>openbsd code in freebsd. >> >>http://www.usenix.org/publications/library/proceedings/bsdcon03/tech/leffler_crypto/leffler_crpto_html/ > > > Thanks, that's a useful reference. > > For handling overload of crypto hardware, I was thinking of falling back > to software processing (really, I guess this is just load balancing > between hardware and software). I did a prototype scheduler (never committed) that took into account per-device performance metrics and outstanding operations. The s/w crypto driver was just part of the mix. As a result fallback on overload and using s/w crypto for small operands (where the setup overhead is too costly) just fell out. Sam