Optical Burst Switching (OBS) { A New Paradigm for an Optical

Internet
Chunming Qiao1 Myungsik Yoo
Dept of CSE Dept of EE

Lab for Advanced Network Design, Evaluation and Research (LANDER)

University at Bualo
Bualo, New York 14260

ABSTRACT
To support bursty trac on the Internet (and especially WWW) eciently, optical burst switch-
ing (OBS) is proposed as a way to streamline both protocol and hardware in building the future
generation Optical Internet. By leveraging the attractive properties of optical communications and
at the same time, taking into account its limitations, OBS combines the best of optical circuit-
switching and packet/cell switching. In this paper, the general concept of OBS protocols and in
particular, those based on Just-Enough-Time (JET), is described, along with the applicability of
OBS protocols to IP over WDM. Specic issues such as the use of ber delay-lines (FDL) for
accommodating processing delay and/or resolving con icts are also discussed. In addition, the
performance of JET-based OBS protocols which use an oset time along with delayed reservation
to achieve ecient utilization of both bandwidth and FDLs as well as to support priority-based
routing is evaluated.

Keywords: bursty trac, ber delay line (FDL), IP, optical networks, reservation, WDM

1
Contacting information: phone:(716)-645-3180 ext. 140, fax: (716)-645-3464 and email: qiao@computer.org.
This research is sponsored in part by a grant from NSF under contract number ANIR-9801778
1 Introduction
The emergence of Terabit switches/routers, whose line speed has approached OC-48 (2.5 Gb/s) and
may soon reach OC-192 (10Gb/s), makes it natural to provide direct WDM interconnects between
these switches and routers, leading to current activities in building the so-called \Optical Internet"
and \IP over SONET (over WDM)". These networks may be regarded as the rst generation Optical
Internet where switching is still performed in the electronic domain. In this paper, we study some
issues related to using all-optical WDM networks as a layer directly beneath IP. Given the signicant
progress made and the continuing advances expected in the DWDM networking technology, which
provides new and strong incentives to building a exible, ecient and bandwidth-abundant ber-
optic network infrastructure capable of providing ubiquitous services, we believe that such future
(2nd) generation Optical Internet will be not only desirable as a way to support a higher degree of
data transparency for the benet of certain applications but also feasible in the near future.
During the past several years, static and slowly recongurable WDM networks have been a
focus, which is understandable given the constraints imposed by the devices and components, and
the basic need to provide lightpaths to an upper layer such as SONET. In order for a WDM optical
layer to provide dierentiated services in an eective and feasible way, as well as to circumvent
the current and/or fundamental economic and technological limits, we proposed an architectural
framework allowing for several interoperable virtual optical networks (VONs) [27]. Under such
a framework, each VON is allocated with appropriate resources (e.g., a subset of wavelengths)
and applies either static or dynamic (i.e., adaptive) control, whichever is more appropriate. For
example, a VON may adopt On-demand reconguration to support bursty trac and short-lived
connections (e.g, see [21,28]), while another VON may adopt Self-reconguration to support steady-
trac and long-lived connections (e.g., see [25, 29]). The former, called a dynamic VON, may use
small but fast switches (e.g. based on either Lithium Niobate directional couplers or broadcast-and-
select star couplers followed by SOAs) as well as wavelength converters. The latter, called a static
VON, may use large but slow switches (e.g. opto-mechanical switches) without the wavelength
conversion capability. It is expected that, as trac nature changes from being voice-dominant to
data-dominant, and at the same time, the device and component technology improves, the optical
layer will evolve from perceived static VONs to a mixture of static and dynamic VONs.
In this paper, we will limit our attention to dynamic VONs, and in particular, those supporting
bursty trac. Such VONs can be referred to as \bursty" VONs which can be used for running

1
IP directly over WDM. There are two important drivers for building such an envisaged Optical
Internet. One is the explosion of the data trac over the Internet, especially the World-Wide-
Web, which is bursty in nature; The other is the desire of the users/applications as well as the
opportunities provided by the breakthrough made in the WDM optical networking technology to
streamline both software (e.g. ATM signaling protocols) and hardware (e.g. SONET equipments)
to reduce latency and cost.
The main contribution of this paper is the introduction of the new switching paradigm called
optical burst switching (OBS). OBS can combine the best of the coarse-grained circuit-switching and
the ne-grained packet-switching paradigms while avoiding their shortcomings, thereby eciently
supporting bursty trac generated by upper level protocols or high-end user applications directly.
Using OBS, a control (or set-up) packet is sent rst, followed by a data burst on a separate
wavelength. Such a one-way reservation paradigm is suitable for sending data requiring a high
bit-rate and a low latency but having a relatively short duration compared to the end-to-end
propagation delay of the network.
The rest of the paper is organized as follows. In the next section, we provide a general description
of OBS protocols. In Sec. 3, we describe Just-Enough-Time (JET) [39,40], and also discuss several
specic issues related to JET-based OBS protocols. In Section 4, we evaluate the performance of
JET and its variations. We conclude the paper in Sec. 5.

2 Why OBS and What is OBS

In order to put optical burst switching (OBS) in perspective, we rst describe a framework called
polymorphic control of which OBS is an integral part.

