IEEE COMMUNICATIONS LETTERS, VOL. 15, NO. 9, SEPTEMBER 2011
1013
Centralized Selective Multicast Retransmission Policy for Enhanced Resource Efficiency in EPON-Based Access Networks YoungHwan Kwon, Member, IEEE, Min-Gon Kim, Member, IEEE, Seong Gon Choi, Member, IEEE, and Jun Kyun Choi, Member, IEEE
Abstract—Based on the combination of an EPON and various last-mile connections, EPON-based access networks can provide broadband access services economically. It, however, is imperative to create an optimized multicast retransmission policy in consideration of such a combined architecture. To resolve this issue, this letter proposes a new Centralized Selective Multicast Retransmission Policy (CSMRP) that efficiently retransmits multicast traffic. Performance results show that the proposed CSMRP can effectively reduce bandwidth consumption and queuing delay for enhanced resource efficiency. Index Terms—EPON, multicast, retransmission, packet loss.
I. I NTRODUCTION
E
THERNET Passive Optical Networks (EPONs) easily support multicast delivery by broadcasting data through a shared optical Point-to-MultiPoint (P2MP) connection from an Optical Line Terminal (OLT) to Optical Network Units (ONUs) [1]. As a valid evolution of access networks, EPONbased access networks have been widely deployed to provide broadband access services economically by combining an EPON and various last-mile connections, such as Digital Subscriber Line (DSL) and WiMAX [2]. Due to such broadcast property and cost-efficiency, EPON-based access networks have garnered much attention as one of the most promising access technologies to provide multicast applications efficiently, especially Internet Protocol Television (IPTV) [2]. On the other hand, EPON-based access networks are associated with a fundamental problem of packet loss due to transmission error, which is caused by electro-magnetic interference in DSL and by unstable wireless environments in WiMAX, respectively [3], [4]. Thus, in EPON-based access networks, it is essential to recover lost multicast packets because a few packet losses can seriously degrade IPTV QoS [3]. Therefore, several previous studies have assessed retransmission for multicast applications to recover packet loss. To exemplify, Unicast Retransmission Policy (URP) [5] retransmits lost packets to a receiver with an End-to-End (E2E) unicast retransmission. As a means of reducing duplicated
Manuscript received May 17, 2011. The associate editor coordinating the review of this letter and approving it for publication was A. M. Tehrani. This work was partly supported by the the IT R&D program of MKE/KEIT [10039160, Research on Core Technologies for Self-Management of Energy Consumption in Wired and Wireless Networks]; and by the ITRC support program of MKE/NIPA (NIPA-2011-(C1090-1111-0013)). Y. H. Kwon, M.-G. Kim, and J. K. Choi are with the Korea Advanced Institute of Science and Technology (KAIST), Daejeon, Rep. of Korea. (email:
[email protected],
[email protected],
[email protected]). S. G. Choi is with ChungBuk National University (CBNU), Cheongju, Rep. of Korea (e-mail:
[email protected]). Digital Object Identifier 10.1109/LCOMM.2011.072911.111025
retransmissions among multiple E2E unicast retransmissions in networks, Multicast Retransmission Policy (MRP) [5] retransmits lost packets once to all receivers of a multicast group with an E2E multicast retransmission. As a more advanced solution to overcome the unnecessary retransmissions to unrelated receivers in an E2E multicast retransmission, Selective MRP (SMRP) [6] was proposed to retransmit lost packets selectively to the associated receivers of a multicast group with hop-by-hop constrained forwarding. This forwarding uses hop-by-hop P2P retransmission to retransmit lost packets selectively. However, the legacy policies have their limitations in EPON-based access networks, which consist of a P2MP connection of an EPON and last-mile P2P connections. URP incurs duplicated retransmissions in the P2MP connection due to multiple E2E unicast retransmissions to each receiver. MRP also induces unnecessary retransmissions in last-mile P2P connections due to an E2E multicast retransmission to all receivers of a multicast group. Besides, SMRP regenerates duplicated retransmissions in the P2MP connection because of multiple hop-by-hop P2P retransmissions to each ONU. To complement the aforementioned limitations of the legacy policies in EPON-based access networks, this letter proposes a new Centralized SMRP (CSMRP) to prevent both duplicated and unnecessary retransmissions simultaneously. When retransmitting