Supporting Ethernet in OBS
Sami Sheeshia and Chunming Qiao
Department of computer science and engineering
State University of New York at Buffalo
Computer and Communications Research Lab
Industrial Technology Research Institute, Taiwan
2002 Journal of Optical Computing
various standards to enhance Ethernet
10GbE over SONET
disadvantages and quantifying its inefficiencies
Ethernet over OBS
presenting a viable alterative which avoids the
shortcomings of SONET.
Ethernet over OBS specific integration issues
such as OBS burst size
Multiprotocol Label Switching (MPLS) issues
The paper introduces the likely role that OBS
will play in the development of 10
SONET (Synchronous optical networking) is
being proposed for the same but its
(TDM) is inefficient for transporting bursty
traffic of the Ethernet.
The author claims that OBS provides a better
sharing of network resources and when
coupled with generalized multiprotocol label
switching (GMPLS) provides a robust Ethernet
Recently network providers face new
challenges at the MAN level due to the
increased user requirements
Tremendous pressure to support
high speed WAN at low cost.
So the above factors demands a flexible,
proven MAN architecture combined with
vendor compatible implementations.
promises to play an important role,
offering good speed, end
consistency for providers and users in a cost
10GbE can operate over long link spans(40
km) and other transport layers such as
SONET and DWDM.
area connectivity is been provided by
SONET which were built to carry voice traffic.
But it is
not optimized for the bursty nature
of the present
time data traffic and is not able to scale itself to support the
rapid growth of internet in a cost
effective manner (requires
ADMs, hubs etc).
Gigabit Ethernet has been popular in LAN and the MAN ,
however, Ethernet by itself cannot be considered to be a
carrier class protocol as it does not provide what SONET
Standards such as
Ethernet over WDM, resilient packet ring
(RPR), packet over SONET (PoS)
have been developed
1.Ethernet over WDM
This provides a long haul connectivity(40 km)
between two Ethernet switches.
point connections are done manually or via
multiprotocol lambda switching.
Total number of circuits that can be provisioned are
No traffic grooming or aggregation can be done.
2. Resilient packet ring (RPR)
Combines the advantages of Ethernet and SONET to
give traditional carrier class features.
Has good MAC switching capabilities providing core
fiber and ring operation.
Controlled bandwidth allocation, 50
restoration, bounded delay and jitters etc.
Though RPR in conjunction with SONET and
Ethernet can provide highly efficient MAN, carriers
have opted to improve SONET transport instead.
3.Packet over SONET (PoS)
PoS is a communication protocol for
transmitting packets over SONET.
PoS supports the sending of IP packets over
It aggregates and encapsulates IP datagrams in
Point protocol (PPP) without any
differentiation among different packet flows.
So, lacks multicast and QoS capabilities.
Provides better bandwidth granularity than
SONET and POS.
Virtual Concatenation (VC) technique allows
a better match between Ethernet data rates and
SONET line rate improving circuit use.
GFP is proposed to adapt multiservices to the
SONET by a uniform mapping of Ethernet
frames and client signals.
LCAS provides bandwidth
Does not address issues associated with using
switching technology for data traffic.
A new proposal
Since the standards mentioned either lack end
end optical transparency or do not offer
efficient use of core resources,
technology is taken advantage of to transmit
Ethernet frames over WDM links.
The author explains this provides a
and data optimized alternative
Ethernet over SONET
SONET was designed primarily for voice
applications and is based on circuit oriented
Dual ring and equipment topology enables SONET to
implement fast protection mechanisms.
10GbE physical layer is compatible with SONET.
192 link speed, use of SONET framing).
Costly aspects like TDM support, performance and
management functions have been avoided.
Disadvantages of using SONET rings for
Basically SONETs point
point design, circuit switch
applications give rise to its limitations.
SONET provisions point
point circuit between
ring nodes => allocated a fixed amount of bandwidth.
In Fig.1(a) each node in the ring is allocated only one
quarter of the total bandwidth.
Thus a limit is put on the maximum data burst traffic
rates and increases the queuing delay.
To create a logical mesh as in Fig.1(b) requires
1) circuits which is time consuming and also
waste ring bandwidth.