2.1 Polymorphic Control

The framework of polymorphic control is a product of integrating many individual research ideas
and results on optical network architectures, control and management. As mentioned earlier, under
this framework, an optical layer is \sliced" into static and dynamic virtual optical networks (VONs),
which apply Self-reconguration and On-demand reconguration, respectively.
One of the basic forms of Self-reconguration is scheduled communications [5,9,29,34]. When
the bandwidth (e.g. in terms of the number of wavelengths) in a static VON is limited, the set of
communicating node pairs may be partitioned into a number of subsets such that the node pairs

2
in each subset can communicate at the same time. Each subset is to be allocated a super time-
slot during which data can be transmitted or received between the communicating nodes in that
subset, and the number of time-slots determines the schedule length. A schedule specifying, among
other things, the time-slot during which a given node pair can communicate, and the path and
wavelength it will use, is then determined. Based on such a schedule, the VON can go through a
pre-determined sequence of congurations by appropriately changing the switch settings inbetween
two super time-slots. In this way, external electronic control and its associated implementation
overhead and performance degradation are minimized.
In scheduled communications, two important performance measures are the schedule length
and the bandwidth (i.e. wavelength) requirement, which relate to each other. With sucient
bandwidth, scheduled communications become embedded communications as a special instance,
where the schedule length is one, or in other words, communications among the entire set of
communicating nodes are accommodated at the same time [25, 29]. In a similar approach, which
may also be considered as a form of Self-reconguration, a logical topology (analogous to a static
VON) containing the set of communicating nodes is devised and embedded even when bandwidth
is limited [22, 32], such that these nodes may communicate at the same time but a message from
its source to its ultimate destination may go through more than one lightpaths, thus requiring O/E
and E/O conversions at the nodes where two lightpaths meet.
In VONs adopting On-demand reconguration, where the performance measures include through-
put, utilization, delay and blocking probability, dynamically changing trac patterns are supported
by transferring data in two basic fashions, namely circuit-switching and packet-switching. With
circuit-switching, connections (or lightpaths) between source and destination pairs are established
before data is transferred, and released after the transfer is completed. Both centralized con-
trol [1, 2, 7, 18] and distributed control [21, 26, 30, 42] have been studied, and in either case, it is
common to use out-of-band signaling (i.e. a separate control network with a dedicated wavelength).
With packet-switching [4, 8, 10], each intermediate node stores an incoming packet, and then for-
wards it to the next node based on its header and a locally stored routing table. Distributed
control is natural and in-band signaling is more often used than out-of-band signaling. Note that
alternately, a ow of packets can be switched based on the match between a label carried by each
packet's header and a label stored at each node, which is set up either by previous packets of the
same ow (as in IP-switching [23]) or by the network (as in Tag-switching [33]). A bursty VON is

3
a dynamic VON that adopts a novel paradigm called optical burst switching (OBS), which can be
used to support MPLS (Multi-protocol Label Switching) in an IP over WDM environment.

2.2 Motivation
The main motivation for considering optical burst switching (OBS) is that some trac in broadband
multimedia services is inherently bursty. More specically, recent studies have shown that, in
addition to trac in a local Ethernet and between remote Ethernets (i.e. WAN trac), trac
generated by Web browsers, wide-area TCP connections (including FTP and TELNET trac
carried over TCP connections), and variable-bit-rate (VBR) video sources are all self-similar (or
bursty at all time scales) [3,19]. More importantly, some studies have concluded that, contrary to
the common assumption based on Poisson trac, multiplexing a large number of self-similar trac
streams results in bursty trac [13,24].
Existing switching paradigms in optical networks are not suitable for supporting bursty trac.
Specically, using optical circuit-switching via wavelength routing [7, 22, 31], a lightpath needs to
be established rst from a source node to a destination node using a dedicated wavelength on
each link along a physical path. The bandwidth, therefore, would not be eciently utilized if
the subsequent data transmission does not have a long duration relative to the set-up time of the
lightpath. In addition, given that number of wavelengths available is limited, not every node can
have a dedicated lightpath to every other node, and accordingly, some data may take a longer
route and/or go through O/E and E/O conversions. Furthermore, the extremely high degree of
transparency of the lightpaths limits the network management capabilities (e.g. monitoring and
fast fault recovery).
An alternative to optical circuit switching is optical or photonic packet/cell switching in which a
packet is sent along with its header [4,8,10]. While the header is being processed by an intermediate
node, either all-optically or electronically (after an O/E conversion), the packet is buered at the
node in the optical domain. However, high-speed optical logic, optical memory technology, and
synchronization requirements are major problems with this approach. In particular, the limited
buering time that can be provided to optical signals prevents worm-hole routing and virtual cut-
through routing [11,17], which are popular in systems with electronic buers, from being deployed
eectively in optical networks.
In order to provide high-bandwidth transport services at the optical layer for bursty trac in

4
a exible, ecient as well as feasible way, what is needed then is a new switching paradigm that
can leverage the attractive properties of optical communications, and at the same time, take into
account its limitations. Optical burst switching (OBS) is intended to accomplish exactly that.

2.3 An Overview of OBS

In OBS, a control packet is sent rst to set up a connection (by reserving an appropriate amount
of bandwidth and conguring the switches along a path), followed by a burst of data without
waiting for an acknowledgement for the connection establishment. In other words, OBS uses one-
way reservation protocols similar to tell-n-go (TAG) [37,38], also known in ATM as fast reservation
protocol (FRP) or ATM Block Transfer with Immediate Transmissions (or ABT-IT) [16,35]. This
distinguishes OBS from circuit-switching as well as from other burst-switching approaches using
protocols such as Reservation/scheduling with Just-In-Time switching (RIT) [15] and tell-and-wait
(TAW), also known in ATM as ABT-DT (Delayed Transmissions) [6,36], all of which are two-way
reservation protocols.
OBS also diers from optical or photonic packet/cell switching mainly in that the former can
switch a burst whose length can range from one to several packets to a (short) session using one
control packet, thus resulting in a lower control overhead per data unit. In addition, OBS uses
out-of-band signaling, but more importantly, the control packet and the data burst are more loosely
coupled (in time) than in packet/cell switching. In fact, they may be separated at the source as well
as subsequent intermediate nodes by an oset time as in the Just-Enough-Time (JET) protocol to
be described later. By choosing the oset time at the source to be larger than the total processing
time of the control packet along the path [39,40], one can eliminate the need for a data burst to be
buered at any subsequent intermediate node just to wait for the control packet to get processed.
Alternatively, an OBS protocol may choose not to use any oset time at the source, but instead,
require that the data burst go through, at each intermediate node, a xed delay that is no shorter
than the maximal time needed to process a control packet at the intermediate node. Such OBS
protocols will be collectively referred to as TAG-based since their basic concepts are the same as
that of TAG itself.
One way to support IP over WDM using OBS is to run IP software, along with other control
software as a part of the interface between the network layer and the WDM layer, on top of every
optical (WDM) switch. In the WDM layer, a dedicated control wavelength is used to provide the

