EXPERT RESPONSE
There are many possible explanations for packet loss within a wide area network. The most common causes are as follows:
- Congestion somewhere along the path. Congestion occurs when the amount of traffic destined for a particular link exceeds the capacity of that link. The switch or router will buffer some packets, but eventually packets must be discarded. Congestion is most common at the entrance to a wide area network -- where the high capacity LAN meets a lower capacity access link -- and at the exit from the wide area network -- where the service provider's large network meets the lower capacity access link.
- Policing or shaping in the service provider network. Service providers often police or shape customer traffic to ensure fair service and/or to enforce service parameters like a frame relay CIR (committed information rate). In this case traffic may be dropped, even though there is no real congestion.
- Loss due to bit errors corrupting the packet contents. With modern transmission technology this is less likely, but some links may have high bit error rates. Corrupted packets may be detected at Layer 2 and the packets will be discarded in the network.
It is often difficult to determine when and how loss is affecting a particular application. Even small amounts of loss (less than 1%) can have a very dramatic effect on application performance. Standard TCP backs off dramatically when loss is observed on a link with a high round trip time (RTT). With just 1% loss, an FTP over a network with a 100 ms RTT may drop to less than 1 Mbps, even if there is actually 10 or 100 Mbps of capacity available.
WAN optimization appliances sometimes include the capability to both measure loss between locations, and to mitigate for loss by using forward error correction or forward packet recovery techniques.
|