multicast packets, the OLT broadcasts the retransmitted packets once to ONUs during the Round Trip Time (RTT) to prevent duplicated retransmission in an EPON. Subsequently, the associated ONUs selectively unicast the retransmitted packets to the associated receivers, whereas other unrelated ONUs drop the retransmitted packets in order to avoid unnecessary retransmission in last-mile connections. In the following sections, the detailed operation of the proposed CSMRP will be presented in EPON-based access networks and its effectiveness in terms of bandwidth consumption and queuing delay is then shown compared to the legacy policies. II. P ROPOSED C ENTRALIZED S ELECTIVE M ULTICAST R ETRANSMISSION P OLICY (CSMRP) This section describes the characteristics of EPON-based access networks and the detailed operation of CSMRP. Fig. 1 presents the architecture of EPON-based access networks. Essentially, they consist of a P2MP connection from an OLT to ONUs in an EPON, and P2P connections from ONUs to receivers through DSL Access Multiplexer (DSLAM) or WiMAX Base Station (BS). Due to the broadcast property in the EPON, the OLT easily transfers multicast packets to a group of ONUs without a copy function. The OLT has a cache
c 2011 IEEE 1089-7798/11$25.00 ⃝
1014
IEEE COMMUNICATIONS LETTERS, VOL. 15, NO. 9, SEPTEMBER 2011
Multicast groups
1 DSLAM
OLT 1
WiMAX BS
n Cache
n
Splitter
pkti
ONU1
(iii) NACK handling
1
ONU2
DSLAM
Group 1
n
Multicast server n
1
n
ONUk A shared P2MP connection
Fig. 1.
1
1
n
1
n
OLT
(v) Selective unicasting retransmission pkti
ONU1
1
Multicast server 1
(iv) Broadcasting retransmission pkti
(i) Data caching pkti
1
1
Multicast packets : n Optical fiber : DSL connection : WiMAX connection :
WiMAX BS
pkti
Receiver1-1 Receiver1-2
Group n
P2P connections
EPON-based access network architecture for CSMRP.
Receiver1-3
Receiver2-1
Fig. 2.
pkti
pkti (ii) Loss detection
pkti
pkti
pkti
ONU2
Multicast transmission :
for retransmission. For multiple last-mile P2P connections per ONU, each ONU copies multicast packets and transmits those to each P2P connection. In this combined architecture, URP or SMRP induces duplicated retransmissions in the P2MP connection due to multiple E2E unicast retransmissions from an OLT to receivers in URP or multiple hop-by-hop P2P retransmissions from an OLT to ONUs in SMRP. Also, MRP causes unnecessary retransmissions in the P2P connections in the event that an E2E multicast retransmission includes unrelated receivers. Hence, it is clear that a new retransmission policy is needed because the legacy policies are not designed and optimized to support EPON-based access networks. To resolve this issue, we propose CSMRP to efficiently retransmit for multicast traffic in EPON-based access networks. The detailed operation of CSMRP is presented in Fig. 2 with the multicast group 1 in Fig. 1. CSMRP operates with the following five procedures: (i) data caching, (ii) loss detection, (iii) Negative ACKnowledgement (NACK) handling, (iv) broadcasting retransmission, and (v) selective unicasting retransmission. First, the OLT sends multicast packets to all receivers of a multicast group via ONUs after it conducts data caching to save the multicast packets in its cache with the caching time of the i𝑡ℎ multicast packet (𝑐𝑡𝑖 ) for later multicast retransmission. The receivers then detects gaps in the sequence numbers of the received packets for loss detection [5]. After detecting loss, associated receivers will set a timer and request retransmission to the OLT by sending NACKs with the sequence number of the lost packet. If the timer expires without receiving the retransmitted packet, the receivers reset the timer and resend NACKs to the OLT [5]. When an associated ONU receives the NACKs, NACK handling is done by the ONU to save the interface number to the receiver and the sequence number of the lost packet in the retransmission list of the ONU which relays the NACKs to the OLT. After receiving the first NACK of the lost packet, the OLT conducts broadcasting retransmission to broadcast the retransmitted packet in its cache once to all ONUs through a P2MP connection. It does so by ignoring other NACKs from other associated ONUs during a preset time, 𝑡𝑡𝑖 , which represents the Round Trip Time (RTT) of the i𝑡ℎ multicast packet from the OLT to a receiver; it is given 𝑡𝑡𝑖 = 𝑟𝑡𝑖 − 𝑐𝑡𝑖 , where 𝑟𝑡𝑖 is the receiving time of the first NACK for the i𝑡ℎ multicast packet. The unrelated ONUs then drop the
pkti
pkti NACK transmission :
Packet dropped without retransmission
Retransmission :
Transmission error :
Operation of CSMRP.
