This deliverable reports and summarises the experimental results obtained during the first trial. The primary objective of these experiments was to verify the AQUILA architecture for providing QoS in the IP network (described in previous deliverables D1201 and D1301). In particular, the reported results cover the following areas: evaluation of network services, experiments with legacy applications supported by defined network services, validation of admission control algorithms, validation of resource management functions (e.g. resource pool mechanisms) and performance evaluation of the signalling system.

On the basis on the analysis of the results obtained during the first trial, the main conclusions and hints for the next stage of the AQUILA project are the following:

• Regarding network services:
  
  
  • Implementation of each network service is in accordance with the assumed specification;
  • Efficiency of AC algorithms agree with the assumptions;
  • Tuning the appropriate values of traffic descriptors for real applications is sometimes very difficult to do. For instance, in the case of the NetMeeting application it can be done only experimentally; therefore, there is the suggestion to simplify traffic descriptors;
  • It was proved that mixing streaming and elastic traffic inside one network service should be avoided;
  • PCBR and PVBR network services, dedicated for serving streaming traffic, guarantee the assumed target QoS requirements (like packet delay characteristics, packet loss ratio);
  • PCBR network service is well suited for applications generating constant bit rate traffic, like WinSIP;
  • PVBR network service is well suited for applications generating variable bit rate traffic, like NetMeeting,
  • PMM and PMC network services, dedicated for serving elastic traffic, guarantee the target throughput requirements, while fail in guaranteeing target packet loss rate; this requires re-design of particular mechanisms associated with these services;
  • The capacity allocated to the PMM or PMC service is fairly shared among all accepted TCP-controlled flows (in the case of PMM service the throughput is proportional to the declared SR value while in the case of PMC to the calculated equivalent bandwidth);
  • PMM service is well suited for serving traffic produced by greedy TCP-controlled sources (like FTP) and adaptive streaming video (like Real-Player) while PMC service is better suited for serving traffic produced by non-greedy TCP sources;
    
    
• Regarding RCL layer:
  
  
  • Resource pool mechanism works correctly for TCL1, TCL3 and TCL4, but should be re-designed for TCL2, the trial will continue;
  • In the initialisation phase, most of the signalling is produced by the connection between the RCA and database; the second largest part of signalling is produced by ACAs (for configuration of edge devices);
  • During reservation set-up the largest contribution to signalling traffic is produced by ACA logging; the second largest contribution to signalling came from the ACA; the third largest contribution to signalling was the database communication;
  • The set-up and release times were reasonable for production use. Times for making and releasing the reservation were the same, about two seconds each;
  • The most critical point of failure is the database; the second critical point is the RCA.

An extended summary of the first trial results is presented in chapter 3. In Annexes A and B (chapter 6 and 7) a detailed description of trial scenarios and results is included.