MAC addr takeover
mtr at cutaway.com
Fri Oct 16 19:11:12 MDT 1998
Oh, there are absolutely issues with both methods, which is
why you have both methods. There are lots of companies
(believe it or not) that don't use switches at all. And,
again, things vary from switch to switch as to the behavoir,
as with client boxes. The difference is there is usually
one type of switch to deal with, not 10 or 20.
Many HA implementations have a client protocol for
propagating events, and using these client daemons to update
ARP tables works well in a lot of situations. I just did a
sap deal where the CI was protected with hacmp for instance,
and the only ARP issues where with the 4 or 5 application
servers, which weren't protected because there were a few of
them. The we never worry about the clients, so MAC takeover
is a pain in the ass and buys you little or nothing.
God bless vendor centric customers ;-)
David Lang wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> the problem is not a changing mac address on a switch port, but the mac
> address changing from one switch port to another without (in some cases
> like a OS crash) a disconnect from the switch. two examples
> 3com switch 3000 15 min timeout on a mac address, immediate timeout if the
> port is disconnected (link lost on that port)
> Cisco Catalyst 5000 15 min timeout on a mac address, no reduction if the
> port is disconnected.
> everything will clear up in a few min (these values are settable, but if
> set to low your switch wasts bandwidth and cpu figuring things out
> repeatedly) but during this time there is no way to tell the switch that
> the mac address is no longer on one port and is instead on the other port.
> Daid Lang
> On Fri, 16 Oct 1998, Yiming Zhang wrote:
> > Date: Fri, 16 Oct 1998 15:44:48 -0700
> > From: Yiming Zhang <yzhang at integrix.com>
> > To: linux-ha at muc.de
> > Subject: Re: MAC addr takeover
> > At 06:23 PM 10/16/98 -0400, you wrote:
> > >On Fri, 16 Oct 1998, David Lang wrote:
> > >> but on the other hand you have to update the MAC tables in your ethernet
> > >> switch if you are using one. is this really easier?
> > >
> > >It's just a single "task", and you just have to concentrate on making it
> > >work with that one type of switch.
> > >
> > >That beats the heck out of having to deal with a plethora of M$-bred myopic,
> > >non-compliant, and just plain stupid/broken networked devices.
> > >
> > >If your switch is broken (wrt physical address mapping), at least you
> > >probably have control over it and can replace it with something that works...
> > Agree,
> > HA cannot and should not guarantee broken switch also works.
> > Let's forget about HA, in a non-HA setting, if a workstation on LAN changed
> > a NIC, and reboot (to send gratuitous ARP), the switch should timely
> > capture this mapping (instead of until cache timeout).
> > If this basic need cannot met, the switch should be
> > replaced even when we have no HA.
> > (But we should mention the broke switch issue in installtion manual).
> > Zhang
> -----BEGIN PGP SIGNATURE-----
> Version: PGP for Personal Privacy 5.0
> Charset: noconv
> -----END PGP SIGNATURE-----
More information about the Linux-HA