5
\static/physical" links between these IP entities. Specically, it is used to support packet switching
between (physically) adjacent IP entities which maintain topology and routing tables. To send data,
a control packet is routed from a source to its destination based on the IP addresses it carries (or
just a lable if MPLS is supported) to set up a connection by conguring all optical switches along
the path. Then, a burst (e.g. one or more data IP packets, or an entire message) is delivered
without going through intermediate IP entities, thus reducing its latency as well as the processing
load at the IP layer. Note that, due to the limited \opaqueness" of the control packet, OBS can
achieve a high degree of adaptivity to congestions or faults (e.g,. by using de ection-routing), and
support priority-based routing as in optical cell/packet switching, as to be discussed later.
In OBS, the wavelength on a link used by the burst will be released as soon as the burst
passes through the link, either automatically according to the reservation made (as in JET) or
by an explicit release packet. In this way, bursts from dierent sources to dierent destinations
can eectively utilize the bandwidth of the same wavelength on a link in a time-shared, statistical
multiplexed fashion. Note that, in case the control packet fails to reserve the bandwidth at an
intermediate node, the burst (which is considered blocked at this time) may have to be dropped.
OBS can support either reliable or unreliable burst transmissions at the optical layer. In the
former, a negative acknowledgement is sent back to the source node, which retransmits the control
packet and the burst later. Such a retransmission may be necessary when OBS is to support some
application protocols directly, but not when using an upper layer protocol such as TCP which
eventually retransmits lost data.
In either case, a dropped burst wastes the bandwidth on the partially established path. However,
since such bandwidth has been reserved exclusively for the burst, it would be wasted even if one
does not send out the burst (as in two-way reservation). Similar arguments apply to optical
or photonic packet switching as well. In order to eliminate the possibility of such bandwidth
waste, a blocked burst (or an optical packet) will have to be stored in an electronic buer after
going through O/E conversions, and later (after going through E/O conversions), relayed to its
destination. Fiber-optical delay lines (FDLs) providing limited delays at intermediate nodes, which
are not mandatory in OBS when using the JET protocol, would help reduce the bandwidth waste
and improve performance in OBS as to be discussed next. Note that, when using TAG-based OBS
protocols (or optical/photonic packet switching), FDLs (or optical buers) are required to delay
each optical burst when the control packet (or the packet header) is processed, but do not help

6
improve performance.
Summarizing the above discussions, switching optical bursts achieves, to certain extent, a bal-
ance between switching coarse-grained optical circuits and switching ne-grained optical pack-
ets/cells, and combines the best of both paradigms, as illustrated in Table 1.

Optical Switching Bandwidth Latency Optical Proc./Sync. Overhead Adaptivity

(paradigm) Utilization (set-up) Buer (per unit data) (trac & fault)
Circuit low high not required low low
Packet/Cell high low required high high
Burst high low not required low high

Table 1: A comparison between three optical switching paradigms

3 The JET Protocol And Its Variations

The proposed Just-Enough-Time (or JET) protocol for OBS has two unique features, namely, the
use of delayed reservation (DR) and the capability of integrating DR with the use of FDL-based
buered burst multiplexers (BBMs), which are to be described in this section. These features make
JET and JET-based variations especially suitable for OBS when compared to TAG-based OBS
protocols and other one-way reservation based OBS protocols that lack either or both features.
S 1 2 D S 1 2 D
Τ control
control δ
Τ ∆ P s
burst δ
s
∆
burst δ
Τ s
∆

(a) (b)

Figure 1: OBS using the JET protocol

Figure 1 illustrates the basic concept of JET. As shown, a source node having a burst to transmit
rst sends a control packet on a signaling channel (which is a dedicated wavelength) towards the

7
destination node. The control packet is processed at each subsequent node in order to establish an
all-optical data path for the following burst. More specically, based on the information carried in
the control packet, each node chooses an appropriate wavelength on the outgoing link, reserves the
bandwidth on it, and sets up the optical switch. Meanwhile, the burst waits at the source in the
electronic domain. After an oset time, T , whose value is to be determined next, the burst is sent
in optical signals on the chosen wavelength (at say, 2.5Gb/s).

3.1 The Use of Oset Time

For simplicity, let us assume that the time to process the control packet, reserve appropriate
bandwidth and set up the switch is
time units at each node, and ignore the receiving and
transmission time of the control packet. In a TAG-based OBS protocol (or optical or photonic
packet switching), a burst is sent by the source along with the control packet without any oset
time (i.e., T = 0 in Figure 1). In addition, at each subsequent intermediate node, the burst waits
for the control packet to be processed, and the two are sent to the next node without any oset time
either. In this way, both the control packet and the burst will be delayed for
units, which will be
referred to as the per-node control latency. Accordingly, the minimum latency of the burst including
the total propagation time, denoted by P , but excluding its transmission time, is P +

H , where
H is the number of hops along the path (e.g., in Figure 1, H = 3).
In JET, we can choose the oset time T to be

H , as shown in Figure 1 (a), to ensure that
there is enough headroom for each node to complete the processing of the control packet before
the burst arrives. In this way, the burst will not encounter a longer latency than using TAG-based
OBS protocols. In fact, we may partition
into roughly two parts: let  be the time to process
the control packet and initiate other operations such as switch setting, and s = (
; ) be the time
required to nish these operations. Using JET, the control packet can be sent out to the next node
immediately after spending  time units at each intermediate node, or in other words, s time units
before the burst. This eectively overlaps the switching setting time at a node with the time for the
control packet to propagate to (and possibly get processed at) the subsequent node. Consequently,
one can reduce the oset time to T = 
H + s, and latency to P + T , which are s
(H ; 1) smaller
0 0

than those using TAG-based OBS protocols. In the rest of the paper, we will ignore the dierence
between
and  (and between T and T ), and just use \processing delay" to refer to the per-node
0

control latency.

