NLB Custom port query

Hello all,

I have a Front End - Back End system as follows;

I have two FE web servers connected to two BE application servers. On the BE
servers I have NLB configured. So in effect, the NLB virtual server sits
between the FE and BE servers. All traffic is directed to the NLB virtual
server and is balanced between the back end servers.

I have an port rule set for the NLB virtual server. The rule is set to;

Listen on all IP addresses

Port Range: from 1000 to 1000 - Custom application port

Protocol: TCP

Filtering Mode: Multiple Host - Afinity: None - Load Weight: equal

I am experiencing the following problem. If an application server running on
port 1000 fails, the NLB Virtual Server still directs traffic to the failed
service. This causes the FE web servers to time out and generate errors.

Is this by design?

How can I get around this problem?

What tools are there to monitor the actual NLB sessions?

Thanks,

Martin

Hello,

this is by design. NLB is not application aware. If the node still
response on the network, it keeps part of the nlb cluster.

Workarounds:
-Previously MS included Sentinel in 2000 RK. But as far as i remember
it's check only web (it test a specific page on the node, if it's down,
it removes it from the nlb)

-SCOM: You can create your own MP that will detect an application
failure and so remove the node from NLB.

-Home made. The same as SCOM, but full home made (a script with an
infinite loop...).

An MSCS Cluster would detect the resource failure (well it depends how
big is the failure), but is a shared storage approach.


The first thing to do is having this application to:
-Logs when something goes bad
-Implement a method to get its health


Cordialement,
Mathieu CHATEAU
french blog: http://www.lotp.fr
english blog: http://lordoftheping.blogspot.com


Martin Williams a écrit :
> Hello all,
>
> I have a Front End - Back End system as follows;
>
> I have two FE web servers connected to two BE application servers. On the BE
> servers I have NLB configured. So in effect, the NLB virtual server sits
> between the FE and BE servers. All traffic is directed to the NLB virtual
> server and is balanced between the back end servers.
>
> I have an port rule set for the NLB virtual server. The rule is set to;
>
> Listen on all IP addresses
>
> Port Range: from 1000 to 1000 - Custom application port
>
> Protocol: TCP
>
> Filtering Mode: Multiple Host - Afinity: None - Load Weight: equal
>
> I am experiencing the following problem. If an application server running on
> port 1000 fails, the NLB Virtual Server still directs traffic to the failed
> service. This causes the FE web servers to time out and generate errors.
>
> Is this by design?
>
> How can I get around this problem?
>
> What tools are there to monitor the actual NLB sessions?
>
> Thanks,
>
> Martin
>
>