Re: igb performance/load udp issue
On 12/21/2011 1:46 AM, Jack Vogel wrote: > I was fighting with UDP issues before the latest checkin, so you should > look at THAT version, 2.3.1 in HEAD please. Hi Jack, Is there a stand alone version of 2.3.1 that we can try on RELENG_9 and RELENG_8 ? ---Mike -- --- Mike Tancsa, tel +1 519 651 3400 Sentex Communications, m...@sentex.net Providing Internet services since 1994 www.sentex.net Cambridge, Ontario Canada http://www.tancsa.com/ ___ freebsd-net@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-net To unsubscribe, send any mail to "freebsd-net-unsubscr...@freebsd.org"
Re: igb performance/load udp issue
On Tue, 20 Dec 2011 22:46:46 -0800, Jack Vogel wrote: I was fighting with UDP issues before the latest checkin, so you should look at THAT version, 2.3.1 in HEAD please. Jack I'm actually running the HEAD version now, but my problems still persist. I did notice that the excessive load is now spread across multiple cores where before it was isolated to one core at a time. This version also doesn't suffer from the horrible performance I experienced with v2.2.5. Thanks, Clint ___ freebsd-net@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-net To unsubscribe, send any mail to "freebsd-net-unsubscr...@freebsd.org"
Re: igb performance/load udp issue
On Wed, 21 Dec 2011 16:34:56 +0100, Luigi Rizzo wrote: On Wed, Dec 21, 2011 at 08:04:54AM -0700, Clint wrote: On Tue, 20 Dec 2011 22:46:46 -0800, Jack Vogel wrote: >I was fighting with UDP issues before the latest checkin, so you >should >look at THAT version, 2.3.1 in HEAD please. > >Jack > I'm actually running the HEAD version now, but my problems still persist. I did notice that the excessive load is now spread across multiple cores where before it was isolated to one core at a time. This version also doesn't suffer from the horrible performance I experienced with v2.2.5. as someone mentioned, it may have to do with the sender using multiple queues thus causing reordering in the frames, which then the receiver has to sort (and often, the receive queue is a list so O(N) complexity per packet.) Can you run the igb driver with just a single queue ? In the bootloader, or before loading the module using "kenv", you should set hw.igb.num_queues = 1 This should prevent gratuitous reordering and perhaps fix your performance. cheers luigi I've already tried that as well, but with the v2.0.3 driver, didn't help. I'll try it again with the HEAD driver tonight. Thanks, Clint ___ freebsd-net@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-net To unsubscribe, send any mail to "freebsd-net-unsubscr...@freebsd.org"
Re: igb performance/load udp issue
On Wed, Dec 21, 2011 at 08:04:54AM -0700, Clint wrote: > On Tue, 20 Dec 2011 22:46:46 -0800, Jack Vogel wrote: > >I was fighting with UDP issues before the latest checkin, so you > >should > >look at THAT version, 2.3.1 in HEAD please. > > > >Jack > > > > I'm actually running the HEAD version now, but my problems still > persist. I did notice that the excessive load is now spread across > multiple cores where before it was isolated to one core at a time. This > version also doesn't suffer from the horrible performance I experienced > with v2.2.5. as someone mentioned, it may have to do with the sender using multiple queues thus causing reordering in the frames, which then the receiver has to sort (and often, the receive queue is a list so O(N) complexity per packet.) Can you run the igb driver with just a single queue ? In the bootloader, or before loading the module using "kenv", you should set hw.igb.num_queues = 1 This should prevent gratuitous reordering and perhaps fix your performance. cheers luigi ___ freebsd-net@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-net To unsubscribe, send any mail to "freebsd-net-unsubscr...@freebsd.org"
Re: igb performance/load udp issue
On Wed, 21 Dec 2011 16:34:56 +0100, Luigi Rizzo wrote: On Wed, Dec 21, 2011 at 08:04:54AM -0700, Clint wrote: On Tue, 20 Dec 2011 22:46:46 -0800, Jack Vogel wrote: >I was fighting with UDP issues before the latest checkin, so you >should >look at THAT version, 2.3.1 in HEAD please. > >Jack > I'm actually running the HEAD version now, but my problems still persist. I did notice that the excessive load is now spread across multiple cores where before it was isolated to one core at a time. This version also doesn't suffer from the horrible performance I experienced with v2.2.5. as someone mentioned, it may have to do with the sender using multiple queues thus causing reordering in the frames, which then the receiver has to sort (and often, the receive queue is a list so O(N) complexity per packet.) Can you run the igb driver with just a single queue ? In the bootloader, or before loading the module using "kenv", you should set hw.igb.num_queues = 1 This should prevent gratuitous reordering and perhaps fix your performance. cheers luigi I've already tried that as well, but with the v2.2.3 driver, didn't help. I'll try it again with the HEAD driver tonight. Thanks, Clint ___ freebsd-net@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-net To unsubscribe, send any mail to "freebsd-net-unsubscr...@freebsd.org"