8
 It is important to note that the burst can be sent without having to wait for an acknowledgement
from its destination. At 2.5 Gb/s, a burst of 500 Kbytes (or 4,000 average-sized IP packets) can
be transmitted in about 1.6ms. However, an acknowledgement would take 2.5ms just to propagate
over a distance of merely 500km. This explains why one-way reservation protocols are generally
better than their two-way counterparts for bursty trac over a relatively long distance. Once a
burst is sent, it passes through the intermediate nodes without going through any buer, so the
minimal latency it encounters would be the same as if the burst is sent along with the control
packet as in optical packet switching. Of course, if a burst is extremely small, one may just as well
send the data along with the control information using packet-switching.

3.2 DR for Ecient Bandwidth Utilization

Figure 2 (a) illustrates why delayed reservation (DR) of bandwidth is useful in achieving ecient
bandwidth utilization. Using a TAG-based OBS protocol, the bandwidth on the outgoing link is
reserved from t 1 , the time node X nishes the processing of the (rst) control packet. In JET, one
0

may also reserve the bandwidth in the same way. However, it is natural to delay the bandwidth
reservation till t1 , the time the (rst) burst arrives. Here, t1 > t 1 and their dierence is the value
0

of the oset time between the burst and its corresponding control packet at node X .
Note that, a way to determine the arrival time of a burst, e.g. t1 , when the processing time of
a control packet may vary from one node to another, is to let the control packet carry the value of
the oset time to be used at the next node. This value can be updated based on the processing
time encounted by the control packet at the current node. In the above example, immediately after
the control packet succeeds in reserving the bandwidth, its transmission is scheduled, say, at t 1 . 00

The value of the oset time to be used at the next node is then obtained by subtracting t 1 ; t 1
00 0

from the current value. Obviously, some guardbands around the bursts may still be needed to
accommodate possible jitters but due to the limited space, such a topic will not be addressed in
this paper.
In addition to taking into account the arriving time of the burst, t1 , what is more important is
that in JET, the bandwidth may be reserved until t1 + l1 , where l1 is the burst duration, instead
of until innity. This will increase the bandwidth utilization and reduce the probability of having
to drop a burst. For example, in both cases shown in Figure 2(a), namely t2 > t1 + l1 and t2 < t1 ,
respectively, the second burst will be dropped at node X if has no buer for the burst when using

9
TAG. However, when using JET, the second burst will not be dropped in case 1, nor in case 2,
provided that its length is shorter than t1 ; t2 .
1st control packet 1st burst
t’1 t1 t 1+ l 1
offset
case 2 case 1
X
Arrival Time
t’2 t2
2nd control packet 2nd burst
(a)

t’1 t1 t 1 +l 1

case 3
X
dmin Arrival Time
dmax

t’2 t2
(b)

Figure 2: Delayed reservation (DR) and its usefulness with or without buer

Note that, DR goes hand in hand with the use of oset time. In addition, although burst
length may vary, we may assume that the length of a burst is known before the corresponding
control packet is sent. This assumption is natural in some applications such as le transfer or
WWW downloading. However, if the burst length is unknown, one may delay the control packet
until either the entire burst arrives (from an upper layer), or a certain length is reached. To take
advantage of the use of an oset time in JET, thereby reducing the pre-transmission latency, an
alternative is to send out the control packet as soon as possible by using an estimated value of the
burst length. If it is an over-estimation, another control (release) packet may be sent to release
the extra bandwidth reserved. If it is an under-estimation, then the remaining data will be sent as
one or more additional bursts. JET may also support an entire session by reserving the bandwidth
to innity, and use an explicit release packet when the circuit is no longer needed (i.e. the session
ends).

3.3 Intelligent Buer Management

As mentioned earlier, JET does not mandate the use of buer, nor its size in terms of the maximal
number of bursts or bits that can be stored (or delayed) simultaneously, and/or maximal delay
it can provide to each burst at a node. Nevertheless, the dropping probability can be further
reduced, and both bandwidth utilization and performance can be further improved, if a burst can
be buered (or delayed) at an intermediate node. Figure 3 shows two possible designs of buered
10
burst multiplexer (BBM) based on ber-delay lines (FDLs), which can be used in an \output-
buered" photonic switch/router. In either case, a burst coming from any of the three inputs can
be buered for a maximal of B = 2n+1 ; 1 time units. The dierence between the shared BBM
shown in Figure 3 (a) and the dedicated BBM shown in Figure 3(b) is that the latter is more
complex (and costly) but more powerful. Again, due to the limited space, we will not address the
issue of designing cost-eective BBMs further in this paper. Nevertheless, we note that bursts at
dierent wavelengths may share the same FDL, and adding wavelength converters in a node may
increase the utilization of the FDLs even further [12, 14]. With current technology, a FDL-based
buer containing a few kilometers of ber, and thus providing a few tens of s delay is feasible [20].
0
n
B = 20 + 21 + +2
1

0 1 n
2 2 2 2
3

(a) (b)

