Network design question. Core redundancy.

Unanswered Question
Mar 5th, 2008
User Badges:

Hi there,

I currently have a network which consists of one 2970G as a core switch and a number of 2950T and 2970G switches at the edge (see attachment 1.jpg).


What I would like to accomplish is redundancy in the core so if one switch fails traffic will automatically flow through the second core switch instead. (see attachment 2.jpg).


I don't know if this is possible to accomplish with 2 2970G, STP and redundant connections to the edge switches. Money has been set aside to invest in new core switches if needed.


Any tips, hints or recommendations is appreciated. Thank you.



Attachment: 
  • 1
  • 2
  • 3
  • 4
  • 5
Overall Rating: 4 (3 ratings)
Loading.
Joseph W. Doherty Wed, 03/05/2008 - 04:36
User Badges:
  • Super Bronze, 10000 points or more

If your switches support any of the various STPs, then yes you can implement your second design.


If your switches offer a STP faster than the original, use if for faster convergence.


When setting up STP for your design, you'll normally want one core switch as the primary root and the other as the secondary root. You'll also want to insure one of the each edge switch's uplinks blocks, not the link between the two core switches.

massimiliano.se... Wed, 03/05/2008 - 04:44
User Badges:
  • Silver, 250 points or more

Hi,

I think it's better for future use, for scalability and for performance you use a layer 3 switch at the network core.

Sure, with 2970 you may use Spanning Tree Protocol, which provides layer 2 redundancy...if you have multiple VLAN you may use also load balancing on links that are running spanning tree (for example tuning the port priorities and so on..).

If you use a layer 3 switch you can configure interVLAN routing and HSRP (redundancy at Layer 3) and you may use Spanning Tree (layer 2 redundancy) too and achieve the redundancy at Layer 3 and Layer 2...example of switch Layer 3 is Catalyst 3560.

This is just my opinion.

I hope this helps.

Best regards.

Massimiliano.



h2odata_ab Wed, 03/05/2008 - 06:35
User Badges:

Thanks to the both of you.


It sounds like the way to go is to upgrade the core switches to 3560's perhaps as the 2970 is no longer in production.


STP sounds like the way to go though in our fairly small implementation. At present we have about 15 Edge switches. Would stacking work in my proposed solution? I've never done it and I get the feeling that stacking might complicate things no?

h2odata_ab Wed, 03/05/2008 - 12:54
User Badges:

I was just told that manegement wants whatever gives us the most flexibility and uptime regardless of cost.


What would suggest without going over the top? The facility consists of multiple buildings connected via 1 gigabit fiber. The most important is the data center which looks very much like my example above (1.jpg) at present ie. a fairly small room with 1 serverack with 2 switches,a patchrack with 6 edge switches and 1 core swith. Redundancy to the outer edge of the network (other buildings) is not a requirement. All in all there's around 300 connected devices. Not a major site.

Joseph W. Doherty Wed, 03/05/2008 - 16:43
User Badges:
  • Super Bronze, 10000 points or more

What I would suggest is a stack of 3750s, containing two 3750G-12S, for termination of the gig fiber links, and two or more 3750G (10/100/1000) copper switches, either 24 or 48 port, POE or not.


This new stack becomes both your core switch and your server edge. You want to insure there's at least one extra switch of both fiber and copper such that loss of individual switch does not leave you short of ports.


The stack itself will continue to operate if an individual member stack fails.


Insure any dual connected switches, or servers, are not connected to the same physical switch.


For the dual connected edge switches, I believe you can channel two links to the stack. This doubles the available bandwidth and negates the need for STP (although should still be enabled). Again, the channel connections on the 3750s should not attach to the same individual stack member.


You can continue to remain pure L2, or activate L3 within the 3750 stack.

h2odata_ab Thu, 03/06/2008 - 02:11
User Badges:

Thank you Joseph. Some food for thought there. Etherchannels sounds like a good idea.

Actions

This Discussion