Re: [Lxc-users] Dreadful network performance, only to host from container
Performance issues with networking and veth devices can be often linked to the implementation of hardware acceleration in various kernel drivers for various NICs. I found this from a lot of toggling of the acceleration tunables with ethtool. I'm sure there's a deeper issue with the drivers, but it was just a trial and error discovery on my part. -- Matt Bailey 303.871.4923 Senior Software Specialist University of Denver, UTS http://du.edu http://system42.net On Fri, May 28, 2010 at 3:12 AM, Daniel Lezcano daniel.lezc...@free.fr wrote: On 05/28/2010 02:58 AM, Toby Corkindale wrote: On 28/05/10 05:55, Matt Bailey wrote: /usr/sbin/ethtool -K br0 sg off /usr/sbin/ethtool -K br0 tso off Might fix your problem, YMMV; this worked for me. Bam! Problem fixed. All I needed was the 'sg' option - tso wasn't enabled anyway. Now getting a healthy 15-16 mbyte/sec. Great ! Thanks for that.. Is this a bug in a driver somewhere that I should, or just something one always needs to be aware of with LXC? (and thus should go in a FAQ) The true is the first time I see this problem solved by this trick. I suppose that has something related to the capabilities of your nic and the bridge inherit them. Dunno ... Matt, how did you find that ? Is it a problem spotted with the other virtualization solution (xen, vmware, qemu, openvz, ...) ? Do you have some pointer describing the problem/solution ? So we can add a FAQ with a good description / diagnostic of the problem. Thanks -- Daniel -- ___ Lxc-users mailing list Lxc-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/lxc-users
Re: [Lxc-users] Dreadful network performance, only to host from container
On 05/28/2010 02:58 AM, Toby Corkindale wrote: On 28/05/10 05:55, Matt Bailey wrote: /usr/sbin/ethtool -K br0 sg off /usr/sbin/ethtool -K br0 tso off Might fix your problem, YMMV; this worked for me. Bam! Problem fixed. All I needed was the 'sg' option - tso wasn't enabled anyway. Now getting a healthy 15-16 mbyte/sec. Great ! Thanks for that.. Is this a bug in a driver somewhere that I should, or just something one always needs to be aware of with LXC? (and thus should go in a FAQ) The true is the first time I see this problem solved by this trick. I suppose that has something related to the capabilities of your nic and the bridge inherit them. Dunno ... Matt, how did you find that ? Is it a problem spotted with the other virtualization solution (xen, vmware, qemu, openvz, ...) ? Do you have some pointer describing the problem/solution ? So we can add a FAQ with a good description / diagnostic of the problem. Thanks -- Daniel -- ___ Lxc-users mailing list Lxc-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/lxc-users
Re: [Lxc-users] Dreadful network performance, only to host from container
Hi, Send ifconfig br0from the host ifconfig eth0 from the container and the version of lxc you're using. Do you have anything special with the /etc/sysctl.conf? On a completely blank container with no tuning, I get with scp; host-container squashfs.img 100% 639MB 33.6MB/s 00:19 container-host squashfs.img 100% 639MB 29.0MB/s 00:22 Both tests inside the container. The limiting resource here is cpu for the encryption. I'm on kernel 2.6.34/fc12 for this. Andy On Thu, 2010-05-27 at 17:51 +1000, Toby Corkindale wrote: Hi, I have several LXC containers, running on Ubuntu Lucid's current kernel, 2.6.32-22. Network performance from the containers to the LAN is fine. However copying files (scp/rsync/whatever) between the containers and their host is extremely slow - a steady 32 kb/sec. I am using bridged networking, with an mtu of 1500 on the bridge and all the related