Professional Documents
Culture Documents
5, MAY 2011
Abstract—Building on the Ethernet Passive Optical Network (NG-PONs). NG-PONs are mainly envisioned to (𝑖) achieve
(EPON) and Gigabit PON (GPON) standards, Next-Generation higher performance parameters, e.g., higher bandwidth per
(NG) PONs (𝑖) provide increased data rates, split ratios, wave- subscriber, increased split ratio, and extended maximum reach,
lengths counts, and fiber lengths, as well as (𝑖𝑖) allow for
all-optical integration of access and metro networks. In this than current EPON/GPON architectures [2], and (𝑖𝑖) broaden
paper we provide a comprehensive probabilistic analysis of EPON/GPON functionalities to include, for instance, the con-
the capacity (maximum mean packet throughput) and packet solidation of optical access, metro, and backhaul networks as
delay of subnetworks that can be used to form NG-PONs. Our well as the support of topologies other than conventional tree
analysis can cover a wide range of NG-PONs through taking the structures [3].
minimum capacity of the subnetworks forming the NG-PON and
weighing the packet delays of the subnetworks. Our numerical In this paper, we evaluate the capacity (maximum mean
and simulation results indicate that our analysis quite accurately packet throughput) and packet delay of a wide range of NG-
characterizes the throughput-delay performance of EPON/GPON PONs through probabilistic analysis and verifying simulations.
tree networks, including networks upgraded with higher data More specifically, we analyze the capacity and delay of
rates and wavelength counts. Our analysis also characterizes the various subnetworks from which NG-PONs can be formed,
trade-offs and bottlenecks when integrating EPON/GPON tree
networks across a metro area with a ring, a Passive Star Coupler thus enabling analytical capacity and delay characterization
(PSC), or an Arrayed Waveguide Grating (AWG) for uniform and for a wide range of NG-PONs built from the examined
non-uniform traffic. To the best of our knowledge, the presented subnetworks. Two important applications for our analysis are:
analysis is the first to consider multiple PONs interconnected via (A) The obtained results provide insight into the performance
a metro network. limitations of candidate NG-PON architectures and thus in-
Index Terms—Metro area network, packet delay, passive form network operators seeking to upgrade their installed
optical network, throughput-delay analysis. TDM PONs. (B) Neither IEEE 802.3ah EPON nor ITU-
T G.984 GPON standardizes a specific dynamic bandwidth
I. I NTRODUCTION allocation (DBA) algorithm. The design of DBA algorithms
T HE Passive Optical Network (PON) is one of the most is left to manufacturers which aim at equipping network
widely deployed access networks due to its unique operators with programmable DBA algorithms that adapt to
benefits, including transparency against data rate and signal new applications and business models and thus make PONs
format as well as high data rates and reliability [1]. The two future-proof. Our capacity and delay analysis provides an
major state-of-the-art PON standards IEEE 802.3ah Ether- upper throughput bound and a delay benchmark for polling-
net PON (EPON) and ITU-T G.984 Gigabit PON (GPON) based medium access control with gated service [4] which
consist both of a single upstream wavelength channel and can be used to evaluate the throughput-delay performance of
a separate single downstream wavelength channel, whereby current and future DBA algorithms for NG-PONs.
both channels are operated with time division multiplexing This paper is structured as follows. In the following section,
(TDM). EPON and GPON are expected to coexist for the we review related work on the analysis of PON access and
foreseeable future as they evolve into Next-Generation PONs metro packet networks. In Section III we give overviews
of the EPON and GPON access networks. In Section IV,
Paper approved by J. A. Salehi, the Editor for Optical CDMA of the IEEE we present NG-PONs that either (𝑖) upgrade PONs or (𝑖𝑖)
Communications Society. Manuscript received July 19, 2010; revised October
15, 2010. interconnect multiple PONs across a metropolitan area. We
F. Aurzada and M. Scheutzow are with the Department of Mathemat- conduct the capacity and delay analysis of the subnetworks
ics, Technical University Berlin, 10623 Berlin, Germany (e-mail: {aurzada, forming NG-PONs in Section V. In Section VI, we compare
ms}@math.tu-berlin.de).
M. Reisslein is with the School of Electrical, Computer, and Energy numerical throughput-delay results obtained from our analysis
Engineering, Arizona State University, Tempe, Arizona 85287-5706, USA with simulations and illustrate the application of our capacity
(e-mail: reisslein@asu.edu). analysis to identify bottlenecks in NG-PONs. We briefly
N. Ghazisaidi was with the Optical Zeitgeist Laboratory, INRS, University
of Québec. He is now with Ericsson Inc., 200 Holger Way, CA 95134, San summarize our contributions in Section VII.
Jose, USA (e-mail: navid.ghazisaidi@ericsson.com).
M. Maier is with the Optical Zeitgeist Laboratory, INRS, University of
Québec, Montréal, QC, H5A 1K6, Canada (e-mail: maier@emt.inrs.ca). II. R ELATED W ORK
Supported by the DFG Research Center M ATHEON “Mathematics for key
technologies” in Berlin. In this section we briefly review related work on the analysis
Digital Object Identifier 10.1109/TCOMM.2011.030411.100418 of passive optical networks and metropolitan area networks.
