![[INET 2000]](/web/20160103035746im_/http://www.isoc.org/inet2000/cdproceedings/inet00sm.gif)
![[ Up ]](/web/20160103035746im_/http://www.isoc.org/inet2000/cdproceedings/up.gif)
![[Prev]](/web/20160103035746im_/http://www.isoc.org/inet2000/cdproceedings/prev.gif)
![[Next]](/web/20160103035746im_/http://www.isoc.org/inet2000/cdproceedings/next.gif)
This paper evaluates the robustness of multicast routing and the performance evaluation of multicast service using the experimental network. For the performance evaluation of multicast service, we use digital video (DV) transmission service, as the practical application for the next generation Internet. The evaluation system uses protocol independent multicast (PIM), which is a reverse path forwarding (RPF) multicast routing protocol. The packet loss observed at the end host is evaluated, compared with the case where the system provides the multicast service using N of the multiple unicast packet transmission. We can make sure the multicast system has better system scalability, regarding the number of receivers, than the unicast-based multicast system has. PIM operates correctly and stably, even when the network has a routing loop. Finally, we operate the DV multicast service over the nationwide JB project IPv6 network using the PIM-SM, to ensure the correct operation.
Most of the data communications over the currently operating Internet uses unicast data transmission with TCP or UDP. Many existing applications do transfer exactly the same data to a large number of receivers. Also, upcoming Internet applications will includes realtime multicast services, such as digital video (DV) program multicasting service or interactive multi-party multimedia conferences or games. We have to establish a stable and cost effective multicast data transmission infrastructure for the next generation Internet applications.
With multicast data transmission, when a sender transfers a single packet toward the particular multicast group, the network copies the packet to transfer it only to the particular receivers belonging to the corresponding multicast group. Since the sender does not need to send the same data to every receiver, the network can provide an efficient multicast data transmission services. With the multicast data transmission, the required processing power at the sender and the required bandwidth for the sender can be smaller than those with the unicast data transmission. Since the total number of packets transferred by the network with multicast data transmission is smaller than that with unicast data transmission, the possibility of inappropriate routing protocol operation due to the large amount of user packet transmission with multicast data transmission would be less than that with unicast data transmission. Also, the diversity of the data reception time by the receivers must be smaller than with the unicast data transmission, and the data reception delay by the recievers must be smaller than that with the unicast transmission.
On the other hand, the multicast service may have the following technical concerns:
The system evaluated in this paper has the following features:
We used a PC router that can operate the PIM-SM with IPv6, and a PC host that can send and receive the DV data using the PIM-SM with IPv6. Using these PC-based routers and hosts, we developed both the local area and the wide area experimental networks. The wide area experimental network accommodated the live traffic with conventional applications, as well as the experimental DV multicast application. Also, the experimental network was not a loop free topology. Therefore, the experimental network would create a transitional routing loop with some network status.
Using the experimental network, we evaluate the performance of packet transmission, compared with the performance where the multicast service is provided by the multiple unicast packet transmissions. Also, we evaluate the stability of routing protocol operation with high speed DV data transmission. It was shown that the experimental system works well with the high speed DV data multicast application, even when the system has a routing loop. As a further study item, it is recognized that we should evaluate the effectiveness of the packet scheduling mechanism to provide different quality of service among system control information (e.g., routing protocol information) and user information.
Section 2 evaluates the performance of packet transmission, compared with the case where the multicast service is provided by the multiple unicast transmissions. Section 3 evaluates the robustness and the stability of the multicast routing protocol. Section 4 describes the experimental operation of DV stream multicast service over the nationwide JB project IPv6 network using IPM-SM. Finally, section 5 gives a brief conclusion.
Figures 1 and 2 show the evaluation system. Sender (Snd1) sends the packets to two receiver hosts (Rcv1 and Rcv2) through the routers (R1 and R2). Rcv1 and Rcv2 receive the same data from Snd1, i.e., multicasting from Snd1 to Rcv1 and Rcv2. R1 and R2 are connected through the OC-3 ATM link, and the hosts (Snd1, Rcv1 and Rcv2) are connected to the routers through the 100MBase-T links.