devices (virtual and real). I have tried setting 'setfd' to 0 on br0. Config includes: lxc.network.type = veth lxc.network.flags = up lxc.network.link = br0 lxc.network.ipv4 = 0.0.0.0 lxc.network.name = eth0 I couldn't spot anything quite like this on the mailing list archives, although I did see in february this: http://www.mail-archive.com/lxc-users@lists.sourceforge.net/msg00071.html but it didn't seem to be resolved. Any thoughts? Thanks, Toby -- ___ Lxc-users mailing list Lxc-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/lxc-users Andrew Phillips Head of Systems www.lmax.com Office: +44 203 1922509 Mobile: +44 (0)7595 242 900 LMAX | Level 2, Yellow Building | 1 Nicholas Road | London | W11 4AN The information in this e-mail and any attachment is confidential and is intended only for the named recipient(s). The e-mail may not be disclosed or used by any person other than the addressee, nor may it be copied in any way. If you are not a named recipient please notify the sender immediately and delete any copies of this message. Any unauthorized copying, disclosure or distribution of the material in this e-mail is strictly forbidden. Any view or opinions presented are solely those of the author and do not necessarily represent those of the company.-- ___ Lxc-users mailing list Lxc-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/lxc-users
Re: [Lxc-users] Dreadful network performance, only to host from container
On 27/05/10 18:06, atp wrote: As requested: ifconfig br0 from the host br0 Link encap:Ethernet HWaddr 00:1e:37:4d:8c:d8 inet addr:192.168.1.206 Bcast:192.168.1.255 Mask:255.255.255.0 inet6 addr: fe80::21e:37ff:fe4d:8cd8/64 Scope:Link UP BROADCAST RUNNING MULTICAST MTU:1500 Metric:1 RX packets:3867723 errors:0 dropped:0 overruns:0 frame:0 TX packets:1849343 errors:0 dropped:0 overruns:0 carrier:0 collisions:0 txqueuelen:0 RX bytes:3451303555 (3.4 GB) TX bytes:382610461 (382.6 MB) ifconfig eth0 from the container eth0 Link encap:Ethernet HWaddr 36:d1:4f:d9:51:59 inet addr:192.168.1.88 Bcast:192.168.1.255 Mask:255.255.255.0 inet6 addr: fe80::34d1:4fff:fed9:5159/64 Scope:Link UP BROADCAST RUNNING MULTICAST MTU:1500 Metric:1 RX packets:1416 errors:0 dropped:0 overruns:0 frame:0 TX packets:495 errors:0 dropped:0 overruns:0 carrier:0 collisions:0 txqueuelen:1000 RX bytes:1020033 (1.0 MB) TX bytes:37512 (37.5 KB) and the version of lxc you're using. It's close to the git head, master branch. Last commit was 0093bb8ced5784468daf8e66783e6be3782e8fea on May 18th. (The version that originally shipped with ubuntu was giving me errors about not being able to pivot_root) Do you have anything special with the /etc/sysctl.conf? I think these came with the system, are they likely to be problematic? net.ipv4.conf.default.rp_filter=1 net.ipv4.conf.all.rp_filter=1 net.ipv4.tcp_syncookies=1 vm.mmap_min_addr = 65536 fs.inotify.max_user_watches = 524288 kernel.shmmax = 38821888 On a completely blank container with no tuning, I get with scp; host-container squashfs.img 100% 639MB 33.6MB/s 00:19 container-host squashfs.img 100% 639MB 29.0MB/s 00:22 Both tests inside the container. The limiting resource here is cpu for the encryption. mm, yeah, I'd be waiting all week to copy an equivalently sized file like that. Although if i copy it to another host on the network, then back again, it's all fine :/ I'm on kernel 2.6.34/fc12 for this. I'm on 2.6.32-22/ubuntu 10.04 thanks, Toby -- ___ Lxc-users mailing list Lxc-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/lxc-users
Re: [Lxc-users] Dreadful network performance, only to host from container