Figure 3: An example of (a). a shared BBM and (b). a dedicated BBM

Given that optical buer is a scarce and expensive resource, JET makes eective use of buer
in two ways. Firstly, in JET, each burst will wait (i.e. be buered) at the source in the electronic
domain during the oset time and there is no need to buer it at any intermediate node at all
when its control packet encounters no blocking along the path. Secondly, in case a control packet
is blocked at an intermediate node, DR and BBM can be integrated in that DR is also useful in
increasing the eectiveness of the available buer through intelligent buer allocation and man-
agement, just as DR is useful in increasing the bandwidth utilization as discussed in the previous
subsection. For example, refer to case 3 shown in Figure 2 (b) where t1 < t2 < t1 + l1 . Clearly,
if the second burst can be delayed (or buered) for at least dmin = t1 + l1 ; t2 , then it needs not
to be dropped. In JET, node X can determine if it has a sucient number of FDLs in the BBM
at the output port. If so, it reserves a minimum number of FDLs necessary to provide a delay
d  dmin using DR (that is, reserve the FDLs only for the period from t2 to t2 + l2 instead of
from t2 to innity); Otherwise, no FDL will be reserved. Note that, without using DR, the node
0

cannot know how much delay is actually necessary. Consequently, it will have to reserve the entire
buering capacity available in the BBM at t2 , say, dmax ( B ) as depicted in Figure 2 (b). Such
0

11
a blind and brutal-force reservation will waste some of the buering capacity when dmax turns out
to be excessive (i.e. if t2 + dmax > t1 + l1 ). Even worse, it will waste the entire buering capacity
0

when dmax turns out to be insucient (i.e. if t2 + dmax < t1 + l1 ) since the burst will have to be
0

dropped after all.

3.4 Adaptive Routing and Priority Schemes

A critical design issue in OBS is how to reduce burst dropping probability. With no or very limited
buering, the dropping probability of a burst may be improved by implementing adaptive routing
and/or assigning it with a higher priority.
As mentioned earlier, a TAG-based OBS protocol does not use any oset time. Instead, a data
burst goes through a xed delay (FDL) at each intermediate node to account for the processing
delay encounted by the corresponding control packet. This facilitates the use of a dierent path to
a given destination each time a source sents a new burst (or retransmits a dropped burst), as well
as de ection-routing at intermediate nodes when a burst is blocked.
A JET-based OBS protocol can also support multi-path routing from a given source to a given
destination as long as the (approximate) number of hops along each path is known. To support
de ection-routing at an intermediate node when there is no bandwidth to reserve on the primary
outgoing link, the control packet chooses an alternate outgoing link, and sets the switch accordingly
so that the data burst will also follow the alternate path. If a minimal oset time based on the
primary path was used, and the alternate path is longer (in terms of number of hops), then the data
burst needs to be delayed further in order to make up for the increase in the total processing delay
encounted by the control packet along the alternate path. This can be accomplished by letting
the data burst go through some FDLs at one or more nodes before the oset time goes to zero,
even if no blocking occurs at these nodes. We note that a JET-based protocol can support limited
adaptivity even without using FDLs. Specically, one can use an extra oset time at the source to
account for a possible increase in the total processing delay of the control packet due to de ection
routing.
In addition to being useful for de ection-routing, having an additional oset time can increase
the priority of a burst. This is because the corresponding control packet will likely to succeed in
reserving the bandwidth into the future, given that very few other control packets arriving earlier
(or around the same time) might have reserved (or want to reserve) the bandwidth that much in

12
advance. This property of an additional oset time can be utilized to improve fairness by assigning
a higher priority to bursts which must travel for a longer distance (in terms of the number of hops)
from their sources to destinations. We will call this variation of JET which implements such a
priority scheme JET-FA (for fairness), whose performance will be shown in the next section.

4 Performance Evaluation
To evaluate the performance of JET and its variations, we assume that bursts have an exponentially
distributed duration with an average denoted by L, and that the destination of each burst is evenly
distributed among all nodes except its source. For a burst whose source-destination distance (or
path length) is H (hops), the minimum oset time is set to

H . We will use k to denote the
number of channels (e.g. wavelengths) per link, b to denote the ratio of the maximal buering time
(i.e B in Figure 3) to L, and c to denote the ratio of
to L.

4.1 The Eectiveness of DR

We rst report simulation results that compare the performance of the JET protocol (which uses
DR) with a variation called NoDR, which does not uses DR, although both use the same (minimum)
oset time.
A torus (or meshed-ring) network having N = n2 nodes which may be considered as having n
horizontal rings and n vertical rings is simulated. Each node in the network is connected to a local
host and four other nodes, and the propagation delay between any two nodes is denoted by p. To
send a burst or control packet, a shortest path from its source to its destination will be used, whose
average path length, Havg , is approximately n=2. When multiple shortest paths exist in a torus,
one is randomly chosen for each burst. In addition, loss-less communications is assumed in which
a dropped burst is retransmitted after a random backo time, whose value is evenly distributed
between 0 and 2L with an average of L.
The performance metrics used are (average) link utilization, denoted by U , as a function of the
(average) delay of a burst due to blocking, denoted by D. Note that, based on such a denition,
the delay, D, will be 0 if no blocking occurs using an one-way reservation protocol and a minimum
oset time value. As a comparison, using a two-way reservation protocol, the minimum delay will
be about Havg
(p +
), which is the time for an acknowledgment to be received by the source.
Simulation results have been obtained using the following default parameter values: N = 4  4,

13
 0.8

JET (b=1)
NoDR (b=1)
0.7 JET (b=0)
NoDR (b=0)

0.6

0.5

link utilization
0.4

0.3

0.2

0.1

0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
D (msec)

