On Thu, 20 Jan 2005 02:34:16 -0800 (PST) sai narasimhamurthy <[EMAIL PROTECTED]> wrote:
> Is there any possible/available implementation of a > Zero copy on recieve for Linux kernels? ( With no > socket copying from the NIC device buffers)...... > > Sai The problem is that with existing interfaces you need to do page and TLB flushing. On SMP systems this is more expensive than the copy. See Intel paper from OLS 2004 http://www.finux.org/Reprints/Reprint-Ronciak-OLS2004.pdf Conclusions We had several surprises along the way, but feel confident that at least with our current code base, we can conclude that using a page-flip methodology to receive network data is less ef- ficient than simply doing a copy. The major contributors to this counterintuitive result seem to be cache issues (especially obvious in the "touched data" tests), and a heavier cost asso- ciated with the work necessary to prepare and complete the page-flip. - To unsubscribe from this list: send the line "unsubscribe linux-net" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html
