| Author |
Message |
Phil
Guest
|
 Posted:
Thu Jan 13, 2005 6:57 am Post subject:
Explain why change from unicast to multicast prevents port f |
|
|
I'm trying to get my head around if switching from unicast to
multicast NLB will solve my switch port flooding problem
Scenario:
- Assume cluster IP address is c.c.c.c
- Dual NIC so can use multi or unicast if want inter-host
communication
- Just consider the cluster NIC for now
I have a general knowledge of Ethernet and IP but not an in depth
understanding so I am going to put my thoughts down in laymans terms
It might even help some people understand better
So, please correct me....
Unicast
-------
1 MaskSourceMac set to 1
2 All hosts have MAC address overridden to u-u-u-u-u-u on their
Cluster NIC
3 Inbound packet for c.c.c.c arrives at router
4 Router sends ARP request
5 One of the cluster host sends ARP reply saying "MAC address for
c.c.c.c is u-u-u-u-u-u"
6 Question:Which host sends this reply?
7 Router then knows to send c.c.c.c traffic to MAC u-u-u-u-u-u
8 Because MaskSourceMac is set to 1, any ethernet frames from a
cluster host to the switch have their MAC addresses masked
9 This means that the switch doesn't know which port has MAC
u-u-u-u-u-u so sends traffic to all ports
10 Hence flooding
11 Question:Is the ARP reply totally independent of the assigning a
MAC address to a port?
12 i.e. surely if an ARP reply comes back from a port saying
"send c.c.c.c traffic to MAC u-u-u-u-u-u" then that port has MAC
u-u-u-u-u-u?
Multicast
---------
1 All hosts have MAC address m-m-m-m-m-m as well as their unique
factory ones set on the cluster NIC
2 Inbound packet for c.c.c.c arrives at router
3 Router sends ARP request
4 The ARP reply says "MAC address for c.c.c.c is m-m-m-m-m-m
5 Question:Which host sends this reply?
5 Such a mapping in an ARP reply is rejected by some routers and so
administrator must add a static ARP entry in the router mapping the
Cluster IP Address to its MAC Address
6 Router now broadcasts and traffic for c.c.c.c to the multicast MAC
m-m-m-m-m-m
7 Question:Will this not also result in flooding of other non-cluster
host ports?
So, have I got it about right?
Many thanks in advance for any replies.
Phil. |
|
| Back to top |
|
 |
Glenn L
Guest
|
| Posted:
Thu Jan 13, 2005 7:38 am Post subject:
Re: Explain why change from unicast to multicast prevents po |
|
|
inline............
--
Glenn L
CCNA, MCSE 2000/2003 + Security
"Phil" <phil.marsden.google@softwire.co.uk> wrote in message
news:76fe3757.0501121657.63c1d0f4@posting.google.com...
| Quote: | I'm trying to get my head around if switching from unicast to
multicast NLB will solve my switch port flooding problem
Scenario:
- Assume cluster IP address is c.c.c.c
- Dual NIC so can use multi or unicast if want inter-host
communication
- Just consider the cluster NIC for now
I have a general knowledge of Ethernet and IP but not an in depth
understanding so I am going to put my thoughts down in laymans terms
It might even help some people understand better
So, please correct me....
Unicast
-------
1 MaskSourceMac set to 1
2 All hosts have MAC address overridden to u-u-u-u-u-u on their
Cluster NIC
[SNIP] |
not sure what you mean here....only one node at a time owns the cluster IP
address and I think the MAC address as well.
| Quote: | 3 Inbound packet for c.c.c.c arrives at router
4 Router sends ARP request
5 One of the cluster host sends ARP reply saying "MAC address for
c.c.c.c is u-u-u-u-u-u"
6 Question:Which host sends this reply?
[snip] |
whichever node currently owns the cluster IP address resource.
| Quote: | 7 Router then knows to send c.c.c.c traffic to MAC u-u-u-u-u-u
8 Because MaskSourceMac is set to 1, any ethernet frames from a
cluster host to the switch have their MAC addresses masked
[SNIP] |
I guess I don't follow you on this one. I don't know what a MaskSourceMac
is.
| Quote: | 9 This means that the switch doesn't know which port has MAC
u-u-u-u-u-u so sends traffic to all ports
10 Hence flooding
11 Question:Is the ARP reply totally independent of the assigning a
MAC address to a port?
12 i.e. surely if an ARP reply comes back from a port saying
"send c.c.c.c traffic to MAC u-u-u-u-u-u" then that port has MAC
u-u-u-u-u-u?
