Let me try and describe.
IBM has a product called Network Dispatcher
(http://www.software.ibm.com/network/dispatcher/) which forwards TCP
and UDP packets for load balancing among a group of clustered servers.
The operating system of the servers does not matter as long as the
loopback device can be either aliased or set. What this does is to
give a server with a valid IP address the ability to respond to a
request sent to another IP address. In this case the "cluster"
address. With the 2.0.x kernel all an admin had to do to alias the
loopback device was to issue the following command:
ifconfig lo:1 <cluster_addr> netmask <valid_netmask> up
where cluster_addr is the IP address of the cluster clients are coming
in on and valid_netmask is the netmask of the network that the
cluster_addr is in.
This does not work using the 2.2.x kernel. What appears to happen is
that once the server machine has its loopback device configured it
drops off the network. It cannot be reached via Ping, Telnet, etc
until the loopback device is deleted. We thought we found a
workaround by using a netmask of all zeros (0.0.0.0) or all ones
(255.255.255.255) but further investigation shows that this server
will respond (incorrectly) to an ARP. It's my understanding that
loopbacks should not respond to ARP requests. We've also tried adding
the -arp when aliasing the loopback to no avail.
We've tested this with kernels 2.2.5-15, 2.2.10, 2.2.11 and 2.2.12-20
and this same concept works on AIX, Solaris, HP-UX, OS/2, Windows NT
and OS/390.
Thanks for any information you can provide on this buglet.
Shane Owenby
shane@raleigh.ibm.com
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/