0090-6778/11$25.00 ⃝
c 2011 IEEE
AURZADA et al.: CAPACITY AND DELAY ANALYSIS OF NEXT-GENERATION PASSIVE OPTICAL NETWORKS (NG-PONS) 1379
EPONs employ medium access control with an underlying on the bandwidth demands, which are signalled with the
polling structure [5]. Building directly on the literature on multipoint control protocol (MPCP). The ONU uses the MPCP
polling systems, e.g., [6], Park et al. [7] analyze an EPON REPORT message to report bandwidth requirements to the
model with random independent switchover times. This EPON OLT. The OLT passes received REPORT messages to its
model holds only when successive upstream transmissions are DBA algorithm module to calculate the upstream transmission
separated by a random time interval sufficiently large to “de- schedule [5]. Then, the OLT issues upstream transmission
correlate” successive transmissions, which would significantly grants by transmitting a GATE message specifying the start
reduce bandwidth utilization in practice. In an EPON, the time and length of the transmission window to each ONU. The
service (upstream transmission) of an Optical Network Unit transmission window may comprise multiple Ethernet frames,
(ONU) follows immediately (separated by a guard time) after whereby EPON does not allow for fragmentation. EPON
the upstream transmission of the preceding ONU to ensure carries Ethernet frames natively, i.e., without encapsulation.
high utilization. The switchover time is therefore generally The GPON offers several combinations of
highly dependent on the round-trip delays and the masking upstream/downstream data rates with a preferred rate of
of the round-trip delays through the interleaving of upstream 2.488 Gb/s downstream and 1.244 Gb/s upstream for up to
transmissions [4]. Subsequent analyses have strived to model 60 km reach and a maximum split ratio of 1:128 [19], [21].
these correlated transmissions and switchovers with increasing Both upstream and downstream transmissions are based on
fidelity [8]–[11]. a periodically recurring time structure with a fixed frame
GPONs have received relatively less research interest than length of 𝛿 = 125 𝜇s. Each upstream frame contains dynamic
EPONs. To the best of our knowledge we conduct the first bandwidth report (DBRu) fields. Each downstream frame
delay analysis of GPONs in this paper. Similarly, WDM PONs contains a physical control block (PCBd), which includes a
have received relatively little research attention to date. The bandwidth map (BWmap) field specifying the ONU upstream
call-level performance of a WDM PON employing Optical transmission grants. Unlike the EPON, the GPON deploys
Code Division Multiple Access (OCDMA) [12] was analyzed the GPON encapsulation method (GEM) involving a 5-byte
in [13]. To the best of our knowledge packet-level analyses GEM header and allows for Ethernet frame fragmentation.
of WDM EPONs has so far only been attempted in [14, Two DBA methods are defined for GPON: (𝑖) status-reporting
Section 2.4] where an WDM EPON with offline scheduling DBA based on ONU reports via the DBRu field, and (𝑖𝑖)
(also known as interleaved polling with stop [5]) was analyzed non-status-reporting DBA based on traffic monitoring at the
and in [15], [16] where the capacity of WDM PONs was OLT. Recent GPON research has focused on the design and
examined. In contrast, we analyze in this paper the capacity evaluation of status-reporting DBA algorithms [22]. The
and packet delay for WDM PON channels operating in online- impact of the various types of guard times and overhead
scheduling PONs that are integrated with metro networks. fields on the bandwidth efficiency of both EPON and GPON
Metropolitan area networks have received significant at- has been thoroughly investigated in [23], [24].
tention over the past two decades. The capacity and de-
lay performance of packet ring networks with a variety of
MAC protocols has been analyzed in numerous studies, see IV. NG-PON S
e.g., [17]. The impact of fiber shortcuts in ring networks has
Generally, NG-PON technologies fall into two cate-
been analyzed in [18]. In this paper, we analyze to the best
gories [25]: (𝑖) Evolutionary NG-PON technologies provide
of our knowledge for the first time a comprehensive NG-
improved performance, such as higher data rates, with legacy
PON that interconnects multiple PONs via a metro network
optical distribution networks and co-exist with legacy PONs.
combining a ring and star networks.
(𝑖𝑖) Revolutionary (disruptive) NG-PON technologies, such as
optical code division multiplexing (OCDM), provide enhanced
III. OVERVIEW OF EPON AND GPON services on new distribution networks. We focus on evolution-
ary NG-PON technologies that are expected to replace current
Typically, both EPON and GPON have a physical tree
state-of-the-art GPON and EPON solutions in the near- to mid-
topology with the OLT at the root [19]. The OLT connects
term.
through an optical splitter to multiple ONUs, also known as
Optical Network Terminals (ONTs). Each ONU can serve a
single or multiple subscribers. To facilitate DBA, both EPON
A. High-speed TDM PON
and GPON use polling based on a report/grant mechanism.