Figure 4: Performance evaluation of various protocols

k = 4, c = 0:1, and b = 1 if BBMs are used (otherwise, b = 0). In addition, we let L = 40sec,
which is equivalent to a burst of 100K bits (or 100 average-sized IP packets) transmitted at 2.5Gb/s,
and p = 2:5msec, which is equivalent to a link length of 500km. The link utilizations achieved by
JET and NoDR with or without using BBMs are shown in Figure 4. Note that, if dierent L and p
are used, the shape and the relative position of the four curves in the Figure will not change much,
only the scale of the X-axis (the delay D) will.
The results indicate that JET which uses DR alone (i.e. b = 0) can achieve the same performance
as NoDR which uses BBMs alone (i.e. b = 1). In addition, JET (b = 1) can outperform NoDR
(b = 0) by about 80%, and the other two by at least 50%. Since the performance improvement
of JET (b = 1) over NoDR (b = 0) is larger than the sum of the improvement of NoDR (b = 1)
over NoDR(b = 0) and the improvement of JET(b = 0) over NoDR (b = 0), one may conclude that
the use of DR can improves not only the bandwidth utilization, but also the buer eectiveness
(through intelligent buer allocation and management).
We have also compared JET with TAG-based OBS protocols, and the results (although not
shown) have indicated that there is no signicant dierence between the performance of the two as
long as c is small (in fact, their performance will be the same when c = 0). If, however, c is large
(e.g. c  1), and the FDLs, which are required by the TAG-based OBS protocols to delay a burst
while its corresponding control packet is being processed, can be used in JET for resolving con icts,
JET may outperform TAG-based protocols simply because the former will have an eectively larger
b.

14
4.2 Scalability Analysis
In this subsection, we compare JET with NoDR by varying the values of the parameters k, b, c
and N . Note that since both JET and NoDR will use the same b value ( b = 1 by default), any
performance improvement of JET over NoDR will be entirely due to the use of DR.
Figure 5 shows the link utilization improvement ratio of JET over NoDR as k varies. The
absolute values of link utilization U achieved by the two protocols under various k are also shown
at right as a reference. These results have been obtained under the assumption that when k > 1,
wavelength conversions can be performed so that a burst can go out on any wavelength that is free
on the outgoing link. This is why U increases with k in both protocols. However, one may also
observe that in JET, such an increase in utilization is larger than in NoDR, and as a result, the
improvement ratio is higher when k > 1 than when k = 1. This can be explained as follows. When
all the wavelengths on the outgoing link are in use, and thus a blocking occurs, a control packet
in JET can pick a wavelength on which the bandwidth will be released rst based on the known
release time of each wavelength, and reserve an appropriate amount of buer. On the other hand,
in NoDR, the blocked control packet cannot predict which wavelength will be available rst, and
thus can take little advantage of the multiple choices among wavelengths.
120 0.9
k=8 JET (k=8)
k=4 JET (k=4)
110 k=1 0.8 NoDR (k=8)
NoDR (k=4)
JET (k=1)
100 0.7 NoDR (k=1)
improvement % in link utilization

90 0.6
link utilization

80 0.5

70 0.4

60 0.3

50 0.2

40 0.1

30 0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 0 0.4 0.8 1.2 1.6 2 2.4 2.8 3.2 3.6 4 4.4
D (msec) D (msec)

Figure 5: Bandwidth utilization improvement of JET over NoDR when k = 1, 4 and 8.

From Figure 5, one may also observe that when k = 4, the improvement ratio initially increases
to close to 100% with the delay, then gradually decreases, and nally settles down at 50%. This can
be explained by examining the right half of the gure. Specically, this is caused by the following
\push-and-pull" eects. One is that U increases with D in NoDR (albeit in JET as well), and the
other is that the rate at which U increases slows down as D increases beyond a certain value. For

15
similar reasons, the improvement ratio when k = 8 is lower than when k = 4 for most values of D.
The utilization improvement of JET over NoDR when b varies is plotted in Figure 6. As can
be seen, the improvement ratio is higher for a larger b value, but converges to around 50% for all
b values when trac is heavy. Note that as with a larger k, a larger b leads to a higher absolute
value of U , which suggests that JET is scalable to both k and the maximal buering time.
80

b=5
b=2
b=1
70 b=0.5
improvement % in link utilization

60

50

40

30

20
0 0.25 0.5 0.75 1 1.25 1.5 1.75 2 2.25 2.5 2.75
D (msec)

Figure 6: Bandwidth utilization improvement ratio when b = 0:5, 1, 2, and 5.

Figure 7 shows the utilization improvement ratio of JET over NoDR for dierent values of N .
The peak improvement ratio is higher for a larger N , and this is because as N increases, so is
Havg , which results in a longer oset time used in JET, and thus a larger amount of bandwidth
(and buering capacity) saved when compared to NoDR. Note that the peak improvement ratios
of 44%, 55% and 75% are obtained at D = 2:5msec, 5msec and 7.5msec, respectively, in the 4  4,
8  8 and 12  12 tori. These values of D are still low when compared to the minimum delay of a
two-way reservation protocol (which would be about 5msec, 10msec and 15msec, respectively, in
these tori).
The eect of c on the utilization improvement ratio of JET protocol over NoDR is shown in
Figure 8 where c varies from 0:1 to 1. A larger c means a relatively larger
, and thus a larger
amount of oset time. It also means that more bandwidth will be wasted in NoDR (as discussed
in the case for a larger N ). Hence, as expected, when c increases, the improvement ratio increases
proportionally. In particular, when c = 1, the improvement ratio can reach up to 400%. Since the
transmission speed may increase much faster than the processing speed as in the past few decades,
one may expect c will increase with time, making JET more attractive and desirable.

16
 80

75 12X12
8X8
4X4
70

65

improvement % in link utilization

60

55

50

45

40

35

30

25
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14
D (msec)

Figure 7: Bandwidth utilization improvement ratio when N = 4  4, 8  8 and 12  12.

