Hi David Hughes,
My scenario is as follows: There are a set of application servers in the HQ and WAN connections between remote Branch Offices and the HQ. The WAN connections speed varies from 256 Kbps to 6 Mbps, and transmission medium can be by satellite, microwave, fiber optics and copper wires with a digital modem.
My question is this: Based on the brief scenario description, what things should I worry about when assessing the WAN connectivity network? Which points are important for identifying latency and low response times for centralized application access? What can we do to provide better network services? Is it relevant to know the Layer 1 Transmission technology that will be in place? What would be the impact, if any, in interactive real time application access? If voice is the application, are there any concerns?
There are lots of questions here, so let me attempt to provide some high level guidance. For one, I would not make the service provider selection based on the Layer 1 technology. More importantly, I would see what kind of Service Level Agreement (SLA) is offered by your service provider. The one that best meets your demands will largely dictate the underlying technology. Things to consider on this front include:
- Typical/peak round trip time, jitter, loss
- Service up-time
- Availability in all of your required locations
For reliable quality voice over IP you will want low loss and jitter, and preferably support for prioritization within the network using Quality of Service (QoS). Service up-time requirements will depend on how business critical network connectivity is for your remote locations.
Lastly, lots of factors can affect latency and loss. One of the biggest contributors to latency is the distance between locations -- you cannot defy the speed of lot. The TCP protocol can also contribute to latency, as can chattiness within the application itself. So you may want to consider some TCP acceleration techniques (like selective acknowledgements and adjustable window sizes) and/or data reduction to minimize traffic across the WAN. Packet loss typically appears on MPLS and IP VPN networks where routers can become oversubscribed. If you are exploring either of those WAN technologies, you should expect some degree of loss on your WAN despite service provider claims). Techniques like Forward Error Correction (FEC) can be quite effective in those situations.
This was first published in August 2007