Wired Intelligent Edge

Hi experts,

 

I am designing a topology for my customer I am stuck with planning the uplinks for one of the VSF fabrics. The topology is like this:

 

I don't know how to plan the uplinks for the VSF fabric at the bottom. The oversubscription ratio is 312/20=15.6, which is good for my scenario. But I don't know where to connect the uplinks, one 10G link on the second 2930F and the other one on the fifth 2930F? In that case the uplinks will be 2 hops and 3 hops away in the VSF fabric, asymmetry, is this fine or accepted? Or do I have to connect one 10G link one every two switches, although because I'de planned for seven switches, I will also have asymmetry.

I have a similar situation with the VSF fabric at the left, because it has 3 members and I'd planned 4G for the uplink (to have a reasonable oversubscription), where do I have to connect the 4 links? Two on the upper switch and two on the lower switch?

My doubt is not related for planning the bandwitdh/oversubscription, but for the access switches where to connect the uplinks.

Please your help.

 

Regards,

Julián

Another approach would be:

• Stack Acceso Piso 2: (2x1G+2x1G+2x1G)=6x1G aggregated to Stack Core Soltano (where 3x1G on 1st Aruba 3810M and the other 3x1G on the 2nd one) <-- that way any VSF Member is uplinked to both Core members.
• Stack Acceso Soltano: as is. That's good.
• Stack Acceso Piso 5: Why not connect 1st, 3rd, 5th and 7th (so a 4x10G instead of just 2x10G) to Core Stack? maybe 1st and 5th to 1st Aruba 3810M, 3rd and 7th to 2nd Aruba 3810M

That's trying to create some symmetry...clearly - other than symmetry and oversubscription ratios - there are other concerns/requirements/restrictions that you should take into account and that will let you decide to uplink differently (How the traffic flow will be? North-South? East-West? where are Servers? and Clients? are Clients equally distributed on each VSF Member? and so on...).

Hi parnassus,

 

Thanks for your support, valuable information. For Stack Acceso Piso 5, and taking into account your suggestion and connecting the 10G uplinks to 1st, 3rd, 5th and 7th members, I was planning to use 2x10G (two SFP+ modules) to form the VSF links between the members, but if so and using 2930F, I would run out of SFP+ modules for the uplinks, since every member has only 4 SFP+ modules. I see two options:

 

1. Change all the members to 2930M, but it will increase the cost too much.
2. Use 1x10G (one SFP+ module) to form the VSF links between the members, but maybe is not enough bandwidth.

What do you think about this?

 

Regards,

Julián

It all boils down to how much traffic will be there in the East-West direction (between VSF Members, speaking about Stack Acceso Piso 5 VSF)...maybe a possible scenario would be: to let 4x10G uplinks to Aruba 3810M Core...and aggregating two or three 1G ports together per each VSF Port on each VSF Member...3x1G could be not enough or could be OK...it depends on traffic to be sustained in case the VSF is really heavily used (a VSF node fails)...especially if the traffic stays local to each VSF Member. 

Hi,

 

Actually between VSF members there isn't much traffic, most of the traffic goes to the stack core. Then I think instead of aggregating two or three 1G ports together per each VSF link on each VSF member, with one 10G port per VSF link should be enough. Many thanks for your support.

 

Regards,

Julián 

Yes, I totally understand...I'm still a fan of ports aggrgation (and VSF/IRF/VSX ISL Ports aren't excluded) because I really care about enhanced resiliency compared to obtaining a pure throughput gains (better two than one...)...the optimum is to aggregate fat pipes...but also thin ones are good.

Ah ok, I didn't forget about resiliency on the VSF links. I am sure you have more hands-on on this than me, does SFP+ modules or 1G ports fails often on these switches? I guess not, but who knows...

 

Regards,

Julián