Demystifying DWDM for the DCI
In previous articles, we discussed how using passive Dense Wave Division Multiplexing (DWDM) can help network operators achieve very high per-fiber capacity using inexpensive passive DWDM filters in 40 channel or even 80 channel variants. Using these filters and putting 10Gbps circuits on each channel can push the data rate per-fiber connection in excess of 400Gbps without the expensive (and often hard-to-get) network switches.
If it is so easy and inexpensive, why aren’t all the data centers defaulting to using this on every fiber end? Well, that’s where things get a little tricky.
Whenever you say “DWDM” to a Data Networking person (and even some service provider engineers), their default thought tends to go straight for large, complex, and expensive DWDM systems. Like Reconfigurable Optical Add Drop Multiplexing (ROADM) systems that are completely automated and perform optical switching, sub-signal aggregation, and even some L2 functions.
The truth is, DWDM is simply the combination and separation of circuits by wavelength -- and only a small part of those larger systems. It is the basic technology that allows users to put 40+ distinct circuits on a given fiber, then separate them at the far end to connect to the individual switch ports.
(As stated in the previous articles this is often done passively, requiring no electrical power, software, annual maintenance agreement, etc. -- and at a fraction of the cost of those more complex active systems.)
So again, I ask: “Why aren’t more data center interconnects using this technology?”
Well, DWDM system design -- or transport engineering -- is usually not taught during Data Networking education courses. DWDM or transport are often thought of as completing ways of architecting a network, which means there are usually two camps: You are either a Data Network Engineer, or a Transport Engineer. Either way, one typically needs the other at some point in their network.
The purpose of this article is not to say you don’t need complex, software-controlled transport devices in your network. The truth is you likely do. What we are singling out here are a few applications where you can get what you need:Fiber capacity between two places quickly, inexpensively, and without sending anyone to school to get certified.
These applications can be:
- Point to point Data Center Interconnects (DCI) on leased, or owned fiber.
- Connections between campus facilities.
- Network facilities between rooms or floors.
Using Passive DWDM can:
- Reduce or eliminate leased or new fiber builds.
- Maximize the data rate per-fiber of installed fiber plants.
- Drastically reduce Capex cost of high-capacity switches, complex DWDM systems, and reliance on service providers to maintain the connections.
- Increase capacity of DCI connections in days not months.
How can we do this in a way we can understand?
It is relatively simple, and really comes down toOptical Link Engineering.
If you take the physical map of your network and zoom in on one span where there is a capacity bottleneck, it becomes a lot easier. For simplicity’s sake, we will focus on connecting 10G switch ports, across a single span between 2km and 50km long, making the math fairly simple. For these locations we just need to focus on two primary factors: Link Budget vs Link Loss, and Dispersion.
Link Budget vs. Link Loss
Every optic or transceiver has a minimum transmit power, and a minimum receiver sensitivity. By subtracting these two values, you are left with the link budget -- or the total amount of power loss the signal can experience and still be legible by the receiver.
In a standard connection, you would calculate (or measure) the total loss of the fiber, patch panels, cassettes, and splices between the two optics. And if that is less than the link budget, then it should work . . . right?
Well, passive DWDM only adds a little more math to the Link Engineering. The optics at each end would need to be specific DWDM optics, and the filters will add more insertion loss at each end -- but it is still, pretty much the same math.
For 10G DWDM optics, the link budget is typically in the 23db range.So if a fiber span, with DWDM filters, has less than 23db of loss the link should work. Simple math, right?
Another important factor we account for is Chromatic Dispersion (CD). This is a characteristic of single-mode fiber where, as a signal travels along a fiber route, it spreads out and can arrive at the far end slightly ahead or slightly behind schedule, making it difficult to be deciphered by the receiver.
The optics we are using will also establish how much dispersion it can tolerate before the signal becomes undetectable. This value is typically measured in picoseconds per kilometer per nanometer (ps/km/nm) or even simply by the optic’s distance rating. For instance, a DWDM optic rated for 80km is often limited to 1360 ps/nm/km of dispersion. This is calculated based on traveling 80km on SMF28 type fiber with a CD rating of approximately 17 ps/nm/km.
So, there you have it.If your link falls inside the specifications defined by the optics on each end, you can deploy passive DWDM to maximize the capacity of your fiber plant, and save loads of time and money.
But what if the span exceeds the link budget or dispersion rating? No problem! The addition of Erbium Doped Fiber Amplifiers (EDFA) --to boost the signal power and/or passive Dispersion Compensation Modules (DCM) to account for excess dispersion between the DWDM filters -- can help extend the reach and ensure the optics on each end perform to expectation to years to come.
Often when Transport Engineers speak to Data Network Engineers, it can seem like they are speaking different languages. That is to be expected. Specialized jargon or terminology, approaches to problems, and education can be vastly different.
If what your network truly needs is fiber capacity, lower cost of fiber infrastructure, and flexibility of lightning-fast circuit turn up, passive and even amplified DWDM networks could be the perfect solution.