Greetings,
I thought I'd close this one out. I had our network guy change all the 1G trunk connections to LACP, along with the 10G stand-alone connection. The 4x1G connections are still doing the broadcast, apparently because it is in a single mode ifgrp. I was hoping this would the broadcasts but it doesn't :-)
Thanks to everyone for their help and suggestions.
Jeff
On Sun, Aug 5, 2012 at 5:44 PM, Stuart Kendrick skendric@fhcrc.org wrote:
For what is is worth, here's a diagram, complete with /etc/rc syntax and Cisco IOS syntax, for doing precisely this (LACP groups within a second-level interface group). http://www.skendric.com/philosophy/uptime/Toast-Ethernet-IP.pdf
BTW: I find that yanking cables is an incomplete way to test host / Ethernet / IP configurations ... it doesn't handle a range of failure modes. For example, when an Ethernet switch reboots, link stays up (or toggles only briefly), but packet forwarding is disabled for many seconds, or even minutes, while the switch performs its hardware checks (on every single port ... takes a long time, if the switch contains a lot of ports) and loads its OS -- schemes which rely on link-only pass the 'yank the cable' test but do not pass the 'reboot the Ethernet switch' test. http://www.skendric.com/philosophy/uptime/Testing-the-Transport-Side-of-High... [The ONTAP section here is dated ... it claims that the LACP approach is the /only/ NIC HA method which ONTAP supports ... that might have been true when I wrote it, with the 7.2.x versions of ONTAP I was using ... but per this discussion, that's inaccurate today: the default ARP polling mechanism is also a possibility.]
hth,
--sk
Stuart Kendrick FHCRC
On 8/1/2012 2:47 AM, Steiner, Jeffrey wrote:
You can use LACP. This is a commonly misread part of the documentation, which states you can't use LACP groups "AS" second-level interface groups, but you can use them "IN" a second-level interface group. The intent of the documentation was to say that you can't take a pair of LACP ifgrps and combine them into another LACP ifgrp and then have some sort of giant LACP trunk where some bits are active and some aren't. There's no problem with making something like a 4-port LACP to one switch, another 4-port LACP trunk to another switch, and then combining them together as an active-passive second-level ifgrp.
-----Original Message----- From: toasters-bounces@teaparty.net [mailto:toasters-bounces@teaparty.net] On Behalf Of Jeff Cleverley Sent: Tuesday, July 31, 2012 6:32 PM To: toasters@teaparty.net Subject: Arp broadcasts from single vifs
Greetings,
I thought I posted about this a while back but don't seem to be able to find anything in my email or web searches. My apologies if this is a duplicate.
We have some 7.3.5.1P4 filers running static multimode vifs, combined into a second level single vif for failover. One vif has 4 or 6 1gig links, and the other is a 10gig network port. We use the prefer to run the 10gig as the primary. All this seems to work fine and pulling cables, etc fails over properly.
My question is whether there is another way to set up a failover configuration where the inactive links are not broadcasting the "who has xxx" arp requests every 5-6 seconds? With 20 standby links, this generates ~345K broadcast per day. This is causing arpwatch to create a very large file and be moderately useless due to the noise. We do have switches that could do LACP/dynamic multimode, but the network guide says these cannot be put in a second level vif.
Is there anything we can do to get failover vifs and not have all the chatter?
Thanks,
Jeff
Toasters mailing list Toasters@teaparty.net http://www.teaparty.net/mailman/listinfo/toasters