450
c=1
c=0.5
400 c=0.2
c=0.1

350
improvement % in link utilization

300

250

200

150

100

50

0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5
D (msec)

Figure 8: Bandwidth utilization improvement ratio when c = 0:1, 0.2, 0.5, and 1

4.3 QoS Support

In this subsection, we show how an additional oset time can be used to support priority and
support QoS (e.g., to achieve fairness) without requiring the use of buer at intermediate nodes.
The performance of the JEt-FA protocol is evaluated under a slightly dierent model in which
a blocked burst will be dropped and not retransmitted. Figure 9 shows the average dropping
probability of a burst as a function of the distance between its source and destination, denoted by
H , when k = 8, b = c = 0, and the trac load (relative to link capacity) is 0.5 and 0.6, respectively.
Two basic JET-FA protocols which assign an additional oset time which is a multiple of H (and
L) are simulated. Specically, in JET-FA(5L) and JET-FA(10L), a burst traveling H hops uses
an additional oset time of 5L  H and 10L  H , respectively. Note that, when L = 40sec and

17
H = 6, the maximal extra oset time would be only 6  10L or 2.4 msec, which is quite tolerable
for bursts that have to travel for 6 hops with an average per-hop distance of, say 500 km. As can
be seen, the fairness is improved over the JET protocol. However, as a trade-o, the throughput of
JET-FA can be slightly lower than JET. Nevertheless, with a large enough k, JET-FA can achieve
approximately the same throughput as JET.
0.35
JET-FA, load=0.6 (5L)
JET-FA, load=0.5 (10L)
JET, load=0.6
0.3 JET, load=0.5

0.25
dropping probability

0.2

0.15

0.1

0.05

0
1 2 3 4 5 6
number of hops

Figure 9: Improve the fairness using JET-FA protocols

Note that, we may apply the same idea in order to provide dierentiated services without using
buer in a WDM network. Specically, we can use an additional oset time when sending every
burst belonging to a high priority class. For example, assume that the durations of the low priority
bursts have an exponential distribution with an average L. If the additional oset time used is 5L,
which is longer than at least 99% of the low-priority bursts, then at most 1% of the low-priority
bursts may block a high priority burst. As a result [41], the average dropping probability of the
high priority bursts will be at least 10 times lower when k = 8, and several orders of magnitude
lower when k = 32 (although the average blocking probability of all bursts remains unchanged).

5 Concluding Remarks
In this paper, we have proposed a novel paradigm called optical burst switching (OBS) as an
ecient way to support the bursty data trac, e.g. IP trac, on top of WDM networks. We have
described two major types of OBS protocols, both of which use an out-of-band control packet to
set up the optical switches for the following data burst. One, which is based on tell-and-go (TAG),
delays a burst at every intermediate node using, for example, ber-delay lines (FDLs). The other,
which is based on Just-Enough-Time (JET), delays the transmission of the burst at its source by

18
an oset time, and thus can either eliminate the need for or make more ecient use of FDLs. We
have also proposed a time-stamping technique to facilitate the use of an oset time and delayed
reservation in JET-based protocols.
Performance evaluation results have indicated that JET-based OBS protocols can achieve a
good bandwidth utilization by using delayed reservation, and improve fairness by assigning an
additional oset time (which is equivalent to a higher priority) to bursts traveling through more
hops. As a future direction, we note that, with the limited degree of opaqueness provided by control
packets, and the ability to achieve better utilization of the network resources, OBS can be used to
eciently support multicasting at the optical layer to take advantage of the inherent multicasting
capability of some optical switches as well as the knowledge of the physical topology of the WDM
layer.

Acknowledgement
The authors would like to thank Yousong Mei and Xijun Zhang of LANDER at University of
Bualo for their valuable comments.

References
[1] K. Bala, T. Stern, and K. Bala. Algorithms for routing in a linear lightwave network. In
Proceedings of the IEEE InfoCom, pages 1{9, 1991.
[2] R. A. Barry and P. A. Humblet. Models of blocking probability in all-optical networks with
and without wavelength changers. In Proceedings of IEEE Infocom, pages 402{412, April 1995.
[3] J. Beran, R. Sherman, M. Taqqu, and W. Willinger. Long-range dependence in variable-bit-
rate video tarc. IEEE Transactions on Communications, 43:1566{1579, 1995.
[4] D. Blumenthal, P. Prucnal, and J. Sauer. Photonic packet switches - architectures and exper-
imental implementations. IEEE Proceedings, 82:1650{1667, November 1994.
[5] M.S. Borella and B. Mukherjee. Ecient scheduling of nonuniform packet trac in a
WDM/TDM local lightwave network with arbitrary transceiver tuning latencies. IEEE Journal
on Selected Areas in Communications, 14:923{934, 1996.

