From mboxrd@z Thu Jan 1 00:00:00 1970 From: Eric Dumazet Subject: Re: 2.6.21 -> 2.6.22 & 2.6.23-rc8 performance regression Date: Mon, 01 Oct 2007 10:20:07 +0200 Message-ID: <4700ADB7.7050102@cosmosbay.com> References: <200709301425.37564.nickpiggin@yahoo.com.au> <20070930223503.M8966@nuclearcat.com> <47008CB0.7010808@cosmosbay.com> <20071001.001259.28812610.davem@davemloft.net> <20071001080339.M21850@nuclearcat.com> Mime-Version: 1.0 Content-Type: text/plain; charset=UTF-8; format=flowed Content-Transfer-Encoding: QUOTED-PRINTABLE Cc: David Miller , linux-kernel@vger.kernel.org, netdev@vger.kernel.org To: Denys Return-path: In-Reply-To: <20071001080339.M21850@nuclearcat.com> Sender: linux-kernel-owner@vger.kernel.org List-Id: netdev.vger.kernel.org Denys a =C3=A9crit : > Well, i can play a bit more on "live" servers. I have now hot-swap se= rver with > full gentoo, where i can rebuild any kernel you want, with any appli= ed patch. > But it looks more like not overhead, load becoming high too "spiky", = and it is > not just permantenly higher. Also it is not normal that all system be= coming > unresposive (for example ping 127.0.0.1 becoming 300ms for period, wh= en usage > softirq jumps to 100%). > > =20 Could you try a pristine 2.6.22.9 and some patch in=20 secure_tcp_sequence_number() like : --- drivers/char/random.c.orig 2007-10-01 10:18:42.000000000 +0200 +++ drivers/char/random.c 2007-10-01 10:19:58.000000000 +0200 @@ -1554,7 +1554,7 @@ * That's funny, Linux has one built in! Use it! * (Networks are faster now - should this be increased?) */ - seq +=3D ktime_get_real().tv64; + seq +=3D ktime_get_real().tv64 / 1000; #if 0 printk("init_seq(%lx, %lx, %d, %d) =3D %d\n", saddr, daddr, sport, dport, seq); Thank you > On Mon, 01 Oct 2007 00:12:59 -0700 (PDT), David Miller wrote > =20 >> From: Eric Dumazet >> Date: Mon, 01 Oct 2007 07:59:12 +0200 >> >> =20 >>> No problem here on bigger servers, so I CC David Miller and netdev >>> on this one. AFAIK do_gettimeofday() and ktime_get_real() should >>> use the same underlying hardware functions on PC and no performance >>> problem should happen here. >>> =20 >> One thing that jumps out at me is that on 32-bit (and to a certain >> extent on 64-bit) there is a lot of stack accesses and missed >> optimizations because all of the work occurs, and gets expanded, >> inside of ktime_get_real(). >> >> The timespec_to_ktime() inside of there constructs the ktime_t retur= n >> value on the stack, then returns that as an aggregate to the caller. >> >> That cannot be without some cost. >> >> ktime_get_real() is definitely a candidate for inlining especially i= n >> these kinds of cases where we'll happily get computations in local >> registers instead of all of this on-stack nonsense. And in several >> cases (if the caller only needs the tv_sec value, for example) >> computations can be elided entirely. >> >> It would be constructive to experiment and see if this is in fact=20 >> part of the problem. >> =20 > > > -- > Denys Fedoryshchenko > Technical Manager > Virtual ISP S.A.L. > > > =20