I am rolling out a new wireless network to one of our sites, a 490,000 sq.ft. manufacturing facility. The existing network is a hot mess that the company has no interest in fixing, with no fault tolerance at all between the core and edge switches. All edge switches connect to one core via either a single pair of OM1 fiber or via copper, and several of those edge switches have other switches or even hubs downstream from them. Don't get me started, it's not getting fixed.
The existing edge switches do not have the open ports to accommodate the ~64 access points being installed. The wireless, once operational, will be for our ERP solution and therefore business critical, so a degree of fault-tolerance is required. Not getting crazy with it, but just anything other than having four switches in a daisy-chain would be an improvement, so I want to build a disparate edge network for the wireless equipment.
What I've planned for is to run four-strand armored OM3 MM 10gb fiber to each of four new IDF cabinets, and to have the switch at each of those cabinets connect via an LACP/dynamic LAG group back to a fiber aggregation switch stack, one landing in the top ag switch and one in the bottom. Those ag switches will be subsequently cross-connected to the core. This is a pretty standard Core-Aggregation-Access model.
I'm getting some resistance from my well-meaning manager (who is probably reading this) that this is overbuilt and too expensive for the site. It is a fairly small operation. He has suggested that a more cost effective solution would be to build a ring(Core 1 --> IDF1 --> IDF 2 --> IDF3 --> IDF4 --> Core 2), which will provide comparable fault tolerance for a switch failure or a severed fiber. We would use EAPS/ERPS instead of STP, so convergence times are very minimal were the ring to be compromised.
I feel like my preferred design offers more scalability and better theoretical performance, but I'm having a hard time convincing him/the company to spend the extra money to buy the ag switches (a whopping $4500, wherever will a $3bn company find that money?).
Is he right? Is the cost not worth the difference in performance which I imagine will be completely unnoticed? Are there other architectures I'm not considering that would work better?
I appreciate your thoughts and feedback.
Here is a very quick-and-dirty drawing of each option I've talked of here:
No comments:
Post a Comment