A trend seen in European Networks today is to control packet latency in the network through a mix of measurement, network architecture and system architecture. Understanding the effects of latency has always been a goal in telecom. In the days of voice-only telephony networks, how the human ear/brain perceived latency from endpoint to endpoint was studied thoroughly. The main difference today is that network latency is being measured in microseconds (us) rather than milliseconds (ms).
The most pressing reason for Carriers to address this issue is the impending move of all Mobile Networks over to Ethernet and in some cases to a common infrastructure with the fixed networks. Mobile Network services have an inherent sensitivity to network latency and the expectations of those services cannot be compromised by (perceived) deficiencies in the fixed network.
Another driver is the convergence of voice/data/video over single access media, regardless of the access media type. The sensitivity to latency is increasing along with the amount of bandwidth and the number of simultaneous applications running over that access pipe. Just ask your teenager who is currently trying to get to the top of the gaming community how important a fast network connection is 
.
Recently a number of new reasons at the application layer to measure and control latency as tightly as possible have emerged. There are securities trading (discussed in this post) or other electronic transactions that value speed of transaction over anything else and can measure the impact of each millisecond of latency. There are also network and electronic security related reasons; Certain protocols view any drastic fluctuation in latency as a potential threat; Or authentication protocols that must allow or deny access as fast as possible. I believe this is the tip of the iceberg and we will be seeing a number of new applications that require a minimization of latency.
I currently have two customers who are so diligent in eliminating latency that they are displaying the latency data of their application and trading networks on the web for all to see. One customer said,
“It’s been our best sales tool. Customers actually call us for services after they’ve measured their existing service and compared it to <our service>”
These multimedia applications coupled with the growth in raw bandwidth and the migration of the Mobile Neworks over to a common access and the metro-core infrastructure are pushing the acceptable levels of latency downwards.
The focus on this trend has been in the Metro and Access portions of the network because this is usually 100%, end-to-end under the control of the Service Provider. Furthermore long-haul networks have another factor to contend with; Speed of light latency at .5 milliseconds per 100 kilometers is, as of yet an unsurmountable barrier for the scientists.
Two methods to battle latency in the Metro and Access are slowly (but surely) manifesting themselves into vendors solutions and in Service Providers Network architectures. Let’s take a look at each one in turn:
Packet Optical Networking
Most of the new services in the Metro today are Ethernet based services and many Vendors are responding to a Carrier need to have a collapsed Optical and Ethernet metro solution. This solution makes use of optical switching and Ethernet intelligence to steer packets to their destinations as efficiently as possible (discussed in this post). The Aim is to keep packets at the optical layer as long as possible therefore eliminating any latency incurred by electronically switching or routing the packets at unnecessary points in the network.
Recent studies have also shown that Packet Optical Networking is one of the places where Carriers see a significant amount of their CAPEX going to in the next 5 years. Carriers have already started to invest in this model and all signs point to a significant increase in 2010.
Eliminating the “Store and Forward”
Routers and switches are a necessary part of any network today to provide service. A function common to both routers and switches is the idea of “Store and Forward“. This can easily be explained by the following list of actions that a packet goes through when entering a router or switch:
- Packet enters on media (Ethernet, SONET, TDM, etc.)
- Packet is immediately read into memory (this is the “Storing” of the packet)
- Packet is examined to determine its destination
- Packet is read out of memory and sent to the interface closest to destination (this is the “Forward”)
- Packet is sent on the destination media interface
Each of these steps requires time to complete and even if this time can be measured in milliseconds or microseconds the latency begins to add up. This is one of the factors forcing Carriers to consider how they utilize routers and switches in the parts of their network where the impact of latency could outweigh any advantages that the router or switch could bring.
There is a type of network element recently adapted to data networks (some might call it “new” even though it’s been around for decades) called a “Network Interface Device” or NID for short (discussed in depth in this post). The NID is basically a “chip on a wire” that makes use of sophisticated silicon to forward packets to their destination without the intermediate step of storing the packet into memory.
Since there is a lack of forwarding memory the NID is limited to the role of a CPE or the first access node, however, the latency incurred by such a device is in the single microsecond (us) range. The benefits are obvious as a number of carriers who have gone down this path can attest to.
