MAC addr takeover

Michael Rowan mtr at cutaway.com
Sat Oct 17 10:49:02 MDT 1998


Cary B. O'Brien 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?
> >
> > David Lang
> >
> > On Fri, 16 Oct 1998, Michael Rowan wrote:
> >
> > > Date: Fri, 16 Oct 1998 16:33:57 -0400
> > > From: Michael Rowan <mtr at cutaway.com>
> > > To: Flavius Bindea <flav at club-internet.fr>
> > > Cc: linux-ha at muc.de
> > > Subject: Re: MAC addr takeover
> > >
> > > Flavius Bindea wrote:
> > > >
> > > > Hi,
> > > >
> > > > I'm new on this mail list and I don't have many knowledge about
> > > > HA, but this is a very interesting feature.
> > > >
> > > > Why should MAC takeover be implemented. Is IP takeover not
> > > > enough ?
> > > >
> > >
> > > IP takeover suffers from ARP cache updates.  Client
> > > machines, sometimes numbering in the hundreds or thousands,
> > > all are communicating with the server.  This means they have
> > > the MAC address for the desination IP address in their arp
> > > cache.  When the IP takeover happens, all of these entries
> > > must be purged, either by hand, through gratuitous arps, or
> > > by timing out.  None of the easy solutions work for all
> > > platforms, so its a pain in the ass.
> > >
> > > When you fail both the IP and MAC address over, its a win
> > > since you don't care about the ARP cache entries.  They all
> > > are still valid.
> > >
> 
> (From a lurker)
> 
> Hold on folks.  I see some confusion here. The only hosts that use the
> MAC address to get to the server are the ones on the same IP network.
> I.E. things on the same physical lan (well, can broadcast to each other)
>  that have IDENTICAL network numbers.
> 
> In most configurations, clients would go through at least one
> router to get to the server.  I.E. they would be on different IP
> networks.  Remember, the I in IP stands for InterNetwork.
> 
> It is unlikely that you would have thousands of devices using the
> same network.  You could, but it would probably be too busy and
> get all clogged up.
> 
> -- cary


Yes and no.  In the switch case, its likely that you have a
whole slew of devices connected to the same lan, subdivided
into vlans or the like.  In a case of a switchless network,
its very easy to have a shop exist on one side of a router,
where the router is only used for carrying traffic to
outside subnets.  And these subnets, as often as not, are
internal IP numbers, unpublished and not connected to the
internet, so they can define whatever class subnet they want
(or have 50 subnets, all existing on the same physical
network). 

You point is valid, but whats reasonable for you or I and
whats reasonable to an IS department can be two different
things entirely.  My experience here is almost always
contrary to what I thought when I was building a HA
product.  

mtr



More information about the Linux-HA mailing list