IPv6 not responding on some aliases (recent 8-stable)
Hello, I upgraded my Nov 2010 8.x-something machine to Dec 4th and later Dec 19th userland and kernel: FreeBSD x.saper.info 8.2-STABLE FreeBSD 8.2-STABLE #0: Mon Dec 19 22:13:54 UTC 2011 r...@x.saper.info:/usr/obj/usr/src/sys/IPSEC amd64 Machine has 6 IPv6 addresses configured (out of provider-supplied /64 range). rtsol is used to get link-local default gateway, but addresses are static. What happens: After boot, SOME IPv6 addresses do not respond to anything (ICMPv6 ping, netcat...), for example: 2001:abcd:f:abcd::1000 does not work 2001:abcd:f:abcd::1001 works 2001:abcd:f:abcd::1002 works 2001:abcd:f:abcd::1003 does not work 2001:abcd:f:abcd::1004 works 2001:abcd:f:abcd::1005 does not work after a reboot it changes a bit, for example :1000 starts working There is a jail runnng on IPv4/IPv6: export jail_myjail_ip="eee.ff.gg.227,2001:abcd:f:abcd::1005" Turning the jail off does not make any difference. Turning off services listening on :1003 does not make any difference (tested with rebooting) The problem exhibited previously with 30% chance to connect to port 22 on :1000 (with ICMPv6 fully working, only port 22 affected). but now having cleaned up configuration I come to this result now: no IPv6 connectivity on some, but not all IPv6 addresses. Going out from the "not working" IPv6 addresses also fails: $ ping6 -S 2001:abcd:f:abcd::1005 www.freebsd.org PING6(56=40+8+8 bytes) 2001:abcd:f:abcd::1005 --> 2001:4f8:fff6::22 ^C --- red.freebsd.org ping6 statistics --- 3 packets transmitted, 0 packets received, 100.0% packet loss $ ping6 -S 2001:abcd:f:abcd::1000 www.freebsd.org PING6(56=40+8+8 bytes) 2001:abcd:f:abcd::1000 --> 2001:4f8:fff6::22 16 bytes from 2001:4f8:fff6::22, icmp_seq=0 hlim=54 time=163.839 ms 16 bytes from 2001:4f8:fff6::22, icmp_seq=1 hlim=54 time=163.789 ms ^C --- red.freebsd.org ping6 statistics --- 2 packets transmitted, 2 packets received, 0.0% packet loss round-trip min/avg/max/std-dev = 163.789/163.814/163.839/0.025 ms What's wrong? Must be something obvious... //Marcin FreeBSD x.saper.info 8.2-STABLE FreeBSD 8.2-STABLE #0: Mon Dec 19 22:13:54 UTC 2011 r...@x.saper.info:/usr/obj/usr/src/sys/IPSEC amd64 My configs: kernel: include GENERIC ident IPSEC options IPSEC options IPSEC_NAT_T options IPSEC_FILTERTUNNEL device crypto (IPsec is compiled in, isn't used right now) /etc/rc.conf.local: ifconfig_sis0="inet aa.bbb.ccc.103 netmask 255.255.255.0 broadcast aa.bbb.ccc.255" defaultrouter="aa.bbb.ccc.254" ifconfig_sis0_alias0="inet eee.ff.gg.227 netmask 0x" ifconfig_sis0_alias1="inet eee.ff.gg.228 netmask 0x" ifconfig_sis0_alias2="inet eee.ff.gg.229 netmask 0x" ipv6_ifconfig_sis0="2001:abcd:f:abcd::1000/64" ipv6_ifconfig_sis0_alias0="2001:abcd:f:abcd::1001/64" ipv6_ifconfig_sis0_alias1="2001:abcd:f:abcd::1002/64" ipv6_ifconfig_sis0_alias2="2001:abcd:f:abcd::1003/64" ipv6_ifconfig_sis0_alias3="2001:abcd:f:abcd::1004/64" ipv6_ifconfig_sis0_alias4="2001:abcd:f:abcd::1005/64" ipv6_default_interface="sis0" /etc/rc.conf: # This file now contains just the overrides from /etc/defaults/rc.conf. # Please make all changes to this file, not to /etc/defaults/rc.conf. sshd_enable="YES" ntpdate_enable="YES" ntpdate_hosts="213.186.33.99" fsck_y_enable="YES" named_enable="YES" ipv6_enable="YES" ipv6_ipv4mapping="YES" sendmail_enable="YES" inetd_enable="YES" kerberos5_server_enable="YES" kerberos5_server_flags="--detach --addresses='eee.ff.gg.229' --addresses='2001:41d0:1:d467::1003' --ports='88/tcp 88/udp'" milterdkim_enable="YES" tor_enable="YES" freeswitch_enable="YES" firewall_enable="YES" firewall_type="open" dummynet_enable="YES" #firewall_type="/etc/l.firewall" mysql_enable="YES" rbldnsd_enable="YES" rbldnsd_flags="-r /usr/local/etc/rbldnsd -b eee.ff.gg.229 blacklist.saper.info:ip4set:blacklist" php_fpm_enable="YES" nginx_enable="YES" ezjail_enable="YES" spawn_fcgi_enable="YES" spawn_fcgi_app="/usr/local/sbin/hgwebdir.fcgi" spawn_fcgi_bindport=9002 dovecot_enable="YES" openfire_enable="YES" openfire_javargs="-Xmx256M -Djava.net.preferIPv6Stack=true" /etc/sysctl.conf: #security.bsd.see_other_uids=0 net.inet6.ip6.accept_rtadv=1 ifconfig sis0: sis0: flags=8843 metric 0 mtu 1500 options=82008 ether 00:1c:c0:de:ad:bf inet aa.bbb.ccc.103 netmask 0xff00 broadcast aa.bbb.ccc.255 inet6 fe80::21c:c0ff:fede:adbf%sis0 prefixlen 64 scopeid 0x5 inet eee.ff.gg.227 netmask 0x broadcast eee.ff.gg.227 inet eee.ff.gg.228 netmask 0x broadcast eee.ff.gg.228 inet eee.ff.gg.229 netmask 0x broadcast eee.ff.gg.229 inet6 2001:abcd:f:abcd::1000 prefixlen 64 inet6 2001:abcd:f:abcd::1001 prefixlen 64 inet6 2001:abcd:f:abcd::1002 prefixlen 64 inet6 2001:abcd:f:abcd::1003 prefixlen 64 inet6 2001:abcd:f:abcd::1004 prefixlen 64 inet6 2001:abcd:f:abcd::1005 prefixlen 64
