From mboxrd@z Thu Jan 1 00:00:00 1970 From: Eric Dumazet Subject: Re: kernel 2.6.25-rc7 highly unstable on high load Date: Thu, 27 Mar 2008 15:52:19 +0100 Message-ID: <47EBB4A3.4090102@cosmosbay.com> References: <20080327062502.M51594@visp.net.lb> <20080326.234049.256533912.davem@davemloft.net> <47EB46BD.2080001@cosmosbay.com> <20080327144102.M65267@visp.net.lb> Mime-Version: 1.0 Content-Type: text/plain; charset=UTF-8; format=flowed Content-Transfer-Encoding: QUOTED-PRINTABLE Cc: David Miller , netdev@vger.kernel.org, kaber@trash.net, netfilter-devel@vger.kernel.org To: Denys Fedoryshchenko Return-path: In-Reply-To: <20080327144102.M65267@visp.net.lb> Sender: netfilter-devel-owner@vger.kernel.org List-Id: netdev.vger.kernel.org Denys Fedoryshchenko a =C3=A9crit : > Yes, it seems related to routing. Before such thing was not happening= (maybe=20 > because TRIE was operating better?). > > Here is info at "peak time", i disable nmi_watchdog now, so garbage c= ollector=20 > will not be triggered by nmi watchdog.=20 > > rt_cache|rt_cache|rt_cache|rt_cache|rt_cache|rt_cache|rt_cache|rt_cac= he| > rt_cache|rt_cache|rt_cache|rt_cache|rt_cache|rt_cache|rt_cache|rt_cac= he| > rt_cache| > entries| in_hit|in_slow_|in_slow_|in_no_ro| in_brd|in_marti|in_mar= ti|=20 > out_hit|out_slow|out_slow|gc_total|gc_ignor|gc_goal_|gc_dst_o|in_hlis= t| > out_hlis| > | | tot| mc| ute| | an_dst| =20 > an_src| | _tot| _mc| | ed| miss| verflow= |=20 > _search|t_search| > 247011| 2255410| 162866| 0| 610| 5| 0| = 0| =20 > 33761| 8574| 0| 165475| 165307| 0| 0|10407428|= 174948| > 251810| 43087| 2782| 0| 9| 0| 0| = 0| =20 > 618| 122| 0| 2912| 2909| 0| 0| 241946| = 3796| > 256277| 43035| 2739| 0| 9| 0| 0| = 0| =20 > 595| 121| 0| 2867| 2864| 0| 0| 243724| = 3748| > 260177| 43596| 2647| 0| 8| 0| 0| = 0| =20 > 672| 123| 0| 2778| 2776| 0| 0| 246880| = 4048| > 232741| 42270| 2759| 0| 15| 0| 0| = 0| =20 > 665| 135| 0| 2910| 2907| 0| 0| 233990| = 3938| > 226623| 42615| 2792| 0| 11| 0| 0| = 0| =20 > 723| 132| 0| 2935| 2932| 0| 0| 218378| = 3862| > 233190| 42397| 2778| 0| 8| 0| 0| = 0| =20 > 675| 128| 0| 2913| 2909| 0| 0| 214258| = 3703| > 239093| 42342| 2713| 0| 9| 0| 0| = 0| =20 > 764| 126| 0| 2847| 2845| 0| 0| 216453| = 4080| > 150539| 36992| 7564| 0| 58| 0| 0| = 0| =20 > 539| 283| 0| 7902| 7900| 0| 0| 93459| = 1996| > 154441| 40258| 4422| 0| 12| 0| 0| = 0| =20 > 655| 212| 0| 4644| 4640| 0| 0| 77271| = 1458| > > On one "crash" i got=20 > Mar 27 11:27:15 205.177.186.74 [ 7141.156100] dst cache overflow > Mar 27 11:27:15 205.177.186.74 [ 7141.002291] dst cache overflow > Mar 27 11:27:15 205.177.186.74 [ 7141.162756] dst cache overflow > > So i increase now routes amount more than 512K to 5120K, and will see= more. > =20 You want to tune route cache for your special needs, and not permit it=20 to store 5 millions entries ! # default is a gc every 60 seconds, not good for large caches echo 1 >/proc/sys/net/ipv4/route/gc_interval # default is 8 entries per slot.. echo 4 >/proc/sys/net/ipv4/route/gc_elasticity # avoid a flush every 10 minutes echo 3600 >/proc/sys/net/ipv4/route/secret_interval -- To unsubscribe from this list: send the line "unsubscribe netfilter-dev= el" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html