In each polling cycle, ONUs send their instantaneous up- Higher speeds are needed to support emerging bandwidth-
stream bandwidth demands through report messages to the hungry applications, e.g., high-definition television and video
OLT, which in turn dynamically allocates variable upstream on demand, and to provide sufficient capacity as backhauls of
transmission windows by sending a separate grant message to next-generation IEEE 802.11n wireless LANs with a through-
each ONU. put of 100 Mb/s or higher per device [2]. For both EPON
The EPON is a symmetric network providing a data rate and GPON, standardization efforts have begun to specify
of 1 Gb/s in both upstream and downstream directions. It symmetric or asymmetric data rates of up to 10 Gb/s [26].
provides a reach between OLT and ONUs of up to 20 km for a Recently, the IEEE standard 802.3av for 10 Gb/s EPON was
split ratio initially set to 1:16 [19], [20] and recently reaching approved in September 2009. DBA algorithms for EPON,
1:64 [21]. The EPON has variable-length polling cycles based GPON, and high-speed TDM PON are compared in [24].
1380 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 59, NO. 5, MAY 2011
up
λ TDM TX TDM Data &
Control
TDM ONU
DEMUX
down RX TDM Data &
λ TDM Control
Data RX WDM λ W DEMUX WDM ONU
Λ OLT λ1
Data RX WDM TDM ONU λ 1...W
Λ OLT BPF RSOA Data
Data & up up
RX TDM λ TDM λ TDM
Control WDM WDM ONU TX TDM Data &
Coupler Coupler Control
DEMUX
Circulator
down Data &
λ TDM RX TDM
Control
Data TX LR λ W+L
Λ AWG
Λ OLT &
Λ AWG Λ AWG
Data TX WDM λ W
Fig. 1. Evolutionary upgrade of legacy TDM ONUs to WDM ONUs and long-reach ONUs (LR ONUs) and their coexistence on the same NG-PON fiber
infrastructure.
3) Capacity of the AWG: In the considered NG-PON streams that flow into OLT 𝑘 from adjacent nodes and leave
setting, the potential for collisions exists in the AWG star over the arc of interest, namely the downstream TDM channel.
subnetwork only at the AWG input ports (local scheduling by In particular,
the OLT avoids actual collisions). For the capacity analysis, it ∑ ∑
suffices therefore to focus on the wavelength channels running 𝐵 𝑇,𝑑,𝑘 = Φ(𝜌𝑙,𝑅,𝑇,𝑘 )+ Φ(𝜌𝑙,𝑃,𝑇,𝑘 ),
from the LR ONUs at a given OLT to the corresponding AWG ring node 𝑙 adjacent to OLT 𝑘 OLT 𝑙
(14)
input port. The traffic generated by the LR ONUs of OLT 𝑘
with 𝜌𝑙,𝑅,𝑇,𝑘 denoting the load due to traffic flowing from the
destined toward LR ONUs at OLT 𝑙 must be accommodated
on the 𝑐(𝑘, 𝑙), 𝑐(𝑘, 𝑙) ≥ 0, wavelength channels that are routed adjacent ring nodes 𝑙 over the ring to reach one of the TDM
ONUs at OLT 𝑘 and 𝜌𝑙,𝑃,𝑇,𝑘 denoting the load from an OLT
from the AWG input port of OLT 𝑘 to the AWG output port
𝑙, 𝑙 ∕= 𝑘, over the PSC to a TDM ONU at OLT 𝑘. Adding
leading to OLT 𝑙. The total average rate of traffic 𝜆𝐴 (𝑘, 𝑙) ¯ 𝑇 and the downstream
the average transmission delay 𝐿/𝐶
transmitted from OLT 𝑘 over the AWG to OLT 𝑙 must satisfy:
∑ propagation delay 𝜏𝑇 we obtain the total delay
𝜆𝐴 (𝑘, 𝑙) := ¯ ⋅ 𝑇 𝐴 (𝑖, 𝑗) < 𝐶𝐴 ⋅ 𝑐(𝑘, 𝑙).
𝐿 (11)
( ) 𝐿¯
𝑖∈𝒞𝑘𝐿 , 𝑗∈𝒞𝑙𝐿 𝐷𝑇,𝑑,𝑘,𝐸 = Φ 𝜌𝑇,𝑑,𝑘 + 𝜏𝑇 + − 𝐵 𝑇,𝑑,𝑘 . (15)
𝐶𝑇
4) Summary of Capacity Analysis: We summarize the ca-
pacity analysis by noting that for given traffic patterns 𝑇 (𝑖, 𝑙) The upstream data transmissions experience the following
and 𝑇 𝐴 (𝑖, 𝑙) it is relatively straightforward to obtain from the delay components due to the reporting and granting procedure.
capacity constraints in Sections V-C1–V-C3 bounds on the First, the mean delay from packet generation until transmission
mean aggregate network throughput. In particular, we denote of the corresponding report is the mean residual cycle length
∑[ ] of the upstream TDM channel, which is approximately 𝜏𝑇 .
