RE: Thoughts on crs installation on HP-UX
Date: Sun, 25 Jan 2009 11:32:58 -0800 (PST)
> Neither bonding nor trunking offers a perfect solution for Ethernet
> device HA - if you try to use link aggregation, they have to both be
> connected to the same switch, creating a SPOF, and if they're connected
> to both switches, then either MII link notification (unreliable with
> certain drivers) or ARP pinging (bad mojo happens if an external network
> failure occurs) has to be used to detect failures.
> Not that it won't work fine in most environments, but Linux bonding
> isn't nearly as nice as what the commercial clustering solutions
> (Veritas, SG, etc.) offer.
Are MII link notification and ARP pinging something done on the switch or the workstations (servers)? Is the latter what arping Linux command does?
# arping -I bond0 -c 2 10.111.111.33
ARPING 10.111.111.33 from 10.111.111.32 bond0 Unicast reply from 10.111.111.33 [00:1B:78:9C:A3:0A] 0.645ms
But I think your point is to have a way to detect the failure of a slave interface that makes up the bonding interface. I don't know if arping does that. But we can simply run ifconfig or `ip address show' and grep DOWN. Right? Or check words like "link status definitely down for interface eth3" in /var/log/messages? We rarely had issues with these interfaces. Linux bonding seems to be fairly stable for us. Our two slave interfaces connect to two switches each.
Yong HuangReceived on Sun Jan 25 2009 - 13:32:58 CST