Cluster failover - network level

Just curious how clustering takes care of things on the network level. I
know that the failover node will take over the Cluster Group which takes
care of the Network Name and Network IP resources. But the switch has a map
of IPs to MACs. So wouldn't the switch still be trying to send to the old
port of the downed server? Or does the Cluster send out some type of
agnostic notification to the switch so it knows to rebuild or switch the new
IP to MAC association with the new port for the failover cluster node that
comes online after old active fails?

Thanks

"John" <> wrote in message
news:...
> Just curious how clustering takes care of things on the network level. I
> know that the failover node will take over the Cluster Group which takes
> care of the Network Name and Network IP resources. But the switch has a
map
> of IPs to MACs.

Switches do not store IP addresses at all, as typically you are talking
about a "Layer-2 switch" and IP is "layer-3"
Switches are mainatining a "dynamic" table of MAC's per switch port

the IP to MAC conversion is done by (R)ARP protocol which is a
"broadcasting" protocol which will get the MAC of the second node of the
cluster after failover.

So, this is taken care of by the IP and Ethernet network stacks & protocols,
not cluster.

Rgds,
Edwin.

> So wouldn't the switch still be trying to send to the old
> port of the downed server? Or does the Cluster send out some type of
> agnostic notification to the switch so it knows to rebuild or switch the
new
> IP to MAC association with the new port for the failover cluster node that
> comes online after old active fails?
>
> Thanks
>

Correct, but the clients have an ARP cache. So when they try to use Outlook
to connect, won't they include the MAC information from the ARP cache into
the packet which then the switch shouuld see that the MAC address was sent
to an old port. But then the old active isn't listening on that port
anymore. That's what I'm curious about.

"Edwin vMierlo [MVP]" <> wrote in
message news:eXz4%...
>
> "John" <> wrote in message
> news:...
>> Just curious how clustering takes care of things on the network level. I
>> know that the failover node will take over the Cluster Group which takes
>> care of the Network Name and Network IP resources. But the switch has a
> map
>> of IPs to MACs.
>
> Switches do not store IP addresses at all, as typically you are talking
> about a "Layer-2 switch" and IP is "layer-3"
> Switches are mainatining a "dynamic" table of MAC's per switch port
>
> the IP to MAC conversion is done by (R)ARP protocol which is a
> "broadcasting" protocol which will get the MAC of the second node of the
> cluster after failover.
>
> So, this is taken care of by the IP and Ethernet network stacks &
> protocols,
> not cluster.
>
> Rgds,
> Edwin.
>
>> So wouldn't the switch still be trying to send to the old
>> port of the downed server? Or does the Cluster send out some type of
>> agnostic notification to the switch so it knows to rebuild or switch the
> new
>> IP to MAC association with the new port for the failover cluster node
>> that
>> comes online after old active fails?
>>
>> Thanks
>>
>
>

That conversion will actually be pretty quick.

What happens when it fails over
1) group goes offline on Node 1 (at this point the IP address is not bound
to the MAC address anymore, and clients already get timeouts/discards, so
the ARP cache will start refreshing here)
2) the group ownership moves and starts onlining the group, including the IP
address which now binds to a new MAC on the Node 2

as all of this is still taking a couple of seconds at least, the IP is down
long enought for ARP to refresh its cycle.

Never seen this being a real problem

Rgds,
Edwin.



"John" <> wrote in message
news:...
> Correct, but the clients have an ARP cache. So when they try to use
Outlook
> to connect, won't they include the MAC information from the ARP cache into
> the packet which then the switch shouuld see that the MAC address was sent
> to an old port. But then the old active isn't listening on that port
> anymore. That's what I'm curious about.
>
> "Edwin vMierlo [MVP]" <> wrote in
> message news:eXz4%...
> >
> > "John" <> wrote in message
> > news:...
> >> Just curious how clustering takes care of things on the network level.
I
> >> know that the failover node will take over the Cluster Group which
takes
> >> care of the Network Name and Network IP resources. But the switch has
a
> > map
> >> of IPs to MACs.
> >
> > Switches do not store IP addresses at all, as typically you are talking
> > about a "Layer-2 switch" and IP is "layer-3"
> > Switches are mainatining a "dynamic" table of MAC's per switch port
> >
> > the IP to MAC conversion is done by (R)ARP protocol which is a
> > "broadcasting" protocol which will get the MAC of the second node of the
> > cluster after failover.
> >
> > So, this is taken care of by the IP and Ethernet network stacks &
> > protocols,
> > not cluster.
> >
> > Rgds,
> > Edwin.
> >
> >> So wouldn't the switch still be trying to send to the old
> >> port of the downed server? Or does the Cluster send out some type of
> >> agnostic notification to the switch so it knows to rebuild or switch
the
> > new
> >> IP to MAC association with the new port for the failover cluster node
> >> that
> >> comes online after old active fails?
> >>
> >> Thanks
> >>
> >
> >
>

Thanks Edwin. It's not a problem. Was more just curious how it works on
the network side of things.