On 05/27/2010 10:21 AM, Toby Corkindale wrote: On 27/05/10 18:06, atp wrote: As requested: ifconfig br0 from the host br0 Link encap:Ethernet HWaddr 00:1e:37:4d:8c:d8 inet addr:192.168.1.206 Bcast:192.168.1.255 Mask:255.255.255.0 inet6 addr: fe80::21e:37ff:fe4d:8cd8/64 Scope:Link UP BROADCAST RUNNING MULTICAST MTU:1500 Metric:1 RX packets:3867723 errors:0 dropped:0 overruns:0 frame:0 TX packets:1849343 errors:0 dropped:0 overruns:0 carrier:0 collisions:0 txqueuelen:0 RX bytes:3451303555 (3.4 GB) TX bytes:382610461 (382.6 MB) Can you give the routes of the host please ? ifconfig eth0 from the container eth0 Link encap:Ethernet HWaddr 36:d1:4f:d9:51:59 inet addr:192.168.1.88 Bcast:192.168.1.255 Mask:255.255.255.0 inet6 addr: fe80::34d1:4fff:fed9:5159/64 Scope:Link UP BROADCAST RUNNING MULTICAST MTU:1500 Metric:1 RX packets:1416 errors:0 dropped:0 overruns:0 frame:0 TX packets:495 errors:0 dropped:0 overruns:0 carrier:0 collisions:0 txqueuelen:1000 RX bytes:1020033 (1.0 MB) TX bytes:37512 (37.5 KB) and the version of lxc you're using. It's close to the git head, master branch. Last commit was 0093bb8ced5784468daf8e66783e6be3782e8fea on May 18th. (The version that originally shipped with ubuntu was giving me errors about not being able to pivot_root) Do you have anything special with the /etc/sysctl.conf? I think these came with the system, are they likely to be problematic? net.ipv4.conf.default.rp_filter=1 net.ipv4.conf.all.rp_filter=1 net.ipv4.tcp_syncookies=1 vm.mmap_min_addr = 65536 fs.inotify.max_user_watches = 524288 kernel.shmmax = 38821888 On a completely blank container with no tuning, I get with scp; host-container squashfs.img 100% 639MB 33.6MB/s 00:19 container-host squashfs.img 100% 639MB 29.0MB/s 00:22 Both tests inside the container. The limiting resource here is cpu for the encryption. mm, yeah, I'd be waiting all week to copy an equivalently sized file like that. Although if i copy it to another host on the network, then back again, it's all fine :/ I'm on kernel 2.6.34/fc12 for this. I'm on 2.6.32-22/ubuntu 10.04 thanks, Toby -- ___ Lxc-users mailing list Lxc-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/lxc-users -- ___ Lxc-users mailing list Lxc-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/lxc-users
Re: [Lxc-users] Dreadful network performance, only to host from container
Toby Just FYI in case you were unaware - it seems one of your MXs is black holed. I tried to email you direct, but messagelabs said; toby.corkind...@strategicdata.com.au: 74.125.148.10 does not like recipient. Remote host said: 554 5.7.1 Service unavailable; Client host [74.125.149.113] blocked using sbl-xbl.spamhaus.org; http://www.spamhaus.org/query/bl?ip=74.125.149.113 reply below; --- A quick look, all looks reasonably normal, txqueuelen for br0 is 0, and probably should be more than 1000. The number of packets on eth0 seems very small - 1416 and 495. rx/tx I take it you've checked for things like arp poisoning - two of the containers having the same mac address? What does the flow of packets with tcpdump look like from the container? Are you seeing resets, packet duplicates, retransmissions? Andy The information in this e-mail and any attachment is confidential and is intended only for the named recipient(s). The e-mail may not be disclosed or used by any person other than the addressee, nor may it be copied in any way. If you are not a named recipient please notify the sender immediately and delete any copies of this message. Any unauthorized copying, disclosure or distribution of the material in this e-mail is strictly forbidden. Any view or opinions presented are solely those of the author and do not necessarily represent those of the company. -- ___ Lxc-users mailing list Lxc-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/lxc-users
