MPLS implementation gotchas

When implementing and controlling multiprotocol label switching (MPLS), you should be aware of potential pitfalls before you're locked into a service-level agreement (SLA) with a bad service provider, according to Johna Till Johnson. If you're looking for globalisation and a better QoS, read on to prepare your company for a successful MPLS migration.

President and founding partner of Nemertes Research, Johnson spoke at Interop last week on MPLS technology and implementation issues. She says that although MPLS is a technology and not a service, you need to look at the service side of MPLS to avoid any pitfalls and to buy MPLS-based services when choosing a carrier. To avoid other MPLS pitfalls, read Johnson's list of "gotchas" to get savvy about your implementation:

Gotcha #1: SLAs and carrier expectations

Don't look first at cost. "You really need to look at your service-level agreements (SLAs)," Johnson says. "Have a look at how much you're using in each class of service for each application." Know your service delivery types so that -- should something break or go down -- you know exactly whom to contact and how long your delays are likely to last.

Make sure your definitions with your carrier are the same. For example, does your "network going down" mean to you that a portion of your network goes down? Maybe your carrier's definition requires your entire network to go down. Make sure you're on the same terms and that an apple means an apple and not an Asian pear.

Put together the RFP: First, do not issue an RFP until you do a migration, and think about your change 18 months before your contract ends. If you're six months away and thinking about a migration, Johnson says, "it's not going to be pretty if you're that close to the wire."

Gotcha #2: ROI

Does MPLS save you money? Johnson says, "Yes, in certain cases, but at a cost." It depends on how much you want to implement voice, video and mobility into your organisation. If you're transitioning from data to data, you probably will not see a cost saving.

Don't implement voice and MPLS all at the same time. If you're using VoIP, you don't necessarily need to use MPLS, and visa-versa: You don't necessarily need MPLS to do VoIP. If you want to see a cost saving, you'll probably want to use both. And, if you have a background in voice, you can argue for a move away from your carrier.

When dealing with your carrier, negotiate aggressively, and instead of worrying about cost, worry about quality and your terms of agreement. Expect to go back to the negotiating table a minimum of three times. "Never expect them to give you their best offer the first, second or third time," Johnson says. If you decide to stay with your carrier, whenever you negotiate something, make sure you get all agreements in writing.

Some service providers knock out service classes altogether. But the problem with this is that you'll see more congestion and slowing before you drop connections. If this begins to happen, expect to see that the top-class system you had will start costing more.

Gotcha #3: Network architecture confusion

Architect and implement the network. Johnson says you need to know how many and what kind of network node interfaces (NNIs) you're using. Understand the nature of them, and see whether they will be valid across services and SLAs.

Every class of service is its own network. The routers are routing classes of service differently, which means there's an overlay of five virtual networks within each router. Johnson says you need to "think about the architectural impacts of multiple networks." Network people are responsible for all of the execution. But just because they make the magic happen, doesn't mean they shouldn't consult everyone on the implementation. Take the time to rope the architecture people in to the discussion. If you skip this dialogue, you're going to set yourself up for some disappointment down the road.

Gotcha #4: Carrier hardware lock-in

Carriers want to own the on-site routers to lock you into their service. "I have bad news and I have worse news," Johnson says: In 1999, if you were going to move to another carrier, you had to have a saving of 20% from your new deal/carrier for the migration to be worthwhile. The worse news is that, now, you have to save 30% to justify the move. Having to pull out your infrastructure to buy new routers will only add costs to the already painful switch.

Carriers also want to manage your network so that they can have increased lock-in. Although they say they're moving "up the stack," all that really means is that they are trying to manage your network, Johnson says. Is this something you want your carrier to do? Yes and no. On the plus side, they will be able to take the burden off your internal management, and your carrier will be able to manage cheaply -- "much more than you ever will," she says. The downside is that your carrier will own everything.

Gotcha #5: Lock-in from convergence

Voice data and carrier convergence is good for MPLS -- but bad if you don't want to be locked in. The danger is that it's extremely hard to switch carriers if something goes wrong. The pro is that VoIP is highly redundant (if you do it right), so MPLS is great for doing any-to-any architectures -- and with any-to-any architecture, you get excellent disaster recovery.

Gotcha #6: Security

The difference between ATM and frame relay privacy is that most people don't know how to hack a frame, but most people know how to hack a router. Most companies are interested in protecting their data, so depending on the type of traffic you're sending out, Johnson recommends using encryption layer protection. If you want to be using WLANs, you need encryption all the way to the desktop.

Gotcha #7: Lack of VoIP know-how

Do not cut your team -- especially people who know VoIP. Since there's no voice-specific school and no one has been in the VoIP business for 20 years, VoIP careers are especially valuable because so few people specialise in voice. Those who know data are a commodity, Johnson says, so keep your VoIP people on board.

Gotcha #8: Complexity at the edge

Complexities exist at the edges. If you're implementing MPLS, "you'll see problems at the edge," Johnson says ... "but the pain is worth it, or will be, if there's a huge rate of latency in your network."