Figure 1. Experimental Configuration for Multicast by Multiple
Unicasts
Figure 2. Experimental Configuration for Multicast by Multicast
Session
In the configuration of Figure 1, Snd1 establishes two (unicast) packet flows to each receiver host. This is the case where the multicast service is provided by the multiple unicast connections. Routers (R1and R2) do not need to copy the received data, and the sender host (Snd1) copies the sending data. In the configuration of Figure 2, Snd1 establishes one (multicast) packet flow to the receiver hosts. This is the case where the multicast service is provided by a single multicast connection. The router R2 copies the received data to deliver the data to receivers (Rcv1 and Rcv2). Here, with the configuration of Figure 2, PIM-SM is applied as the multicast routing. Sender host (Snd1) can control the packet transmission rate to the network, with the rate control (i.e., shaping).
All nodes that are routers and hosts are ordinary IBM-compatible PCs with the following specifications:
Table 1 and Figure 3 show the number of received packets and the number of dropped packets at the router (R1), parameterizing the packet transmission rate from the sender host (Snd1).
| Transmission Rate | Number of Received Packets with Unicast | Number of Dropped Packets with Unicast | Number of Received Packets with Multicast | Number of Dropped Packets with Multicast |
|---|---|---|---|---|
| 10Mbps | 7414 | 68 | 7536 | 86 |
| 30Mbps | 7517 | 156 | 7413 | 74 |
| 50Mbps | 7814 | 1012 | 7413 | 87 |
| 100Mbps | 7814 | 2568 | 7586 | 98 |
Figure 3. Packet Loss Rate at the Router
As shown, the router cannot relay the unicast packets, in accordance with the increase of the packet transmission rate from the sender with unicast-based multicast. However, with the multicast service, the router can relay the packets and the packet dropped rate does not increase even when the packet transmission rate from the sender host (Snd1) increases.
The RPF represents Reverse Path Forwarding. RPF is commonly used in the major multicast routing protocols, such as PIM that is used in the evaluation system discussed in this paper. In the RPF system, the multicast packet transmission is executed using the unicast routing information. The multicast session is recognized and managed by the pair of source node IP address ("S") and IP multicast address assigned for the multicast group ("G"), i.e., {S,G}. When the router receives the multicast packet, the router checks the source IP address in it.
if (Recv_I/F = Nxt_Hop_R_I/F)
{forward the received packet to the rest of interfaces}
else
{silently discard the received packet}
endif
Here,
Therefore, the RPF multicast routing does not need any additional routing mechanism for a multicast service, i.e., it only needs the unicast routing information. Also, when the IP multicast packet is received from the wrong interface compared to the unicast routing information, the received multicast packets are automatically and silently discarded.
This mechanism avoids the packet forwarding to the routing loop. When the unicast routing system has the routing loop, the packet transferred into the loop will be discarded due to the mismatch of unicast routing information to the source IP address and the interface receiving the IP multicast packet. This function of RPF multicast routing has a significant benefit, compared with the unicast-based multicast service. With the unicast-based multicast service, the packet has to be looped in the network until the TTL expires. However, with the multicast service using the RPF, it is expected that the packets will be discarded at the entry router to the routing loop.
Figure 4 shows the evaluation system. Sender (Snd1) sends the multicast packets toward two receiver hosts (Rcv1 and Rcv2) through the routers (R1, R2, and R3). All routers run the PIM-SM multicast routing protocol. All the data links in the system are 100MBase-T.
Figure 4. Experimantal Configuration for Evaluation of Multicast Routing
Against Routing Loop
All nodes that are routers and hosts are ordinary IBM-compatible PCs with the following specifications:
In the evaluation system, the R1 has the inappropriate unicast routing information. For R1, the next hop router to transfer the packet toward Snd1is the R3. Therefore, the routing toward the Snd1 has the routing loop among R1, R2 and R3.