Re: [Lxc-users] Dreadful network performance, only to host from container
/usr/sbin/ethtool -K br0 sg off /usr/sbin/ethtool -K br0 tso off Might fix your problem, YMMV; this worked for me. -- Matt Bailey 303.871.4923 Senior Software Specialist University of Denver, UTS http://du.edu http://system42.net On Thu, May 27, 2010 at 4:48 AM, atp andrew.phill...@lmax.com wrote: Toby Just FYI in case you were unaware - it seems one of your MXs is black holed. I tried to email you direct, but messagelabs said; toby.corkind...@strategicdata.com.au: 74.125.148.10 does not like recipient. Remote host said: 554 5.7.1 Service unavailable; Client host [74.125.149.113] blocked using sbl-xbl.spamhaus.org; http://www.spamhaus.org/query/bl?ip=74.125.149.113 reply below; --- A quick look, all looks reasonably normal, txqueuelen for br0 is 0, and probably should be more than 1000. The number of packets on eth0 seems very small - 1416 and 495. rx/tx I take it you've checked for things like arp poisoning - two of the containers having the same mac address? What does the flow of packets with tcpdump look like from the container? Are you seeing resets, packet duplicates, retransmissions? Andy The information in this e-mail and any attachment is confidential and is intended only for the named recipient(s). The e-mail may not be disclosed or used by any person other than the addressee, nor may it be copied in any way. If you are not a named recipient please notify the sender immediately and delete any copies of this message. Any unauthorized copying, disclosure or distribution of the material in this e-mail is strictly forbidden. Any view or opinions presented are solely those of the author and do not necessarily represent those of the company. -- ___ Lxc-users mailing list Lxc-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/lxc-users -- ___ Lxc-users mailing list Lxc-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/lxc-users
Re: [Lxc-users] Dreadful network performance, only to host from container
On 27/05/10 19:52, Daniel Lezcano wrote: On 05/27/2010 10:21 AM, Toby Corkindale wrote: On 27/05/10 18:06, atp wrote: As requested: ifconfig br0 from the host br0 Link encap:Ethernet HWaddr 00:1e:37:4d:8c:d8 inet addr:192.168.1.206 Bcast:192.168.1.255 Mask:255.255.255.0 inet6 addr: fe80::21e:37ff:fe4d:8cd8/64 Scope:Link UP BROADCAST RUNNING MULTICAST MTU:1500 Metric:1 RX packets:3867723 errors:0 dropped:0 overruns:0 frame:0 TX packets:1849343 errors:0 dropped:0 overruns:0 carrier:0 collisions:0 txqueuelen:0 RX bytes:3451303555 (3.4 GB) TX bytes:382610461 (382.6 MB) Can you give the routes of the host please ? $ route -n Kernel IP routing table Destination Gateway Genmask Flags Metric RefUse Iface 192.168.1.0 0.0.0.0 255.255.255.0 U 0 00 br0 192.168.122.0 0.0.0.0 255.255.255.0 U 0 00 virbr0 169.254.0.0 0.0.0.0 255.255.0.0 U 1000 00 br0 0.0.0.0 192.168.1.1 0.0.0.0 UG10000 br0 I have some kvm (via libvirt) virtual machines running on this host as well, hence the virbr0 existing as well. -- ___ Lxc-users mailing list Lxc-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/lxc-users
Re: [Lxc-users] Dreadful network performance, only to host from container
On 28/05/10 05:55, Matt Bailey wrote: /usr/sbin/ethtool -K br0 sg off /usr/sbin/ethtool -K br0 tso off Might fix your problem, YMMV; this worked for me. Bam! Problem fixed. All I needed was the 'sg' option - tso wasn't enabled anyway. Now getting a healthy 15-16 mbyte/sec. Thanks for that.. Is this a bug in a driver somewhere that I should, or just something one always needs to be aware of with LXC? (and thus should go in a FAQ) thanks! Toby -- ___ Lxc-users mailing list Lxc-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/lxc-users