𝑟𝑇 = 𝐿 ¯ 𝜎(𝑖) + 𝜎 𝐴 (𝑖) (12) Second, the upstream propagation of the report and down-
𝑖∈𝒩 stream propagation of the grant add the round trip propagation
as the total generated traffic [in bit/second], which is equiv- delay 2𝜏𝑇 . Third, the queueing delay for the grant prior to its
alent to the mean aggregate throughput of the network. Each transmission on the downstream TDM channel is Φ(𝜌𝑇,𝑑,𝑘 )
capacity constraint results in an upper bound on 𝑟𝑇 . The ¯
without priority for grants, and 𝜌𝑇,𝑑,𝑘 𝐿/(2𝐶 𝑇 ) with priority
tightest bound identifies the network bottleneck limiting the for the grant messages (considered in the following). Adding
mean aggregate throughput. packet queueing, transmission, and propagation delays gives
𝐿¯ ¯
𝜌𝑇,𝑑,𝑘 𝐿
D. Delay Analysis 𝐷𝑇,𝑢,𝑘,𝐸 = 4𝜏𝑇 + Φ(𝜌𝑇,𝑢,𝑘 ) + + . (16)
𝐶𝑇 2𝐶𝑇
We denote 𝜏𝑇 , 𝜏𝑃 , and 𝜏𝐴 [in seconds] for the one-way
2) Delay on the EPON WDM Channels:
propagation delay over the tree network, as well as the PSC
a) Remodulated Carrier for Upstream Transmission:
and AWG star subnetworks, respectively. We require that the
The WDM channels can be continuously used for down-
traffic that is generated at node 𝑖 and destined to node 𝑗 is
stream and upstream transmission. The traffic rate for the
Poisson with packet generation rate 𝑇 (𝑖, 𝑗) [packets/second]
downstream WDM channels is given by 𝜆𝑊,𝑑,𝑘 (8). The
and independent of the traffic for all other combinations 𝑖′ , 𝑗 ′ .
queueing delay can be approximated by the queueing delay
For each traffic rate 𝜆 [bit/second] of Section V-C, we define a
in an M/G/𝑊 queueing system. Since there is no explicit
corresponding (relative) load 𝜌 by dividing 𝜆 by the respective
delay formula for such a system, we further approximate by
transmission capacity 𝐶 [bit/second]. We denote
( 2 ) considering an M/G/1 queue with a server with transmission
𝜎
𝜌 𝐿¯𝐿 + 𝐿¯ rate 𝑊 𝐶𝑊 , i.e., we consider an M/G/1 queue with load
Φ(𝜌) := (13) 𝜌𝑊,𝑑,𝑘 = 𝜆𝑊,𝑑,𝑘 /(𝑊 𝐶𝑊 ),
2𝐶(1 − 𝜌)
( 𝑊,𝑑,𝑘 ) ¯
for the corresponding mean queuing delay in an M/G/1 𝜆 𝐿
𝐷𝑊,𝑑,𝑘,𝐸 ≈ Φ + 𝜏𝑇 + − 𝐵 𝑊,𝑑,𝑘 , (17)
queue [35]. 𝑊 𝐶𝑊 𝐶𝑇
1) Delay on the Up/Downstream EPON TDM channels:
where 𝐵 𝑊,𝑑,𝑘 is defined analogously to (14). For the upstream
The long run average traffic rate on the downstream TDM
direction
( 𝑊,𝑢,𝑘we
) obtain analogously to (16): 𝐷𝑊,𝑢,𝑘,𝐸 ≈ 4𝜏𝑇 +
channel from OLT 𝑘 is 𝜆𝑇,𝑑,𝑘 given in (7), resulting in a load 𝜆 ¯
𝐿 ¯
𝜌𝑇 ,𝑑,𝑘 𝐿
𝜌𝑇,𝑑,𝑘 = 𝜆𝑇,𝑑,𝑘 /𝐶𝑇 . Thus, an initial estimate of the queueing Φ 𝑊 𝐶𝑊 + 𝐶𝑊 + 2𝐶𝑊 .
delay of a packet prior to transmission on the downstream b) Empty Carrier for Upstream Data Transmission:
TDM channel is given by Φ(𝜌𝑇,𝑑,𝑘 ). This delay does not With switching between upstream and downstream transmis-
consider that this traffic may already have traversed adjacent sions, the combined upstream and downstream traffic has to
ring nodes and/or other OLTs 𝑙, 𝑙 ∕= 𝑘, (via the PSC) if be accommodated on the WDM channels, resulting in the load
𝑊,𝑑,𝑘 𝑊,𝑢,𝑘
the considered EPON is part of a metro network (Fig. 2). 𝜌𝑊,𝑘 = 𝜆 𝑊+𝜆 𝐶𝑊 and a queueing delay of approximately
Applying the method of Bux and Schlatter [36] to our setting, Φ(𝜌𝑊,𝑘 ). As detailed in [37] the delay contribution due to
we compensate for the queueing delay at the preceding nodes switchovers can be approximated through a superposition of
by subtracting a correction term 𝐵 𝑇,𝑑,𝑘 from the queueing Poisson processes and incorporated as a shift in the packet
delay Φ(𝜌𝑇,𝑑,𝑘 ). Following [36], the correction term 𝐵 𝑇,𝑑,𝑘 length distribution to mean 𝐿+𝑝 ¯ 𝑠,𝑘 𝐶𝑊 𝜏𝑇 and second moment
is the sum of the queueing delays for the individual traffic 2 ¯
𝜎𝐿 +2𝐿𝑝𝑠,𝑘 𝐶𝑊 𝜏𝑇 +𝑝𝑠,𝑘 (𝐶𝑊 𝜏𝑇 )2 . Using this modified packet
1384 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 59, NO. 5, MAY 2011
length distribution, we obtain the delay on the WDM channels 4) Delay in the Ring/PSC Star Subnetwork: With 𝜏𝑃𝑓 de-
as noting the frame duration [in seconds] on the PSC, a newly
¯ arrived packet at the OLT waits on average 𝜏𝑃𝑓 /2 before its
𝐿
𝐷𝑊,𝑑,𝑘 = Φ(𝜌𝑊,𝑘 ) + 𝜏𝑇 + − 𝐵 𝑊,𝑑,𝑘 . (18) control packet can be sent. The control packet experiences a
𝐶𝑇 propagation delay of 𝜏𝑃 . Once the control packet is received,
The upstream traffic experiences additional delay components the packet enters the virtual queue for the destination OLT.