Multicast
---------
1 All hosts have MAC address m-m-m-m-m-m as well as their unique
factory ones set on the cluster NIC
2 Inbound packet for c.c.c.c arrives at router
3 Router sends ARP request
4 The ARP reply says "MAC address for c.c.c.c is m-m-m-m-m-m
5 Question:Which host sends this reply?
[SNIP] |
I suppose they all would. If they all share the same IP multicast address,
then they also share the same multicast MAC address.
http://www.microsoft.com/resources/documentation/Windows/2000/server/reskit/en-us/Default.asp?url=/resources/documentation/Windows/2000/server/reskit/en-us/intwork/inaf_mul_wrfn.asp
| Quote: | 5 Such a mapping in an ARP reply is rejected by some routers and so
administrator must add a static ARP entry in the router mapping the
Cluster IP Address to its MAC Address
6 Router now broadcasts and traffic for c.c.c.c to the multicast MAC
m-m-m-m-m-m
7 Question:Will this not also result in flooding of other non-cluster
host ports?
|
Multicasting is new to W2K3 clustering and only happens on the private
heartbeat NIC and only for clusters of 3 or more nodes. I am unclear on
what happens if the private network communication fails, and the public
network is cofigured for mixed mode, hence the public network takes over
control of heartbeats. The documentation does not spell this out. Does the
public network start using multicasting? I doubt it.
Easy test though.
Your private heartbeat network should not be going through a router in most
cases. It should be connected through a dumb hub or a layer 2 switch on an
isolated VLAN.
| Quote: | So, have I got it about right?
Many thanks in advance for any replies.
Phil. |
|
|
| Back to top |
|
|
Russ Kaufmann [MCT]
Guest
|
| Posted:
Thu Jan 13, 2005 11:35 pm Post subject:
Re: Explain why change from unicast to multicast prevents po |
|
|
"Phil" <phil.marsden.google@softwire.co.uk> wrote in message
news:76fe3757.0501121657.63c1d0f4@posting.google.com...
| Quote: | I'm trying to get my head around if switching from unicast to
multicast NLB will solve my switch port flooding problem
|
It might. I hope this makes sense.
In unicast, each NLB node replaces its real (hard coded) MAC address with a
new one (generated by the NLB software) and each node uses the same MAC
address. Because of this, a switch is not able to learn the port for the NLB
cluster MAC address and is forced to send the packets destined for the MAC
address to all ports to make sure it gets to the right destination.
In multicast, NLB adds a layer 2 MAC address to the NIC of each node. Each
node basically has two MAC addresses, its real one and its NLB generated
one. With multicast, you can create static entries in the switch so that it
sends the packets only to members of the NLB cluster based on their real
(hard coded) MAC addresses. Basically, you can set up a mapping of the NLB
MAC to the real MACs in the switch. If you don't create the static entries,
it will cause switch flooding just like in unicast.
| Quote: | Unicast
-------
1 MaskSourceMac set to 1
2 All hosts have MAC address overridden to u-u-u-u-u-u on their
Cluster NIC
|
This is the key point to unicast - the real MAC is overridden by the NLB
MAC.
| Quote: | 3 Inbound packet for c.c.c.c arrives at router
4 Router sends ARP request
5 One of the cluster host sends ARP reply saying "MAC address for
c.c.c.c is u-u-u-u-u-u"
6 Question:Which host sends this reply?
|
All of them. They all think they are that MAC. Thus the problem. At least
this is the way that I understand it.
| Quote: | 7 Router then knows to send c.c.c.c traffic to MAC u-u-u-u-u-u
8 Because MaskSourceMac is set to 1, any ethernet frames from a
cluster host to the switch have their MAC addresses masked
|
Not necessarily masked, but overridden. NLB hosts don't have unique MACs in
this environment.
| Quote: | 9 This means that the switch doesn't know which port has MAC
u-u-u-u-u-u so sends traffic to all ports
10 Hence flooding
11 Question:Is the ARP reply totally independent of the assigning a
MAC address to a port?
|
The switch does the mapping of the MAC to the port. ARP has nothing to do
with it other than resolving IP to MAC.
| Quote: | 12 i.e. surely if an ARP reply comes back from a port saying
"send c.c.c.c traffic to MAC u-u-u-u-u-u" then that port has MAC
u-u-u-u-u-u?
|
That is the problem in unicast. There is no single port response.
| Quote: | Multicast
---------
1 All hosts have MAC address m-m-m-m-m-m as well as their unique
factory ones set on the cluster NIC
2 Inbound packet for c.c.c.c arrives at router
3 Router sends ARP request
4 The ARP reply says "MAC address for c.c.c.c is m-m-m-m-m-m
5 Question:Which host sends this reply?
|
All of them.
| Quote: | 5 Such a mapping in an ARP reply is rejected by some routers and so
administrator must add a static ARP entry in the router mapping the
Cluster IP Address to its MAC Address
6 Router now broadcasts and traffic for c.c.c.c to the multicast MAC
m-m-m-m-m-m
7 Question:Will this not also result in flooding of other non-cluster
host ports?
|
It only floods the NLB cluster ports if the mappings are in place.
| Quote: | So, have I got it about right?
|
It sounds like you have it pretty close to the way I understand it. |
|
| Back to top |
|
|
|
|
|
|
FAQ
Memberlist
Register
Profile
Log in to check your private messages
Log in