Layer 2 muticast requires each source to allocate a
separate circuit for each destination.
Thus, a mesh has to be created in which separate
copies of the packet are sent to each destination.
Obviously, result multiple copies traveling around the
ring, wasting bandwidth.
3.Wasted Protection Bandwidth
50% of ring bandwidth and equipment is reserved for
SONET does not give the provider the choice of
when and how much bandwidth to reserve for
Thus SONET does not provide protection in a cost
SONET overhead (TDM capabilities, physical
layer, management functions) for STS
resulting from the transportation of 10GbE frames
over circuits built from SONET links is 3.7%.
But , at low traffic loads, efficiency worsens as
large bursts of Ethernet frames are delivered in a
fraction of SONET time slot.
Carrying increasing data traffic over manually
switched networks makes it
difficult to develop new services thus increasing
Due to the burstiness and variable frame size of Ethernet traffic
there is inefficiency in SONET framing.
Efficiency of EoS is given by
packing efficiency of Ethernet frames into
SONET payload (SPE).
unused remainder (due to low traffic load, size of
incoming frame >>remainder in time slot)
Let the Ethernet frame size be modeled with an
exponential distribution with parameter
The packing efficiency of the Ethernet frames into a SONET Tspe
is given by
maximum number of Ethernet frames that SPE can hold.
size of the burst.
Sum of the first frame to the last frame.
Due to the bursty nature, the time intervals (g) >>>Tspe
Thus as g varies ,
PE becomes proportional to the offered load.
Ethernet over OBS
Lack of statistical multiplexing, leads to poor
Recent arrival of optical cross connects (OXC) can be used to
extend 10GbE over long distances
point to point permanent
no longer shared among different 10GbE.
OBS is a new technology that exploits the large bandwidths of
DWDM by avoiding electronic processing of optical packets.
O conversions in OBS.
basic data block, collection of data frames with same
destination address and QoS parameters.
Optical burst is switched by a predetermined path by a
Improves backbone efficiency and offers excellent
The GMPLS labels are used to identify the
destination edge switch and are mapped to available
wavelengths at each OBS switch.
do not reserve the resources permanently
LOBS are set up dynamically to support required
QoS and torn down once burst are transmitted.
Bandwidth allocated dynamically
burst by burst
GMPLS extends WAN connectivity into the MAN
Complex optical UNI interfaces not required.
3. Multicast traffic:
OBS provides multicast at WDM layer using light
require additional hardware
complicate switch controls
based multicast provides simplification
and scalability at no cost.
4.Protection and Restoration:
GMPLS fast reroute mechanism can be used to
provide protection and restoration by automatically
rerouting traffic on an alternate LOBS.
Accomplishes by precomputing number of LOBS
paths at the same time the primary LOBS paths are
Integrating 10GbE over GMPLS enabled OBS
requires specific details to burst size and GMPLS
OBS efficiency depends on the fraction of time the reserved
bandwidth along a LOBS path is used by the burst .
time at which the bandwidth reservation starts
burst time (burst size L/output data rate bo)
end queuing delay depends whether fixed or
variable size bursts are used.
burst allows paths to be set up as the burst
is being assembled.
In the fixed size the data burst encounters no edge
delay as long as the offset time = time taken to
assemble the burst.
, control packet can only be sent
after entire burst is assembled
minimum edge delay
OBS Burst Size
For better efficiency, the OBS burst must be >>To
not so large to cause significant queuing delays.
High bandwidth, full
duplex operation requires a flow
control mechanism: 10GbE source and destination
exchange flow control packets
to prevent data loss,
throttle frame transmission rate.
control packets are typically 64 bytes, not
assembled into bursts since it affects efficiency.
Elasticity of OBS burst unlike SONET fixed frame slot
size allows it to adapt to the bursty nature of Ethernet.
Ethernet adaptor requires a minimum IFG between 2
successive frames; at 10GbE IFG ~9.6ns or 12 Bytes.
IFG provides receiving station time to update counters,
check CRC of previous frame, mange the buffers.
Two ways to implement IFG:
Insert the 12 Byte spacing at the 10GbE receiving
places new requirements on the processing at the
destination switch to seek the start of each frame.
