From mboxrd@z Thu Jan 1 00:00:00 1970 From: Jesper Dangaard Brouer Subject: Re: [PATCH v3 net-next] tc: remove unused redirect ttl Date: Sun, 3 May 2015 17:28:42 +0200 Message-ID: <20150503172842.42379466@redhat.com> References: <1430543983-10446-1-git-send-email-ast@plumgrid.com> Mime-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 7bit Cc: brouer@redhat.com, "David S. Miller" , Jamal Hadi Salim , Daniel Borkmann , Florian Westphal , netdev@vger.kernel.org To: Alexei Starovoitov Return-path: Received: from mx1.redhat.com ([209.132.183.28]:49959 "EHLO mx1.redhat.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1750887AbbECP2t (ORCPT ); Sun, 3 May 2015 11:28:49 -0400 In-Reply-To: <1430543983-10446-1-git-send-email-ast@plumgrid.com> Sender: netdev-owner@vger.kernel.org List-ID: On Fri, 1 May 2015 22:19:43 -0700 Alexei Starovoitov wrote: > From: Jamal Hadi Salim > > improves ingress+u32 performance from 22.4 Mpps to 22.9 Mpps I like this change and I'm ACKing it. But I would like to comment on these performance numbers. I assume this is a single CPU test with your pktgen RX/STACK_INJECT mode that drops the packet "early (in ip_rcv() I think). It sound impressive to get +0.5 Mpps improvement from such a small change, while in reality this is approx a 1ns improvement. (1/(22.4*10^6)*10^9) = 44.64ns (1/(22.9*10^6)*10^9) = 43.67ns improvement diff = 0.97ns I'm a fan of zooming in on parts of the stack, as it allow us to performance optimize and measure parts of the stack that is normally hard to measure. And your ingress use-case is valid. People should just be aware, that we can easily "over-sell" when "zooming" in... > Signed-off-by: Jamal Hadi Salim > Signed-off-by: Alexei Starovoitov > Acked-by: Florian Westphal > Acked-by: Daniel Borkmann Acked-by: Jesper Dangaard Brouer -- Best regards, Jesper Dangaard Brouer MSc.CS, Sr. Network Kernel Developer at Red Hat Author of http://www.iptv-analyzer.org LinkedIn: http://www.linkedin.com/in/brouer