Multicast tree construction algorithms for multimedia service are largely classified into Source based tree Shared. The Shared tree based multicast is known to show better performance in the aspect of scalability than source based tree. Generally, there have been lots of researches on the algorithm to satisfy QoS constraints through proper Rendezvous Point (RP) in the shared tree. In addition, as the multicast group members join and leave dynamically during the session, the RP shared tree should be reselected to satisfy the QoS of the new members. But, there is not much previous works on the RP reselection algorithm to guarantee peer-to-peer service quality. In this thesis, both of new initial RP selection (iRPSA) and RP reselection (nRPSA) algorithms are proposed, which use RTCP(REAL Time Control Protocol) report packet fields.
Conventional RP selection algorithm can be classified into two algorithms depending on multiple QoS constraints. Multiple QoS constraints based RP selection algorithm considers multiple QoS constraints such as delay, delay jitter, and bandwidth simultaneously. RP selection algorithm without QoS constraints can be divided into three areas, which are random, topology based, and group based RP selection, respectively. Conventional RP selection algorithms, such as, Random, Topology based, and Group based algorithm cannot fully guarantee QoS constraints.
Recent studies on RP selection report the RP selection algorithm considering multiple QoS constraints. In order to provide the real time multimedia service, the complexity of QoSconstraint algorithms should be reduced considerably.
As the multicast group members join and leave dynamically during the session, it is difficult to guarantee the QoS constraints for some of newly joining members. In that case, new RP should be selected to guarantee QoS for every group member. There are two algorithms to select new RP. One is to select new RP considering member dynamic sensitivity that indicates the temporal variation of service sessions. The other is to execute RP reselection algorithm periodically.
However, in general, RP selection causes the process overhead to obtain information for the RP decision and to distribute the new RP information to all group members.
In this thesis, we propose two kinds of new RP selection algorithms based on RTP/UDP/IP-multicast service model. Those are initial RP selection algorithm called iRPSA( RTCP-based initial RP Selection Algorithm) and RP reselection algorithm called nROSA(RTCP-based new RP Selection Algorithm)/iRPSA selects initial RP under QoS constraints obtained by the RTCP receiver report packet sent by every multicast group member. Senders in the multicast group send the RTCP sender report packet and receivers reply with the RTCP receiver report packet.
We can obtain QoS information using the RTCP sender/ receiver reports fields in the RTCP report packet. Delay between two peers can be calculated with the difference among three times such as Delay since last sender report (DLSR), Time of last sender report (TLSR) fields in RTCP receiver report packets and report receiving timestamp. Inter-arrival delay jitter and packet loss rate are obtained from the inter-arrival jitter field and cumulative number of packet lost fields in RTCP receiver report packet, respectively. Inter-destination delay jitter can be calculated with the delay values received from all the other multicast group members.
Boot Strap Router(BSR) in the PIM-SM protocol is required to control RP selection mechanism. After a sender receives FTCP receiver report from all other receivers, it reports the QoS information to BSR. Therefore BSR receives QoS information from every group member. BSR calculates RP based on these QoS information and reports selected RP information to all members.
Moreover, multicast group member size can be calculated by RTCP report. It is possible because every multicast group member sends RTCP report packet to all other members periodically. Direct counting method estimates multicast group size with RTCP report.
On the other hand, nRPSA has three steps to identify new RP location on the network. First step of nRPSA is to estimate multicast group size based on direct counting method. Direct counting method is to calculate multicast group size by discrimination "SSPC of sender" field of RTCP receiver report packets.
The Second step is to calculate QoS sensitivity parameter as the criterion for the time to perform RP recalculation. QoS parameter sensitivity is defined as the ratio of total multicast group member sizes obtained by direct counting method to newly joining group members size which do not satisfy QoS constraints criteria. QoS parameter sensitivity is the threshold to trigger RP reselection algorithm.
Finally, new RP is positioned from QoS parameter sensitivity in the second step. RP selection of nRPSA is classified into three parts based on QoS parameter sensitivity criterion. When the QoS parameter sensitivity calculated is below □,there is no change of the RP location. But, when it becomes over the value □, RP will be moved to the nearest member towards the majority of new members who don't satisfy QoS constraints. iRPSA will be used to calculate new RP location when the QoS parameter sensitivity becomes over the value □.(□)
In this thesis, we have simulated the proposed algorithm with NS simulator. Network topology is generated with Tiers up to 100 nodes. Simulation results show that proposed RP selection algorithm has better performance than random and topology based one . And, RP reselection algorithm improves delay performance after initial RP selection as the multicast group members join and leave dynamically during the session.
