Comments below... -- regards, Andre | http://www.varon.ca
On Wed, Nov 19, 2008 at 2:45 PM, John Osena <[EMAIL PROTECTED]> wrote: > Hi Andrelst, Thad and all, > > I used mode=1 and the martian sources went away. Thanks Andrelst. > > I did some more googling and found this link > http://www.linuxfoundation.org/en/Net:Bonding. It says mode=1 is > preferred for high availability on a multiple switch topology while mode=0 > for high throughput on a single switch topology. It is also a good reading > about bonding. What the site did not mention is that mode=1 will work for multiple switch, since this method essentially uses only one NIC to communicate. Other modes has also some internal issues like arp tables mismatch, tcp congestion and multiple switches connected to different bonded NIC's. From practice, mode=4 is the best (You sidestep all the issues.) because the OS and Switches cooperate among themselves as long as they adhere to the 802.3ad protocol. RHEL and Solaris works reasonably well on this from my experience. Caution for you and others. Out of date documentations for bonding is bad here, to the point that some suggestion is just plain wrong. The technology has matured in a couple of years, and take a grain if salt on the hints. > About the martian sources on mode=0, I suspect it is due to the fact I had a > multiple switch topology and according to the link -- "This mode is the only > mode that will permit a single TCP/IP > connection to stripe traffic across multiple interfaces. It is > therefore the only mode that will allow a single TCP/IP stream to > utilize more than one interface's worth of throughput. This comes at a > cost, however: the striping often results in peer systems receiving > packets out of order, causing TCP/IP's congestion control system to kick in, > often by retransmitting segments." > > Just want to rant, a "service network restart" did not change bonding modes. > A reboot was necessary. Ahhh... the reason it did not work is because the modules are still enabled, and therefore the configuration is still there. Now, I'm assuming that you have some Lights Out/Out of Band Management where you can access the system since ssh will be down momentarily, but you can script it so that you get back your ssh session immediately. What you can do so that you don't need a reboot is: * unload the bonding module using modprobe * a simple "ifconfig up"for the bond device will do the trick. RHEL is smart enough to load the modules and enable the new config, and enable the interface. > Comments are still welcome on the topic and always open to suggestions. > > Cheers! > John > > > > ----- Original Message ---- > From: andrelst <[EMAIL PROTECTED]> > To: [EMAIL PROTECTED]; Philippine Linux Users' Group (PLUG) Technical > Discussion List <[email protected]> > Sent: Wednesday, November 19, 2008 9:57:07 AM > Subject: Re: [plug] linux bonding and then receiving martian sources in the > logs > > Thad, > > But you can be a smart rat and therefore enjoy a balanced work and life? :) > > mode=1 is the easiest to implement and troubleshoot, and is fault > tolerant or has high availability. > But FYI to the list forum here, it does not give you load balancing > (active-active) and high throughput > on the Network. > > -- > regards, > Andre | http://www.varon.ca > > On Wed, Nov 19, 2008 at 12:06 AM, thad <[EMAIL PROTECTED]> wrote: >> Lily Tomlin - "The trouble with the rat race is that even if you win, >> you're still a rat." >> >>> If your switch has LACP or 802.3ad or "etherchannel" in cisco >>> terminology support, mode=4 is the best. >> >> From my previous work we set mode=1 it works fine with rhel. Its >> attached to cisco that has etherchannel support and its the same >> switch we used for mostly etherchanneled aix lpars. _________________________________________________ Philippine Linux Users' Group (PLUG) Mailing List http://lists.linux.org.ph/mailman/listinfo/plug Searchable Archives: http://archives.free.net.ph