due the reporting and granting procedure analogous to the This queue experiences a load of 𝜌𝑃,𝑙 = 𝜆𝑃 (𝑙)/𝐶𝑃 with
analysis leading to (16). Thus, 𝐷𝑊,𝑢,𝑘 := 𝐷𝑊,𝑑,𝑘 + 3𝜏𝑇 + 𝜆𝑃 (𝑙) given in (10). Adding in the data packet transmission
¯
𝜌𝑇 ,𝑑,𝑘 𝐿 and propagation delays, we obtain
2𝐶𝑇 .
3) Delay on GPON: Let 𝛿 denote the frame duration of 125 𝜏𝑃𝑓 𝐿¯
𝜇s of the GPON. A packet generated at an ONU has to wait on 𝐷𝑃 (𝑙) = + 2𝜏𝑃 + Φ(𝜌𝑃,𝑙 ) + − 𝐵 𝑃,𝑙 , (21)
2 𝐶𝑃
average 𝛿/2 for the beginning of the next frame in which it will
be reported in a DBRu field. This next frame has a duration with
∑ the
[∑ correction term 𝐵 𝑃,𝑙 ] =
𝑅,𝑃,𝑚,𝑘,𝑙 𝑇,𝑃,𝑘,𝑙
(transmission delay) of 𝛿 and takes 𝜏𝑇 to propagate to the OLT. OLT 𝑘 𝑚 Φ(𝜌 ) + Φ(𝜌 ) + Φ(𝜌𝑊,𝑃,𝑘,𝑙 )
Once arrived at the OLT, the bandwidth report is processed whereby the inner sum is over the ring nodes 𝑚 adjacent to
and the grant for the packet’s transmission is included in the OLT 𝑘; further, 𝜌𝑅,𝑃,𝑚,𝑘,𝑙 denotes the traffic that originates
BWmap of the next downstream frame. Even with negligible at ring node 𝑚 and flows over the PSC from OLT 𝑘 to OLT
processing time at the OLT, there is a delay of up to 𝛿 until the 𝑙, and 𝜌𝑇,𝑃,𝑘,𝑙 and 𝜌𝑊,𝑃,𝑘,𝑙 denote the analogous traffic loads
beginning of the next downstream frame. More specifically, for the TDM and WDM channels (see [37] for details).
𝑅,𝑃
let 𝜔, 0 ≤ 𝜔 < 𝛿, [in seconds] denote the offset between The packet delay in the ring/PSC star subnetwork 𝐷𝑖𝑗
the upstream and downstream channels defined as follows. At from an OLT/CO 𝑖 to another OLT/CO (or a destination ring
the instant when a new slot starts on the upstream channel, 𝜔 node) 𝑗 is given by Eqn.
∑ (22) in [34]𝑃with the last two sums
𝑃
seconds have passed of the current downstream channel slot, replaced by 𝐷𝑖𝑗 = 𝑘,𝑙 𝑝𝑖𝑗 (𝑘, 𝑙)𝐷 (𝑙). The packet delay
i.e., for 𝜔 = 0, the slots on the upstream and downstream from a ring node to a CO/hotspot (or a destination ring node)
channels are aligned. Then, the time until the beginning of the is given by Eqn. (21) in [34] with the last sum replaced by
𝑃
downstream frame 𝐷𝑖𝑗 .
( containing
( ⌊the BWmap
⌋)) of the considered 5) Delay in AWG Star Subnetwork: The packet delay in
packet is 𝛾1 = 1 − 𝜏𝑇 𝛿−𝜔 − 𝜏𝑇 𝛿−𝜔 𝛿.
