Performance Enhancement Proxies (PEPs) are key elements for meeting basic performance requirements for interactive Internet usage via satellite.
Accordingly, the SatLabs group has produced the specification for an Interoperable PEP (I-PEP). The objective of the activity is to develop an alternative congestion control scheme [to both standard TCP and TCP Vegas currently included into I-PEP stack], fully interoperable with the I-PEP specification and specifically aimed at optimizing web browsing performance in the I-PEP environment, while maintaining good performance for other TCP traffic such as large files. The resulting transport protocol will be named "TCP Noordwijk".
The output of the activity will be a full specification of the new transport protocol, validated by simulation, together with a prototype implementation. A exhaustive test campaign, involving both DVB-RCS network emulators and real DVB-RCS systems, will be performed in order to evaluate protocol performance and verify the matching with the project requirements. Protocol implementation will rely on the Mitre SCPS-TP implementation.
The ITT addresses an important issue that can be considered as an enabling technology for the successful market uptake of satellite services for Internet access. Although threatened in the volume of traffic it generates by the growing importance of video streaming over the Internet, web browsing is still one of the main Internet applications. Even in a scenario of huge growth of video traffic, web browsing will expectedly remain as an important usage of the Internet. The commercial acceptance of Internet browsing over satellite depends critically on finding mechanisms for a satisfactory subjective browsing experience.
The main benefit expected from this activity involves the optimization criteria, which will be to lower latency of HTTP transfers, while preserving also performances of large file data transfers (e.g. FTP).
This new TCP-based protocol will ensure compatibility with a standard TCP implementation and interoperability with regard to the I-PEP specification.
In principle, the capabilities of a congestion control algorithm depend on the kind of protocol control mechanisms and network state information available. For this purpose, the I-PEP specification defines the use of a number of SCPS-TP and TCP mechanisms, options, and extensions, some optional and some mandatory. Given the influence of the available options on the capabilities of the resulting algorithms, the study has made a systematic exploration of the possible combinations, starting from the baseline defined by the mandatory features and from here progressing to the possible combinations of options.
The studies conducted in early stages of the project have shown that is possible to increase the I-PEP performances, avoiding a slow start behaviour and adopting instead a bursty sending strategy. The burst size and the time elapsed between burst sending are calculated by means a statistical analysis of acks interarrival times (acks dispersion).
All TCP Noordwijk functionalities must be implemented at the sender side only in order to guarantee interoperability with current SCPS-TP profile for I-PEP.
The following figures show the evolution of sequence numbers for a single connection and 3 overlapping connections, respectively:
The performance of the transport protocol, especially concerning the interaction between different flows, may also depend on the architecture of the I-PEP in the layers between the transport layer and the DVB-RCS access layer. In order to perform a suitable coverage, the study considers a number of scenarios, each defined by a specific set of assumptions on options and architectural features. Different DVB-RCS resource allocation mechanisms, namely, Constant Rate Allocation (CRA), Volume Based Dynamic Capacity (VBDC), and Rate Based Dynamic Capacity (RBDC) will be evaluated. It would be in principle expected that burst-based control schemes would work better over VBDC and rate-based schemes better over RBDC, but this hypothesis should be systematically investigated.
The project has been divided into two phases:
Phase 1: Analysis, design and preliminary validation of the protocol.
Phase 2: Implementation of prototype software. Detailed validation of the protocol.
The project started on 1st November 2006.
Phase 2 of the project is completed, including:
The result of phase 2 shows performance increases compared to classic transport protocols were evident in most of the testing conditions with regard, for instance, to throughput reached by TCP-Noodwijk connections against TCP-Vegas default transport protocol in VBDC DAMA. These results have suggested a project extension in order to verify the interoperability of TCP-Noordwijk implementation with a commercial I-PEP.
This project allows us to experiment with a relevant international collaboration between the private industry and excellence coming from the Academic world, joining the high competence and capacity of theorizing and designing with the experience in the development of solution market oriented. In this framework, a very important role is represented by ESA as scientific driver, funding entity and a reference point for the European scientific and industrial entities.