Simulation result show that proposed initial RP selection algorithm based on RTP has better delay performance than existing random and topology based one by 32~56%. In addition, proposed RP reselection algorithm has the results to improve the delay between multicast group member by 24~40% after RP reselection according that the group members join and leave dynamically.
Proposed two algorithms initial RP selection and RP reselection algorithm use RTCP protocol which underlies multimedia application service. So, proposed algorithm does not need any special process for collecting network information to decide RP. Proposed algorithm can be used to support real time multimedia services in high-speed Internet wide area network environments.
인터넷에서 멀티미디어 서비스를 지원하기 위한 멀티캐스트 프로토콜은 트리 구성 방식에 따라 공유 트리 방식과 소스 기반 트리 방식으로 나눌 수 있는데, 공유 트리 방식이 확장성 측면에서 보다 우수하다고 알려져 있다. 공유 트리 방식에서 QoS를 만족시키기 위해 고려해야 될 핵심 사항 중 하나인 RP (Rendezvous Point) 선정에 대하여 기존에는QoS 제약 조건에 따라 별도의 과정을 통해 RP를 계산하는 방식으로 연구가 이루어지고 있다.
또한, 멀티캐스트 그룹 멤버가 동적으로 가입 또는 탈퇴를 할 경우에는 초기에 설정된 RP가 신규 멤버까지는 QoS를 만족시키지 못하는 경우가 발생하게 되므로, 신규멤버 까지 QoS를 만족시키는 새로운 RP를 재선정할 필요가 있는데, RP재선정에 관한 기존 연구는 매우 미흡한 실정이다.
본 논문에서 제안하는 방식은 RTP/UDP/IP-multicast 서비스 모델을 기반으로 RP를 선정하는 방식으로 RTCP 패킷으로부터 QoS정보를 수집함으로써 별도의 정보 수집 절차를 필요로 하지 않는다. 이 방식은 RTCP수신자 통보(RTCP receiver report) 패킷의 필드를 이용하여 멀티캐스트 그룹 멤버 각각으로부터 QoS 요소 값을 판별하고, 그 값으로부터 초기 RP로 선정한다.
QoS 요소 값은 RTCP 송신자 통보 (RTCP sender report) 패킷과 RTCP 수신자 통보 (RTCP receiver report) 패킷의 일부 필드를 이용하여 도착시간 지연변이, 지연, 패킷 손실 율, 목적지간 지연 변이 등으로 정의할 수 있다.
RP를 초기 선정하기 위해서는 전체 프로세스를 주관하는 서버가 필요한데, 이는 PIM-SM에서 사용하는 BSR (Boot Strap Router)을 이용한다
또한, 멀티캐스트 그룹 멤버의 동적 변화에 따라 네트워크의 QoS가 달라지게 된다. 따라서, 멤버의 동적 변화를 고려한 RP재선정 방식이 필요하게 된다.
본 논문에서는 RP 재선정 단계를 3단계로 나누어 구분한다. 1단계는 전체 멀티캐스트 그룹의 크기를 추정하는 단계로 직접 산출 방법을 사용하고, 2 단계는 QoS 요소 기준 값을 만족시키지 못하는 그룹 멤버 수를 측정하여 QoS지수 민감도를 계산하는 단계이다. 2단계에서 구한 QoS 지수 민감도를 기준으로 3단계에서는 RP를 재선정 하는 방식을 결정한다.
본 논문에서는 모의 실험을 NS-2 Simulator를 이용하여 실시하였는데, 제안한 알고리즘에 따라 RP를 초기 선정하고 재선정하는 방식을 기존 방식과 비교하여 종단간 최대 지연 및 평균지연으로 성능을 평가하였다.
모의 실험을 통하여 제안된 방식이 초기 RP선정 시 에는 임의 선정, 위상 기반 선정 방식보다 32 ~ 56% 정도, 멀티캐스트 그룹멤버의 동적 변화 시에는 초기 RP를 그대로 이용하는 방식보다 24% ~ 40%의 개선 효과를 보였다.