"Edwin vMierlo [MVP]" <> wrote in
message news:...
> That conversion will actually be pretty quick.
>
> What happens when it fails over
> 1) group goes offline on Node 1 (at this point the IP address is not bound
> to the MAC address anymore, and clients already get timeouts/discards, so
> the ARP cache will start refreshing here)
> 2) the group ownership moves and starts onlining the group, including the
> IP
> address which now binds to a new MAC on the Node 2
>
> as all of this is still taking a couple of seconds at least, the IP is
> down
> long enought for ARP to refresh its cycle.
>
> Never seen this being a real problem
>
> Rgds,
> Edwin.
>
>
>
> "John" <> wrote in message
> news:...
>> Correct, but the clients have an ARP cache. So when they try to use
> Outlook
>> to connect, won't they include the MAC information from the ARP cache
>> into
>> the packet which then the switch shouuld see that the MAC address was
>> sent
>> to an old port. But then the old active isn't listening on that port
>> anymore. That's what I'm curious about.
>>
>> "Edwin vMierlo [MVP]" <> wrote in
>> message news:eXz4%...
>> >
>> > "John" <> wrote in message
>> > news:...
>> >> Just curious how clustering takes care of things on the network level.
> I
>> >> know that the failover node will take over the Cluster Group which
> takes
>> >> care of the Network Name and Network IP resources. But the switch has
> a
>> > map
>> >> of IPs to MACs.
>> >
>> > Switches do not store IP addresses at all, as typically you are talking
>> > about a "Layer-2 switch" and IP is "layer-3"
>> > Switches are mainatining a "dynamic" table of MAC's per switch port
>> >
>> > the IP to MAC conversion is done by (R)ARP protocol which is a
>> > "broadcasting" protocol which will get the MAC of the second node of
>> > the
>> > cluster after failover.
>> >
>> > So, this is taken care of by the IP and Ethernet network stacks &
>> > protocols,
>> > not cluster.
>> >
>> > Rgds,
>> > Edwin.
>> >
>> >> So wouldn't the switch still be trying to send to the old
>> >> port of the downed server? Or does the Cluster send out some type of
>> >> agnostic notification to the switch so it knows to rebuild or switch
> the
>> > new
>> >> IP to MAC association with the new port for the failover cluster node
>> >> that
>> >> comes online after old active fails?
>> >>
>> >> Thanks
>> >>
>> >
>> >
>>
>
>

John,
this behavior is controlled by relying on gratuitous ARP. Your network
devices should be able to handle it properly if you need to use them in
clustered environment. For more info, refer to
http://support.microsoft.com/kb/244331

hth
Marcin

"John" <> wrote in message
news:...
> Just curious how clustering takes care of things on the network level. I
> know that the failover node will take over the Cluster Group which takes
> care of the Network Name and Network IP resources. But the switch has a
> map of IPs to MACs. So wouldn't the switch still be trying to send to the
> old port of the downed server? Or does the Cluster send out some type of
> agnostic notification to the switch so it knows to rebuild or switch the
> new IP to MAC association with the new port for the failover cluster node
> that comes online after old active fails?
>
> Thanks

Handy link ! thanks for posting !

rgds,
Edwin.



"Marcin" <> wrote in message
news:...
> John,
> this behavior is controlled by relying on gratuitous ARP. Your network
> devices should be able to handle it properly if you need to use them in
> clustered environment. For more info, refer to
> http://support.microsoft.com/kb/244331
>
> hth
> Marcin
>
> "John" <> wrote in message
> news:...
> > Just curious how clustering takes care of things on the network level.
I
> > know that the failover node will take over the Cluster Group which takes
> > care of the Network Name and Network IP resources. But the switch has a
> > map of IPs to MACs. So wouldn't the switch still be trying to send to
the
> > old port of the downed server? Or does the Cluster send out some type
of
> > agnostic notification to the switch so it knows to rebuild or switch the
> > new IP to MAC association with the new port for the failover cluster
node
> > that comes online after old active fails?
> >
> > Thanks
>
>

Excellent! Thanks to you both...

"Marcin" <> wrote in message
news:...
> John,
> this behavior is controlled by relying on gratuitous ARP. Your network
> devices should be able to handle it properly if you need to use them in
> clustered environment. For more info, refer to
> http://support.microsoft.com/kb/244331
>
> hth
> Marcin
>
> "John" <> wrote in message
> news:...
>> Just curious how clustering takes care of things on the network level. I
>> know that the failover node will take over the Cluster Group which takes
>> care of the Network Name and Network IP resources. But the switch has a
>> map of IPs to MACs. So wouldn't the switch still be trying to send to
>> the old port of the downed server? Or does the Cluster send out some
>> type of agnostic notification to the switch so it knows to rebuild or
>> switch the new IP to MAC association with the new port for the failover
>> cluster node that comes online after old active fails?
>>
>> Thanks
>
>