This section introduces satellite networks. This section includes the following topics:

You can use satellite networks for WAN connectivity for people in remote locations or when users are in a temporary or mobile environment (for example, a ship or an oil rig). Satellite networks have several characteristics that differ from terrestrial networks. The most prevalent differences include the following:

These characteristics can create challenges for traditional TCP algorithms for various reasons. This chapter helps you to understand and address these characteristics.

When it comes to TCP, latency is the enemy. The higher the latency, the longer it takes ordinary TCP to ramp up and use the available capacity. In satellite networks, latency is typically 10 to 100 times higher than in terrestrial links. Latency in a single-hop satellite network usually varies from 550 ms to 800 ms. Dual-hop satellite network latency commonly ranges from 1100 ms to 1600 ms. With INMARSAT BGAN networks (broadband satellite communications) it is not unusual to observe peak latencies of 2000 ms and higher when congestion is present.

Even though the SteelHead has TCP optimization enabled by default, tuning transport settings for a target satellite environment results in even better performance.

Satellite communications are known for having less than perfect packet delivery. Loss can wreak havoc on TCP throughput. If the loss is high enough, it can cause sessions to time-out. Loss can happen for several reasons on satellite networks. You must determine the primary cause of loss in your network so that you can tune your TCP optimization accordingly.

The two primary reasons for loss in satellite networks are congestion and poor signal quality. Do not ignore the possibility of a simple bad cable or speed and duplex mismatch when troubleshooting. Congestion-based loss is normal and expected when a link or path is saturated. Congestion-based loss triggers a TCP stacks congestion avoidance algorithm so that individual TCP sessions can adapt intelligently. Poor signal quality increases the bit error rate (BER), which is not due to congestion, but rather because of satellite communications. BER is independent of load on the circuit.

Many different events can trigger increased BER. A few of the most common are obstructions (buildings, bridges, and so on), a misaligned satellite dish, and weather. Weather events are temporary and go away in time. With obstructions or poor angles, the loss can be more persistent. Most satellite modems have forward error correction (FEC) built-in for free, and can overcome an increased BER in many situations.

In some cases, the BER loss is too severe for the satellite modem to overcome with FEC. In these situations, you might detect consistent TCP loss. Ideally, you should solve the root cause of the increased BER, but as a network administrator, this might be out of your control. In this situation, you can leverage a TCP Stack optimized for a high error environment.

You can choose from many satellite transport options. Their performance characteristics vary widely. A reasonable estimation is the more terminals sharing a segment, the higher the possibility for variable latency and loss. This section provides a brief summary of popular satellite solutions, along with general assumptions of loss and latency expectations. You should verify these details for your network because they might be different than those in the following table.