Hi Serge,
The simple answer to this is there is no simple answer, especially since we
do not know the affinity rules being used. If this is an NLB cluster that
is balancing Terminal Services then you should check that your
configruation matches the following:
243523 Using Terminal Server with Windows Load Balancing Service
http://support.microsoft.com/?id=243523
In addition to this article I would suggest the following best practice.
The NLB hosts should have 2 network cards, one network card for NLB and one
network card for other kinds of traffic. The IP addresses on the NLB card
should be in the same subnet and the IP address for the other card, called
Admin, should be in an entirely different subnet. This can either be setup
in a configuration called private, where the admin adapters all communicate
with a hub to each other and noting else. Or what is called an internet
configuration where the NLB cluster is in a sort of intermediary. The NLB
cluster in this configuration has the incoming connections on the NLB
interface and then a connection to a LAN where it may have to access some
resources. Since this sounds like a direct Terminal Service setup that does
not have to access data elsewhere I would suggest using the private
configuration described above, along with the settings from the KB article
provided above. It should also be noted that unicast should be used. The
only reason to ever use multicast in NLB is in single network card
configurations or if an application that is being balanced requires NLB.
The reason for this is many switches are not compatible with multicast NLB
traffic and this can cause connectivity issues. Microsoft also does not
recommend using single NIC NLB configurations.
It also sounds like the port rules may not be set correctly per the article
above you want your port rules to look something like this:
Port Range: from 3389 to 3389.
Protocols: TCP Multiple hosts.
Affinity: Depends on requirements.
Load weight: Equal (if you want equal load on each Terminal Server).
From what you have described you probably want the affinity set to none,
This will cause clients to connect to either server without regard to the
server they connected to previously. If there is data on a particular
server this will not be optimal.
The reason you are going to the same server time after time is that the
client has made a connection with that server. By default NLB is set to
single affinity, which is continue going to a node I connected to
previously, the NLB algorithm uses the client IP and source port to
remember the affinity for a client. Hence why you go back to the same
server again and again.
OK, on to the second part of your question. Yes the amount of traffic you
see is normal. By default NLB masks the source MAC address of the NLB
system, as such this causes greater network traffic as the MAC is not set
into the cache of any system or switch on the network.
--
Curtis Koenig
Support Professional
Microsoft Clustering Technologies Support
Microsoft Certified Systems Engineer
Microsoft Certified Systems Engineer - Security
This posting is provided "AS IS" with no warranties and confers no rights.
Please reply to the newsgroup so that others may benefit. Thanks!
--------------------
