[f-nsp] Sym-active LB setup in existing environment

[Mike Lott]

Hi all

I'd like to ask for some advice if possible.

I have received a secondary Foundry ServerIron CGC16-SSL with two port- 
wsm6-ssl-management-module and a single jc-16-port-gig-copper-module.  
This has been purchased so that we can add a level of redundancy into  
our existing network by adding it in a sym-active configuration with  
the existing identical SLB. However, the current topography, outlined  
on a simplistic level, is as follows:


                               |-->DistLayer02--->AccessLayer02--- 
Client02
                               |
                               ||-->DistLayer01--->AccessLayer01--- 
Client01
                               ||
                               |||-->DistLayer00--->AccessLayer00--- 
Client00
GATEWAY<--->CORE
                               |||--->ServerFarm00--->Server00
                               ||
                               ||--->ServerFarm01--->Server01
                               |
                               |--->ServerFarm02--->Server02
                                                 |
                                                 |--->FOUNDRY-SLB00

For the sake of this issue (without going into the Dist/Access  
layers), all the ServerFarms consist of two 3com switches that are  
stacked together, and have an aggregated fibre link (1Gb per fibre  
link) direct to the CORE (two, stacked, 3Com fibre switches). As you  
can see, the SLB is hanging off one of the Server Farms and not in an  
efficient position. For example, a request for a service running on  
Server00 and Server01 (the service being load balanced by the Foundry  
SLB) from Client01 would take the following path:

Client01->AccessLayer01->DistLayer01->CORE->ServerFarm02->FOUNDRY- 
SLB00->ServerFarm02->CORE->ServerFarm00->Server00

Obviously the response from the server would be the reverse of the  
above path as we are not employing DSR (and currently have no  
intention to).

What I would like to achieve is this (again, simplistic - there would  
be cross-links incorporated on each ServerFarm between the two FOUNDRY- 
SLB's and aggregated cross-links between the FOUNDRY-SLB's and the  
CORE switches):

                                   |-->DistLayer02--->AccessLayer02--- 
Client02
                                   |
                                   ||-->DistLayer01--->AccessLayer01--- 
Client01
                                   ||
                                   |||-->DistLayer00--- 
 >AccessLayer00---Client00
                                   |||
                                   |||
                                   ||||--->FOUNDRY-SLB01--->|--- 
 >ServerFarm02--->Server02
                                   ||||                           
||                  |
GATEWAY<--->CORE                          ||                  |--- 
 >ServerFarm01--->Server01
                                      |                           
||                  |
                                      |--->FOUNDRY-SLB00--->|--- 
 >ServerFarm00--->Server00


The "||" between the two FOUNDRY-SLB's is the trunk link that handles  
the syncing of configuration data.

I hope my stick diagram makes some sense and hasn't been garbled.

One of my concerns is with reduction of throughput, creating a  
bottleneck and subsequent slowdown (all servers are running with 1Gb  
cards that are aggregated across the stacked ServerFarm switches) as  
in the current setup everything is aggregated fibre direct to the  
CORE, and with the proposed setup the ServerFarms are connected to the  
CORE via the Foundry-SLB's. There are also some servers attached to  
the ServerFarms that are not behind the Foundry-SLB's and the path  
then becomes inefficient for them as I have removed the fibre links.  
Could I keep the fibre links in place and enable STP (the CORE is  
obviously designated as the Root Bridge)? I don't have a great deal of  
experience with STP, I might add.

To summarise:

1. Is this proposal in fact possible?
2. Would it be possible to keep the fibre links in this setup, or  
would they have to go due to looping?
3. What impact on throughout, in your experience, would there be in  
removing the fibre links completely?

As yet, I have not done any form of current throughput analysis, but I  
would like to hear your general ideas on this proposal, if possible.

Thanks very much

Mike