TCP Extensions for Space Communication (aka SCPS-TP)

R.C. Durst and G.J. Miller (MITRE) and E.J. Travis (Gemini Industries)

One-line summary: Roll together many TCP-for-wireless optimizations, but change them slightly, in order to adapt TCP for communication with satellites, the space shuttle, etc.

Overview/Main Points

Standard problems: long fat pipe; errors from radiation, etc. may be bursty or periodic (orbit); asymmetric (typ. 1000:1); TCP aborts after n retransmissions
Use TCP Vegas (RTT, not loss, indicates congestion). Since doesn't use receive window size as upper bound on pipe filling, enables large windows for long fat network.
Use open loop rate control rather than ACKs to clock out data.
Explicit notification via ICMP of corruption or outage from downstream routers: sources enter persistent mode, probing until outage goes away. When no explicit notification, assume a default cause of loss (configurable).
Use "stateless" TCP header compression (similar to RFC ??? but nonconformant) to avoid "cascading loss" problem.
Use selective NAK rather than go-back-N (similar to RFC 2018, but nonconformant)

Relevance

TCP In Space! About time we started adapting proven protocols rather than rewriting from scratch for each mission (this is what NASA has done to date).

Flaws

Too bad it's non-interoperable, but the changes seem innocuous and could be handled in gateway routers.
Except for the explicit notification of reason for lossage, this is just all of the wireless TCP improvements rolled into one.

Back to index