the AWG star subnetwork is approximately
The downstream frame has a transmission delay of 𝛿
¯
𝜌𝑇,𝑑,𝑘 𝐿 ¯
𝐿
and propagation
( ( ⌊ of 𝜏⌋))
delay 𝑇 . The packet has to wait for
𝐷𝐴 (𝑘, 𝑙) = 3𝜏𝑇 + + Φ(𝜌𝐴 (𝑘, 𝑙)) + 𝜏𝐴 + , (22)
𝛾2 = 1 − 𝜏𝑇 𝛿+𝜔 − 𝜏𝑇 𝛿+𝜔 𝛿 until the beginning of the 2𝐶𝑇 𝐶𝐴
next upstream frame before it can possibly be transmitted. where the first two terms are due to the report-grant cycle us-
Thus, it takes overall on average 5𝛿/2+𝜏𝑇 +𝛾1 +𝜏𝑇 +𝛾2 from ing the upstream TDM channel. Averaging, using the 𝜆𝐴 (𝑘, 𝑙)
the instant the packet is generated to the instant the packet (11) as weights, gives the average packet delay 𝐷𝐴 on the
becomes eligible for upstream transmission. And then, the AWG star subnetwork:
packet is put into a general queue for the upstream channel. In
∑ 𝜆𝐴 (𝑘, 𝑙)
terms of the mean packet delay, this channel can be modeled as 𝐷𝐴 = 𝐷𝐴 (𝑘, 𝑙) ⋅ ∑ . (23)
𝐴 ′ ′
an M/G/1 queue (noting that the specific scheduling discipline OLT 𝑘, 𝑙 OLT 𝑘′ , 𝑙′ 𝜆 (𝑘 , 𝑙 )
does not affect the overall mean packet delay in the GPON, as
long as the channel is operated in work conserving manner, E. Overall Delay
i.e., is not left idle while packets are queued), with corre-
We obtain the overall average packet delay by weighing
sponding delay Φ(𝜌𝑇,𝑢,𝑘 ). Finally, the packet experiences the
¯ 𝑇 and propagation delay 𝜏𝑇 . Overall, the delays 𝐷(𝑖, 𝑗) of the different paths according to their
transmission delay 𝐿/𝐶
packet traffic rates 𝑇 (𝑖, 𝑗). First, for traffic transmitted over the
the mean delay for the TDM upstream channel is ∑
ring/PSC star subnetwork 𝐷𝑅,𝑃 = 𝑖,𝑗 𝐷(𝑖, 𝑗) ∑ ′ 𝑇′(𝑖,𝑗) 𝑇 (𝑖′ ,𝑗 ′ ) .
𝑖 ,𝑗
5𝛿 𝐿¯ For instance, 𝐷(𝑖, 𝑗) = 𝐷𝑖𝑙 𝑅,𝑃
+ 𝐷𝑇,𝑑,𝑙 for traffic from ring
𝑇,𝑢,𝑘,𝐺
𝐷 = + 𝛾1 + 𝛾2 + Φ(𝜌𝑇,𝑢,𝑘 ) + 3𝜏𝑇 + . (19)
2 𝐶𝑇 node/CO 𝑖 to TDM ONU 𝑗 at OLT 𝑙 (resp. 𝐷𝑊,𝑑,𝑙 for traffic
to WDM or LR ONU at OLT 𝑙). Overall, we obtain the mean
The WDM upstream channels experience the analogous
packet delay as
delays for the report-grant cycle but carry the load ∑
𝜆𝑊,𝑢,𝑘 /(𝑊 𝐶𝑊 ), i.e., we obtain 𝐷𝑊,𝑢,𝑘,𝐺 by replacing 𝑅,𝑃 ∑ 𝑖,𝑗 𝑇 (𝑖, 𝑗)
𝐷 = 𝐷 ∑
Φ(𝜌𝑇,𝑢,𝑘 ) by Φ(𝜆𝑊,𝑢,𝑘 /(𝑊 𝐶𝑊 )) in (19). 𝑖,𝑗 𝑇 (𝑖, 𝑗) + 𝐴
LR ONUs 𝑖,𝑗 𝑇 (𝑖, 𝑗)
The TDM downstream channel is analyzed analogously ∑ 𝐴
LR ONUs 𝑖,𝑗 𝑇 (𝑖, 𝑗)
giving +𝐷𝐴 ∑ ∑ 𝐴
.(24)
𝑖,𝑗 𝑇 (𝑖, 𝑗) + LR ONUs 𝑖,𝑗 𝑇 (𝑖, 𝑗)
𝛿 ( ) ¯
𝐿
𝐷𝑇,𝑑,𝑘,𝐺 = + Φ 𝜌𝑇,𝑑,𝑘 + 𝜏𝑇 + 𝑘 − 𝐵 𝑇,𝑑,𝑘 . (20) VI. N UMERICAL AND S IMULATION R ESULTS
2 𝐶𝑇
This section presents numerical results on the throughput-
The delay for the downstream WDM
( 𝑊,𝑑,𝑘 ) channels is obtained by delay performance, first for isolated PONs and then for
( )
replacing Φ 𝜌𝑇,𝑑,𝑘 by Φ 𝜆𝑊 𝐶𝑊 in (20). integrated access-metro networks, obtained from our analysis
AURZADA et al.: CAPACITY AND DELAY ANALYSIS OF NEXT-GENERATION PASSIVE OPTICAL NETWORKS (NG-PONS) 1385
1.2 1.2
TDM
Empty carrier
1 1 WDM w/ W=8
WDM w/ W=2
WDM w/ W=8 Remodulation
0.8 0.8
High−speed TDM WDM w/ W=8
@ 2.5 Gb/s
0.6 0.6
0.4 0.4
High−speed TDM High−speed TDM
@ 10 Gb/s @ 10 Gb/s
Fig. 3. Mean delay 𝐷 on upstream TDM/WDM channels of high-speed TDM Fig. 4. Mean delay 𝐷 vs. mean aggregate throughput 𝑟𝑇 of WDM
and WDM EPON/GPON vs. mean aggregate throughput 𝑟𝑇 for 𝑁𝑇 = 32 EPON/GPON using remodulation and empty carrier.