retransmitted multicast packets without retransmission if they do not relay any NACK. Based upon the broadcasting retransmission, it is possible to prevent duplicated retransmissions in the P2MP connection when using this policy. Moreover, for P2P connections, the associated ONUs do selective unicasting transmission to retransmit the multicast packets selectively to the associated receivers with their retransmission lists; thus, unnecessary retransmission can be also prevented in P2P connections. To support QoS of multicast applications, each receiver requests retransmission continously until its packet loss ratio is less than the loss requirement of multicast applications. For example, the required packet loss ratio of IPTV is 1.0E-7 [7]. For IPTV QoS, receivers continuously requests retransmission until the k𝑡ℎ retransmission request probability (𝑃𝑅 (𝑘)) is less than 1.0E-7. Because 𝑃𝑅 (𝑘) is affected by the loss of NACKs and retransmitted packets, 𝑃𝑅 (𝑘) is given by: 𝑃𝑅 (𝑘) = 𝑃𝐿𝑘 ⋅ (2 − 𝑃𝐿 )𝑘−1 ≤ 10−7 ,
(1)
where 𝑃𝐿 is the packet loss proability of IPTV packet during transmission. The minimum number of transmissions per multicast packet to guarantee IPTV QoS (𝑁𝑚 ) is given by: ⌈ ⌉ 𝑙𝑜𝑔10 (2 − 𝑃𝐿 ) − 7 𝑁𝑚 = . (2) 𝑙𝑜𝑔10 (2𝑃𝐿 − 𝑃𝐿2 ) For IPTV QoS, the multicast packets are retransmitted as 𝑁𝑚 -1 except for the first transmission. Thus, in the proposed policy, multicast packets are cached in the OLT during (𝑁𝑚 − 1) ⋅ 𝑅𝑇 𝑇𝑅𝐸𝑇 , where 𝑅𝑇 𝑇𝑅𝐸𝑇 is the RTT for retransmission. By doing so, the characteristics of both connections can be put into consideration at the same time. Consequently, the bandwidth consumed by multicast applications can be efficiently reduced and the queuing delay at the OLT also can be effectively diminished. III. P ERFORMANCE E VALUATION This section discusses the performance of CSMRP compared to the legacy policies regarding the bandwidth consumption and the average queuing delay. The results were obtained using an OPNET simulator with the following assumptions. (i)
500
400
300
0.022
URP (anal.) URP (simul.) SMRP (simul.) MRP (simul.) CSMRP (simul.)
Average queuing delay (ms)
Total bandwidth consumed (Mbps)
KWON et al.: CENTRALIZED SELECTIVE MULTICAST RETRANSMISSION POLICY FOR ENHANCED RESOURCE EFFICIENCY IN EPON-BASED ACCESS . . . 1015
200
100
0 5
10
15
20
25
30
35
40
45
Number of associated receivers per ONU (r)
The volume of the first transmission traffic (𝑉𝐹 ) is 100Mbps, as ten High Definition (HD) IPTVs and the size of each multicast packet for IPTV is 10528 bits [8]. (ii) The arrival rate of the packets complies with Poisson process. (iii) The bandwidth of an EPON is 1Gbps and that of each P2P connection is 50Mbps. (iv) There are 16 ONUs, and each IPTV is serviced to an identical number of associated receivers per ONU. (v) The packet loss probability (𝑃𝐿 ), which is caused by only transmission error due to physical interference in each P2P connection, is identical in each receiver (=0.005). (vi) Each receiver has a play-out buffer large enough to request retransmission continuously until supporting IPTV QoS. In the case of URP, the additinal bandwidth is dependent on the volume of retransmitted packets. In EPON-based access networks, URP requires two types of additional bandwidth for retransmission: (i) the additional bandwidth in the P2MP connection (𝐵𝑃 2𝑀𝑃 ), multiplied by the number of ONUs (𝑛) and the number of associated receivers per ONU (𝑟); and (ii) the additional bandwidth in the P2P connection (𝐵𝑃 2𝑃 ), as determined by 𝑃𝑅 (𝑘). Thus, 𝐵𝑃 2𝑀𝑃 and 𝐵𝑃 2𝑃 are obtained from:
𝐵𝑃 2𝑀𝑃
= =
𝑉𝐹 {
𝑁∑ 𝑚 −1 𝑘=1
(1 − 𝑃𝐿 )𝑃𝑅 (𝑘)},
𝑟 ⋅ 𝑛 ⋅ 𝐵𝑃 2𝑃 .
