I'm experimenting with back-to-back (crossover) links between Sun boxes and filers. Specifically, I'm trying to use Solaris 8's IP network multipathing failover and fallback with such connections.
The problem is that Solaris is unable to recognize the failure of a NIC directly connected to a filer. In testing, the feature works as expected on our normal switched network.
Solaris's IP network multipathing failover and fallback (what a mouthful!) works by initially sending multicast ICMP echo requests (to ALL-SYSTEMS.MCAST.net). The Solaris box relies on routers and hosts to respond to the multicast requests. Then it sends periodic unicast ICMP echo requests to those specific routers/hosts and knows a NIC has failed when the responses stop.
Snoop on the Solaris box reveals the filer is not responding to the multicast requests. The Solaris box, therefore, never even begins its monitoring because it thinks there is nothing to ping. Here is an excerpt:
192.168.0.5 -> ALL-SYSTEMS.MCAST.net ICMP Echo request (ID: 8194 Sequence number: 1) 192.168.0.5 -> ALL-SYSTEMS.MCAST.net ICMP Echo request (ID: 8194 Sequence number: 1) 192.168.0.5 -> ALL-SYSTEMS.MCAST.net ICMP Echo request (ID: 8194 Sequence number: 1)
The filer is definitely sitting on the other end of the cable:
ping -s 192.168.0.1 PING 192.168.0.1: 56 data bytes 64 bytes from 192.168.0.1: icmp_seq=0. time=1. ms 64 bytes from 192.168.0.1: icmp_seq=1. time=0. ms 64 bytes from 192.168.0.1: icmp_seq=2. time=0. ms 64 bytes from 192.168.0.1: icmp_seq=3. time=0. ms
Solaris box is an Ultra 1 with Solaris 8 10/00. Filer is an F760 running Data ONTAP 5.3.7. We've tested Sun ge to NetApp gigabit and also Sun hme to NetApp quad fast ethernet.
I'm interested in hearing any suggestions that might make this configuration work.
--Brian L. Brush Senior Systems Administrator Paradyne Corporation