TDM ONUs and 𝑁𝑊 = 32 WDM ONUs, respectively.
1/(𝑁 − 1).
2
Remodulation WDM
Fig. 3 compares the mean delay 𝐷 on the upstream 1 w/ W=8 and ring&PSC
5
of the bidirectional ring is set to 100 km and it comprises α=1 (analysis)
𝑁𝑟 = 4 equally spaced ring nodes. Both ring and PSC operate 4.5 α=2 (analysis)
5
β=0.24 (analysis)
[14] W.-R. Chang, “Research and design of system architectures and com-
β=0.5 (analysis) munication protocols for next-generation optical networks," Ph.D. dis-
4.5
β=1 (analysis) sertation, National Cheng Kung University, Tainan, Taiwan, 2008.
β=0.24 (simulation) [15] N. Antunes, C. Fricker, P. Roberts, and J. Roberts, “Traffic capacity of
4 β=0.5 (simulation) large WDM passive optical networks," in Proc. Int. Teletraffic Congress,
β=1 (simulation) Sep. 2010.
3.5 [16] F. Aurzada, M. Scheutzow, M. Reisslein, and M. Maier, “Towards a
fundamental understanding of the stability and delay of offline WDM
3 EPONs," IEEE/OSA J. Optical Commun. Netw., vol. 2, no. 1, pp. 51-66,
Mean Delay (ms)
WDM EPON
w/ ring&AWG Jan. 2010.
2.5 [17] M. A. Marsan, E. Leonardi, M. Meo, and F. Neri, “Modeling slotted
WDM rings with discrete-time Markovian models," Comput. Netw.,
2 vol. 32, no. 5, pp. 599-615, May 2000.
[18] I. Rubin and H.-K. H. Hua, “Synthesis and throughput behavior
1.5 of WDM meshed-ring networks under nonuniform traffic loading,"
IEEE/OSA J. Lightwave Technol., vol. 15, no. 8, pp. 1513-1521, Aug.
1
1997.
High−speed TDM
[19] A. Girard, FTTx PON Technology and Testing. EXFO Electro-Optical
EPON w/ ring&AWG Eng. Inc., 2005.
0.5
[20] M. D. Vaughn, D. Kozichek, D. Meis, A. Boskovic, and R. Wanger,
“Value of reach-and-split ratio increase in FTTH access networks,"
0
0 10 20 30 40 50 60 IEEE/OSA J. Lightwave Technol., vol. 22, no. 11, pp. 2617-2622, Nov.
Mean Aggregate Throughput (Gb/s) 2004.
[21] F. Effenberger, D. Clearly, O. Haran, G. Kramer, R. D. Li, M. Oron,
Fig. 7. Mean delay 𝐷 vs. mean aggregate throughput 𝑟𝑇 of three and T. Pfeiffer, “An introduction to PON technologies," IEEE Commun.
WDM EPONs and three high-speed TDM EPONs interconnected through Mag., vol. 45, no. 3, pp. S17-S25, Mar. 2007.
a ring&AWG for 𝛼 = 2 and different destination traffic non-uniformity [22] J. Jiang and J. Senior, “A new efficient dynamic MAC protocol for the
𝛽 ∈ {0.24, 0.5, 1}. delivery of multiple services over GPON," Photonic Network Commun.,
vol. 18, no. 2, pp. 227-236, Oct. 2009.
[23] M. Hajduczenia, H. J. A. da Silva, and P. P. Monteiro, “EPON versus
APON and GPON: a detailed performance comparison," OSA J. Optical
Netw., vol. 5, no. 4, pp. 298-319, Apr. 2006.
use of our analysis to evaluate the throughput delay perfor- [24] B. Skubic, J. Chen, J. Ahmed, L. Wosinska, and B. Mukherjee, “A
mance of upgrades that increase the transmission line rates or comparison of dynamic bandwidth allocation for EPON, GPON, and
wavelength counts. We also demonstrate the identification of next-generation TDM PON," IEEE Commun. Mag., vol. 47, no. 3, pp.
S40-S48, Mar. 2009.
network bottlenecks using our analysis. [25] J.-I. Kani, F. Bourgart, A. Cui, A. Rafel, M. Campbell, R. Davey, and
S. Rodrigues, “Next-generation PON–part I: technology roadmap and
general requirements," IEEE Commun. Mag., vol. 47, no. 11, pp. 43-
R EFERENCES 49, Nov. 2009.
[26] F. J. Effenberger, “The XG-PON system: cost effective 10 Gb/s access,"
[1] M. Rad, H. A. Fathallah, and L. Rusch, “Fiber fault PON monitoring IEEE/OSA J. Lightwave Technol., vol. 29, no. 4, pp. 403-409, Feb. 2011.
using optical coding: effects of customer geographic distribution," IEEE [27] D. Shea and J. Mitchell, “Architecture to integrate multiple PONs with
Trans. Commun., vol. 58, no. 4, pp. 1172-1181, Apr. 2010. long reach DWDM backhaul," IEEE J. Sel. Areas Commun., vol. 27,
[2] R. Lin, “Next generation PON in emerging networks," in Proc. no. 2, pp. 126-133, Feb. 2009.
