Transcript OBS-1

Optical Burst Switching (OBS)
1
Optical Internet
• IP runs over an all-optical WDM layer
– OXCs interconnected by fiber links
– IP routers attached to OXCs
• Increasing discrepancy between optical
transmission and electronic switching speed
– Want “through” traffic to be switched in the
optical domain to eliminate the electronic
bottleneck at the IP layer
2
Optical Circuit Switching
• Lightpaths are set up between source and destination
nodes
– No optical buffer needed at the intermediate nodes
– Bit rate and protocol transparency
• Setting up a connection takes a few hundreds of ms 
Not suitable for short lived connections
• Bandwidth allocated by one wavelength at a time,
however, most applications only need sub- bandwidth
• No statistical multiplexing  Inefficient bandwidth
utilization when carrying bursty traffic
3
Optical Packet Switching
• High bandwidth utilization due to statistical
multiplexing
• Need to buffer packets at intermediate nodes
• Not feasible in the near future
– Current optical switches (OXCs) too slow for packet
switching
– No practical optical buffer
– Immaturity of high-speed optical logic
4
The Challenge
• How to efficiently support bursty traffic
with high resource utilization as in packet
switching while requiring no buffer at the
WDM layer as in circuit switching?
• Answer: Optical Burst Switching (OBS)
5
OBS
• Burst assembly/disassembly at the edge of
an OBS network
– Multiple IP packets aggregated into a burst at
the ingress node
– Data bursts disassembled at the egress node
– Packets/bursts buffered at the edge during burst
assembly/disassembly
6
OBS
• Separation of data and control signals in the core
– For each data burst, a control packet containing the
header information (including burst length) is
transmitted on a dedicated control channel
• A control packet is processed electronically at each
intermediate OBS node to configure the OXCs
– An offset time between a control packet and the
corresponding data burst
• The offset time is large enough so that the data burst can be
switched all-optically without being delayed at the
intermediate nodes
7
Advantages of OBS
• No optical buffer or fiber delay lines
(FDLs) is necessary at the intermediate
nodes
• Burst-level granularity leads to a statistical
multiplexing gain absent in optical circuit
switching
• A lower control overhead per bit than in
optical packet switching
8
OBS Building Blocks
• Burst assembly: assembly of client layer
data into bursts
• Burst reservation protocols: end-to-end
burst transmission scheme
• Burst scheduling: assignment of resources
(wavelengths) at individual nodes
• Contention resolution: reaction in case of
burst scheduling conflict
9
Burst Assembly
• Aggregating packets from various sources
into bursts at the edge of an OBS network
– Packets to the same OBS egress node are
processed in one burst assembly unit
– Usually, one designated assembly queue for
each traffic class
– Create control packet and adjust the offset time
for each burst
10
Burst Assembly Algorithms
• Timer-based scheme:
– A timer starts at the beginning of each assembly cycle
– After a fixed time T, all the packets that arrived in this
period are assembled into a burst.
• Effect of time out value T
– T too large: the packet delay at the edge will be too
long.
– T too small: too many small bursts will be generated
resulting in a higher control overhead.
• Disadvantage: might result in undesirable burst
lengths.
11
Burst Assembly Algorithms
• Burstlength-based scheme:
– Set a threshold on the minimum burst length.
– A burst is assembled when a new packet arrives
making the total length of current buffered
packets exceed the threshold.
• Disadvantage: no guarantee on the assembly
delay
12
Burst Assembly Algorithms
• Mixed timer/threshold-based assembly
algorithm:
– A burst is assembled when either the burst
length exceeds the desirable threshold or the
timer expires
– Address the deficiency of both timer-based and
burstlength-based schemes
13
Burst Assembly Algorithms
• After a burst is generated, it’s buffered in the
queue for an offset time before being transmitted
• During the offset period, packets may continue to
arrive
– Can’t include the packets in the same burst
– Leaving the packets for the next burst will increase the
average delay.
• Use burst length prediction to minimize the extra
delay
14
Burst Length Prediction
• Let the control packet carry a burst length of l+f(t)
– l : the exact burst length when the control packet is sent
– f(t) : the predicted extra burst length as a result of
additional packet arrivals during the offset time t.
• Assume the total length of packets actually arrive
during the offset time is l(t)
– If l(t) < f(t), part of the bandwidth reserved will be
wasted
– If l(t) > f(t), the extra packets are delayed to the next
burst
15
A Burst Reservation Protocol:
Just-Enough-Time (JET)
• Basic ideas
– Each control packet carries the offset time and
burst length
– The offset time is chosen so that no optical
buffering or delay is required at the
intermediate nodes
– Delayed reservation: the reservation starts at the
expected arrival time of the burst
16
JET
• Control packet is followed by a burst after a
base offset time T    ( h )
H
h 1
– (h): time to process the control packet at hop
h, 1  h  H
– No fiber delay lines (FDLs) necessary at the
intermediate nodes to delay the burst
– At each intermediate node, T is reduced by (h)
17
JET
• Use Delayed Reservation (DR) to Achieve
efficient bandwidth utilization
– Bandwidth on the output link at node i is reserved from
the burst arrival time ts to the burst departure time ts + l
(l = burst length)
– ts = ta + T(i), where T ( i )  T    ( h ) is the offset time
remaining after i hops and ta is the time at which the
processing of the control packet finishes
i
h 1
• The burst is dropped if the requested bandwidth is
not available
– Can use FDLs at an intermediate node to resolve
contention
18
JET
19