The multicast packets are transferred to the first hop router R1. R1 has to discard the all the received multicast IP packets from the Snd1 node, since the interface receiving the IP multicast packets is different from the interface RPF expected. We manually modify the routing information at R1, so that the evaluation system has a routing loop among R1, R2, and R3. By the application of the RPF mechanism, it is expected that R1 silently drops the multicast packets.
The experimental system actually discards the IP multicast packets at R1, when R1 has the wrong unicast routing information to form a routing loop. When we use the multiple unicast connection to provide the multicast service from Snd1 to Rcv1 and Rcv2, the unicast packets do loop among R1, R2, and R3. Moreover, the packet to be looped will be two packet flows (both toward Rcv1 and toward Rcv2). We made sure the routing loop had generated the congestion with unicast-based multicast service. Also, we made sure to avoid packet transmission at the entry router of the (unicast) routing loop, to avoid the network congestion due to the unicast routing loop.
We evaluated our developed multicast system over the nationwide JB project network. The JB network [1] is jointly operated among WIDE project [2], CKP [3] and ITRC [4]. Most of the high-speed links are provided by the JGN (Japan Gigabit Network) operated by TAO [5], by the TTNet [6] and by the CRL [7]. It uses IPv6 as the basic Internet protocol. Figure 5 shows the network configuration of the JB network, regarding only the high-speed links. More than 10 organizations are interconnected by the high-speed datalinks (e.g., OC-12 ATM, Gigabit Ethernet). All organizations are interconnected through the PC-based routers with KAME IPv6 protocol stack, and run the PIM-SM for multicast service. We had two rendezvous point (RP) routers in the network, to create shared multicast trees.
Figure 5. Network Configuration of the JB Project IPv6 Network
Figure 6 shows the overview of DV multicast service over the JB project IPv6 network. The technical workshop held by the WIDE project is multicasted to more than 10 sites over the nationwide JB project network, using the DV video stream. The DV stream is multicasted to the sites using the PIM-SM with IPv6. Each site observes the same DV image simultaneously. The participants from the remote sites can interactively join the workshop using the network. This means that the remote sites can send a DV stream whenever they have a question. The DV stream from the remote sites also multicasted to all the participating sites through the RP routers. The experimental interactive multicast session for one-day workshop using the DV multicasting has been correctly operated using the PIM-SM.
Figure 6. Overview of DV Multicasting over the JB Project Network
This paper evaluates the robustness of multicast routing and the performance evaluation of multicast service using the experimental network. For the performance evaluation of multicast service, we use digital video transmission service as the practical application for the next generation Internet. The evaluation system uses Protocol Independent Multicast (PIM), which is a reverse path forwarding (RPF) multicast routing protocol. The packet loss observed at the end host is evaluated, compared with the case where the system provide the multicast service using N of the multiple unicast packet transmission. We can make sure the multicast system has better system scalability, regarding the number of receivers, than the unicast-based multicast system. PIM operates correctly and stably, even when the network has a routing loop. Finally, we operate the DV multicast service over the nationwide JB project IPv6 network using the PIM-SM, to ensure the correct operation.
The following items would be further study items:
[1] M.Minami, M.Oe, K.Okamura, Y.Kadobayashi, A.Ogawa, K.Nagahashi, H.Esaki; "JB: Design and Architecture of Next Generation Internet Infrastructure in Japan," ICCC99, Tokyo, September 1999.
[2] WIDE Project, http://www.wide.ad.jp
[3] CKP (Cyber Kansai Project), http://www.ckp.or.jp
[4] ITRC, http://www.itrc.net
[5] JGN and TAO, http://www.tao.go.jp/JGN/
[6] TTNet, http://www.ttnet.co.jp
[7] CRL, http://www.crl.go.jp
[8] D.Estrin, et al., "Protocol Independent Multicast-Sparse Mode (PIM-SM): Protocol Specification," IETF RFC 2362, June 1998.
[9] Y. Rekhter, T. Li , "An Architecture for IPv6 Unicast Address Allocation," IETF RFC 1887, December 1995.