OFC/NFOEC, Feb. 2008. [28] G. Talli and P. D. Townsend, “Hybrid DWDM-TDM long-reach PON
[3] K. Grobe and J.-P. Elbers, “PON in adolescence: From TDMA to WDM- for next-generation optical access," IEEE/OSA J. Lightwave Technol.,
PON," IEEE Commun. Mag., vol. 46, no. 1, pp. 26-34, Jan. 2008. vol. 24, no. 7, pp. 2827-2834, July 2006.
[4] G. Kramer, B. Mukherjee, and G. Pesavento, “IPACT: a dynamic [29] M. Maier, “WDM passive optical networks and beyond: the road ahead,"
protocol for an Ethernet PON (EPON)," IEEE Commun. Mag., vol. 40, IEEE/OSA J. Optical Commun. Netw., vol. 1, no. 4, pp. 2617-2622, Sep.
no. 2, pp. 74-80, Feb. 2002. 2009.
[5] J. Zheng and H. Mouftah, “A survey of dynamic bandwidth allocation [30] L. Meng, C. Assi, M. Maier, and A. Dhaini, “Resource management
algorithms for Ethernet Passive Optical Networks," Optical Switching in STARGATE-based Ethernet passive optical networks (SG-EPONs),"
Netw., vol. 6, no. 3, pp. 151-162, July 2009. IEEE/OSA J. Optical Commun. Netw., vol. 1, no. 4, pp. 279-293, Sep.
[6] H. Takagi, Analysis of Polling Systems. MIT Press, 1986. 2009
[7] C. G. Park, D. H. Han, and K. W. Rim, “Packet delay analysis [31] M. Maier and M. Reisslein, “AWG based metro WDM networking,"
of symmetric gated polling system for DBA scheme in an EPON," IEEE Commun. Mag., vol. 42, no. 11, pp. S19-S26, Nov. 2004.
Telecommun. Syst., vol. 30, no. 1-3, pp. 13-34, Nov. 2005. [32] M. Scheutzow, M. Maier, M. Reisslein, and A. Wolisz, “Wavelength
[8] F. Aurzada, M. Scheutzow, M. Herzog, M. Maier, and M. Reisslein, reuse for efficient packet-switched transport in an AWG-based metro
“Delay analysis of Ethernet Passive Optical Networks with gated ser- WDM network," IEEE/OSA J. Lightwave Technol., vol. 21, no. 6, pp.
vice," OSA J. Optical Netw., vol. 7, no. 1, pp. 25-41, Jan. 2008. 1435-1455, June 2003.
[9] B. Lannoo, L. Verslegers, D. Colle, M. Pickavet, M. Gagnaire, and [33] M. Maier, M. Herzog, and M. Reisslein, “STARGATE: the next evolu-
P. Demeester, “Analytical model for the IPACT dynamic bandwidth tionary step toward unleashing the potential of WDM EPONs," IEEE
allocation algorithm in EPONs," OSA J. Optical Netw., vol. 6, no. 6, Commun. Mag., vol. 45, no. 5, pp. 50-56, May 2007.
pp. 677-688, June 2007. [34] M. Maier, M. Herzog, M. Scheutzow, and M. Reisslein, “PROTEC-
[10] M. T. Ngo, A. Gravey, and D. Bhadauria, “A mean value analysis TORATION: a fast and efficient multiple-failure recovery technique for
approach for evaluating the performance of EPON with Gated IPACT," resilient packet ring (RPR) using dark fiber," IEEE/OSA J. Lightwave
in Proc. Int. Conf. Optical Netw. Design Modeling, Mar. 2008, pp. 1-6. Technol., vol. 23, no. 10, pp. 2816-2838, Oct. 2005.
[11] J. Vardakas and M. Logothetis, “Packet delay analysis for priority- [35] L. Kleinrock, Queueing Systems: Volume I: Theory. Wiley, 1975.
based passive optical networks," in Proc. Int. Conf. Emerging Netw. [36] W. Bux and M. Schlatter, “An approximate method for the performance
Intelligence, Oct. 2009, pp. 103-107. analysis of buffer insertion rings," IEEE Trans. Commun., vol. COM-31,
[12] W. Kwong, G.-C. Yang, and J.-G. Zhang, “2𝑛 prime-sequence codes no. 1, pp. 50-55, Jan. 1983.
and coding architecture for optical code-division multiple-access," IEEE [37] F. Aurzada, M. Scheutzow, M. Reisslein, N. Ghazisaidi, and M. Maier,
Trans. Commun., vol. 44, no. 9, pp. 1152-1162, Sep. 1996. “Capacity and delay analysis of next-generation passive optical networks
[13] J. Vardakas, V. Vassilakis, and M. Logothetis, “Blocking analysis for (NG-PONs)–extended version," Technical Report, Ariz. State Univ., July
priority classes in hybrid WDM-OCDMA passive optical networks," in 2010. Available: http://mre.faculty.asu.edu/NG_PONs.pdf.
Proc. Fifth Advanced International Conf. Telecommun., 2009, pp. 389-
394.
1388 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 59, NO. 5, MAY 2011