0.018
(3) (4)
Fig. 3 presents the total bandwidth consumed by retransmitted traffic in EPON-based access networks. Essentially, URP and MRP require larger bandwidth than CSMRP due to the duplicated retransmissions in the P2MP connection or the unnecessary retransmissions in the P2P connections. Although SMRP was proposed to solve the problems of URP and MRP, SMRP requires larger bandwidth than MRP when the number of associated receivers per ONU exceeds 20, as the hop-by-hop P2P retransmissions induce duplicated retransmissions in the shared P2MP connection. As a result, the legacy policies are not optimized to support EPON-based access networks that include both P2MP and P2P connections. On the other hand, CSMRP achieves the lowest level of bandwidth consumption due to the absence of both the duplicated retransmissions in the P2MP connection of an EPON and the unnecessary retransmissions in the last-mile P2P connections. As the number of associated receivers per ONU increases, CSMRP can reduce more and more retransmitted traffic. Even though CSMRP requires additional operations (e.g, NACK handling, broadcasting retransmission, and selective unicasting retransmission) and some memory for additional information
Fig. 4.
URP (simul.) SMRP (simul.) MRP (simul.) CSMRP (simul.)
0.016 0.014 0.012 0.01 5
50
Fig. 3. Total bandwidth consumed by retransmitted traffic in EPON-based access networks.
𝐵𝑃 2𝑃
0.02
10
15
20
25
30
35
40
45
Number of associated receivers per ONU (r)
50
The average queuing delay at the OLT.
(e.g, time value and retransmission list), the processing load of CSMRP is less than those of URP, MRP, and SMRP because the sizes of their retransmitted packets exceed thousands of bytes while the size of additional information for CSMRP is just dozens of bytes. As another effect of CSMRP, the average queuing delay at the OLT can be kept at the lowest level, as shown in Fig. 4. For both URP and SMRP, much longer average queuing delays are presented due to the increase of retransmitted traffic from the duplicated retransmissions in the P2MP connection by these policies. In contrast, MRP and CSMRP can guarantee shorter average queuing delay because these policies minimize the bandwidth consumed by retransmitted traffic due to the absence of duplicated retransmissions in an EPON. IV. C ONCLUSION This letter proposed CSMRP that is specially devised for the characteristics of EPON-based access networks. CSMRP can reduce the bandwidth consumption of retransmissions and can maintain the shortest average queuing delay at the OLT for multicast applications compared to the legacy policies. The effectiveness of CSMRP can be achieved by avoiding duplicated retransmissions in an EPON and unnecessary retransmissions in last-mile connections. As a consequence, CSMRP makes EPON-based access networks more scalable and stable to support multicast applications including IPTV, which is one of the killer applications of next generation internet. R EFERENCES [1] M. Hajduczenia, H. J. A. Silva, and P. P. MonterioCoudreuse, “Development of 10Gb/s EPON in IEEE 802.3av,” IEEE Commun. Mag., vol. 46, no. 7, pp.40–47, July 2008. [2] K. Tanaka, A. Agata, and Y. Horiuchi, “IEEE 802.3av 10G-EPON standardization and its research and development status,” IEEE J. Lightwave Technol., vol. 28, no. 4, pp. 651–661, Feb. 2009. [3] M. G. Luby and J. W. Miller, “The impact of packet loss on an IPTV network,” white paper, Digital Fountain, Jan. 2007. [4] M. Chatterjee, S. Sengupta, and S. Ganguly, “Feedback-based real-time streaming over WiMax,” IEEE Wireless Commun., vol. 14, no. 1, pp. 64–71, Feb. 2007. [5] W. Yoon, D. Lee, and H. Y. Youn, “Comparison of tree-based reliable multicast protocols for many-to-many sessions,” IEE Proc. Commun., vol. 152, no. 6, pp. 923–931, Dec. 2005. [6] O. Holland and A. H. Aghvami, “Constrained multicast retransmission forwarding under bitmapped feedback packets,” Electron. Lett., vol. 42, no. 7, Mar. 2006. [7] J. Asghar, I. Hood, and F. L. Faucheur, “Preserving video quality in IPTV networks,” IEEE Trans. Broadcast., vol. 55, no. 2, pp. 386–395, June 2009. [8] ITU-T Std. Recommendation G.1080, “Quality of experience requirements for IPTV services,” Dec. 2008.