19
 [6] P. E. Boyer and D. P. Tranchier. A reservation principle with applications to the ATM trac
control. Computer Networks and ISDN Systems, 24:321{334, 1992.
[7] I. Chlamtac, A. Ganz, and G. Karmi. Lightpath communications: an approach to high-
bandwidth optical WANs. IEEE Transactions on Communications, 40:1171{1182, July 1992.
[8] I. Chlamtac et al. Cord: contention resolution by delay lines. IEEE Journal on Selected Areas
in Communications, 14:1014{1029, June 1996.
[9] H.S. Choi, H.-A. Choi, and M. Azizoglu. Ecient scheduling of transmission in optical broad-
cast networks. IEEE/ACM Transactions on Networking, 4:913{920, 1996.
[10] R. Cruz and J.-T Tsai. Cod: alternative architectures for high speed packet switching.
IEEE/ACM Transactions on Networking, 4:11{21, February 1996.
[11] W.J. Dally and C.L. Seitz. Deadlock-free message routing in multiprocessor interconnection
networks. IEEE Transactions on Computers, 36:547{553, May 1987.
[12] S. Danielsen et al. WDM packet swicth architectures and analysis of the in uence of tunable
wavelength converters on the performance. IEEE/OSA Journal of Lightwave Technology,
15:219{227, February 1997.
[13] A. Erramilli, O. Narayan, and W. Willinger. Experiemental queueing analysis with long-range
dependent packet trac. IEEE/ACM Transactions on Networking, 4(2):209{223, 1996.
[14] J. Gabriagues and J. Jacob. Exploitation of the wavelength domain for photonic switching in
IBCN. In Proc. of ECOC, volume 2, pages 59{66, 1991. Paris, France.
[15] G.C. Hudek and D.J. Muder. Signaling analysis for a multi-switch all-optical network. In
Proceedings of Int'l Conf. on Communication (ICC), pages 1206{1210, June 1995.
[16] ITU-T Rec. I.371. Trac control and congestion control in B-ISDN. Perth, U.K. Nov. 6-14,
1995.
[17] P. Kermani and L. Kleinrock. Virtual cut-through : A new computer communication switching
technique. Computer Networks, 3:267{286, 1979.
[18] K.C. Lee and Victor O.K. Li. A circuit rerouting algorithm for all-optical wide-area networks.
In Proceedings of IEEE Infocom, pages 954{961, 1994.
20
[19] W. Leland, M. Taqqu, M. Willinger, and D. Wilson. On the self-similar nature of Ethernet
trac (extended version). IEEE/ACM Transactions on Networking, 2(1):1{15, 1994.
[20] F. Masetti, P. Gavignet-Morin, D. Chiaroni, and G. Da Loura. Fiber delay lines optical buer
for ATM photonic switching applications. In Proceedings of IEEE Infocom, volume 3, pages
935{942, 1993.
[21] Y. Mei and C. Qiao. Ecient distributed control protocols for WDM optical networks. In Proc.
Int'l Conference on Computer Communication and Networks, pages 150{153, September 1997.
[22] B. Mukherjee, S. Ramamurphy, D. Banerjee, and A. Mukherjee. Some principles for designing
a wide-area optical networks. In Proceedings of IEEE Infocom, pages 110{118, 1994.
[23] P. Newman, G. Monshall, and T. Lyon. IP switching { ATM under IP. IEEE/ACM Transac-
tions on Networking, 6:117{129, April 1998.
[24] V. Paxon and S. Floyd. Wide area trac: the failure of Poisson modeling. IEEE/ACM
Transactions on Networking, 3(3):226{244, 1995.
[25] C. Qiao and Y. Mei. On the multiplexing degree required to embed permutations in a class of
interconnection networks. In Proceedings of the IEEE Symp. High Performance Computer Ar-
chitecture, pages 118{129, February 1996. (A comprehensive version is to appear in IEEE/ACM
Trans. on Networking).
[26] C. Qiao and Y. Mei. Wavelength reservation under distributed control. In IEEE/LEOS
Broadband Optical Networks, August 1996. (a comprehensive version has been submitted to
IEEE/ACM Transactions on Networking).
[27] C. Qiao, Y. Mei, M. Yoo, and X. Zhang. Polymorphic control for cost-eective design of
optical networks. In NSF DIMACS Workshop on Multichannel Optical Networks: Theory and
Practice, March 1998.
[28] C. Qiao and R. Melhem. Reducing communication latency with path multiplexing in opti-
cally interconnected multiprocessor systems. IEEE Transactions on Parallel and Distributed
Systems, 8(2):97{108, 1997. (A preliminary version appeared in HPCA'96).
[29] C. Qiao, X. Zhang, and L. Zhou. Scheduling all-to-all connections in WDM rings. In SPIE
Proceedings, All Optical Communication Systems: Architecture, Control and Network Issues,

21
 volume 2919, pages 218{229, November 1996. (a modied version has been accepted for
publication in IEEE/ACM Transaction on Networking).
[30] R. Ramaswami and A. Segall. Distributed network control for wavelength routed optical
networks. In Proceedings of IEEE Infocom, pages 138{147, March 1996.
[31] R. Ramaswami and K.N. Sivarajan. Optimal routing and wavelength assignment in all-optical
networks. In Proceedings of IEEE Infocom, pages 970{979, June 1994.
[32] R. Ramaswami and K.N. Sivarajan. Design of logical topologies for wavelnegth-routed optical
networks. IEEE JSAC/JLT Special Issue on Optical Networks, 14(5):840{851, 1996.
[33] Y. Rekhter et al. Tag switching architecture overview. IEEE Proceedings, 82:1973{1983,
December 1997.
[34] George N. Rouskas and Vijay Sivaraman. Packet scheduling in broadcast WDM networks with
arbitrary transceivers tuning latencies. IEEE/ACM Transactions on Networking, 5(3):359{370,
1997.
[35] H. Shimonishi, T. Takine, M. Murata, and H. Miyahara. Performance analysis of fast reserva-
tion protocol with generalized bandwidth reservation method. In Proceedings of IEEE Infocom,
volume 2, pages 758{767, 1996.
[36] Jonathan S. Turner. Managing bandwidth in ATM networks with bursty trac. IEEE Network,
pages 50{58, September 1992.
[37] E. Varvarigos and V. Sharma. The ready-to-go virtual circuit protocol : A loss-free protocol for
multigigabit networks using FIFO buers. IEEE/ACM Transactions on Networking, 5(5):705{
718, October 1997.
[38] I. Widjaja. Performance analysis of burst admission-control protocols. IEE proceedings-
communications, 142(1):7{14, February 1995.
[39] M. Yoo, M. Jeong, and C. Qiao. A high-speed protocol for bursty trac in optical networks.
In SPIE Proceedings, All Optical Communication Systems: Architecture, Control and Network
Issues, volume 3230, pages 79{90, November 1997.
[40] M. Yoo and C. Qiao. Just-enough-time(JET): a high speed protocol for bursty trac in optical
networks. In IEEE/LEOS Technologies for a Global Information Infrastructure, August 1997.
22
[41] M. Yoo and C. Qiao. A new optical burst switching protocol for supporting quality of service.
In SPIE Proceedings, All Optical Networking: Architecture, Control and Management Issues,
volume 3531, pages 396{405, November 1998.
[42] X. Yuan, R. Melhem, and R. Gupta. Distributed path reservation algorithms for multiplexed
all-optical interconnection networks. In Proceedings of the IEEE Symp. High Performance
Computer Architecture, pages 38{47, February 1997.

23