Incorporate the IFG within the burst
maintains the integrity of Ethernet transmission at
the expense of wasted bandwidth within each burst.
Since the IFG, preamble, frame check sequence
(FCS) constitute a large overhead for the 1500 Byte
9000 Byte frames can be used to
For e.g.. 155,520
Byte(SONET slot) carrying 1500
Byte frames incurs 1% overhead whereas 9000
frame requires fewer IFG, preamble resulting in
0.1%, a reduction of ten fold.
Performance comparison of EoS and EoB
under different traffic conditions.
: burst aggregation time
What is Multi
protocol Label Switching (MPLS) ?
A data mechanism to provide a unified service for both circuit
based clients and packet
Supports various network technologies like ATM, IP etc.
MPLS sets up label
switched paths (LSP) for packets , thus
saving time for a LSR (router) to look up the next node for
forwarding the packet to.
Does not have common control and traffic engineering for
wavelength , TDM and fiber switching.
Forwarding Equivalence class (FEC)
Set of packets with similar characteristics is forwarded the same way.
Working of a Label Switched Path
: LSR (Label Switch Router)
: LER (Label Edge Router)
Extends the MPLS functionality for TDM (labels
and FDM (labels= electromagnetic frequency of light waves).
Ethernet tunnels provisioned dynamically to create wide
peer networking where CPE is part of the MPLS
Establishes space division multiplexed paths where labels
indicate the position of the data.
Establishes common control plane between IP service
management and optical layers.
: Virtual Leased Lines
or Ethernet tunnels.
: Customer Premise
: Provider’s Edge
∙ VLAN (Virtual LAN) :
a group of end
multiple LAN segments and
can communicate as if they
were on a single LAN.
CPE exchanges tables with the provider’s edge (PE) switches
using GMPLS signaling.
Ethernet frame arriving at the CPE from local LANS carry the
original Ethernet fields and 802.1p/q headers (p
class expediting and dynamic multicast filtering part of MAC
defines services provided in Virtual LANs).
Frame is mapped to a FEC. FEC lookup yields the outgoing
port and virtual circuit (VC) label, which is added to the frame
and forwarded on the outgoing port toward PE switch.
The PE switch maps the Ethernet frame into an OBS burst
and provides the required signaling by GMPLS label
assignment in one of the following two addressing
Flat assignment scheme
PE constructs a label forwarding base as shown below.
Like CPE label assignment , each label designates a
FEC, one for each frame.
Only provides simple point to point connection.
Increases amount of labels and signaling.
Hierarchical assignment scheme:
Here each VLAN is assigned one VLL or LOBS label as
shown in below.
So labels with same LOBS label aggregated in to the same
Allows easy rerouting of VLAN traffic in event of failure.
Additional processing overhead.
PE looks up the incoming label, determines the VLAN,
adds a LOBS label as shown below.
Frames with same LOBS label are aggregated into the same
EoB path protection and restoration
At WDM level:
At optical layer, schemes match that of SONET but
inflexible, subjects of current research.
At Ethernet level:
Spanning Tree Protocol (STP ) is used to provide redundant
Long convergence time not suitable =>LOBS constantly
setup and torn down.
At GMPLS level
when path fails, OBS switch signals head
switch to use a backup LOBS path.
makes a local decision to redirect the burst.
Timely detour path is needed.
So, preestablished paths is essential in burst traffic.
Shortest reroute time =>decision made as close to the
Since it takes time to inform the head node, local reroute
mechanism is preferred.
Impossible to predict where failure may occur.
Every switch and link along the path is protected, detour
paths setup dynamically at the same time primary paths are
Local reroute requires establishing (N
1) detour paths as
number of OBS switches that the LOBS traverses.
Extending Ethernet services over OBS provides
better scalability and bandwidth efficiency than with
SONET is limited by its TDM nature and by
which does not provide statistical multiplexing.
As DWDM evolves more and more wavelength will
be supported on each fiber.
Ethernet is best for LAN
OBS provides efficient sharing of backbone
GMPLS provides standard control mechanism to
bridge distant MANs, provides protection with no
permanent reservation of resources.
Providers improve their revenue stream and
consumers reduce their network operational cost.