1. Optical Packet/Burst Switching based on : “Optical Packet and Burst Switching Technologies for the Future Photonic Internet” S.J. Ben Yoo
Raimena Veisllari

2. Content Short Introduction
Optical Burst Switching (reservation, contention resolution)
Optical Switches Fabrics/Technologies
Optical Header Processing
Optical Packet Synchronization/Time Switch
Optical Packet Switches
Optical Label Switching
Testbed demonstration of edge/core OLR
Summary

3. Optical Networks evolution
WDM ptp first-generation
The true benefit of optical networking
may rise from avoiding electronics in the data plane
All-optical second generation
Format and protocol transparency
Simplifies hardware requirements in the
data plane?
ROADM and OXC
OPS/OBS
True IP over WDM
Statistically
Multiplexing

4. Optical Networks Evolution
1st-stage (IP over WDM and IP over OTN), static model
2nd-stage (IP over ASON), overlay model: wavelength paths are dynamically assigned among IP routers on demand (usually a single-hop model).
3rd-stage (IP over MGS), peer model/integrated model: wavelength paths OLSP combined with electrical label switched paths ELSP are dynamically routed/provisioned
4th-stage (IP over OBS/OFS), integrated model: this architecture attempts to minimize the need for processing and buffering by aggregating flows of IP packets into optical bursts.
5th-stage (IP over OPS), integrated model: IP packets can be switched in the optical domain. Optical processing/optical buffer will become possible in this stage.
Wei et. al.“High-Performance Hybrid-Switching Optical Router for IP over WDM Integration”

5. The main goal of moving from OCS to OPS/OBS is wavelength granularity vs. subwavelength granularity but loosing the guaranteed delivery, fixed delay
Adapting better to the bursty nature of the Internet traffic but still under research OBS only a couple of commercial companies while OPS still in research?

6. OBS principles
Quickly transport large amount of data without provisioning long-lasting circuits.
Burst Header Cell (BHC) or BCP
Depends on the reservation scheme (one-way or two-way signalling), Usually Hdr info + burst length
If no resources available->contention resolution based on local node
In here the BCP is in-band (in the same wavelength) with the data so you have to set a

7. OBS Control Protocols
Based on the reservatio/contention resolution schemes employed. Differentiate between:
Setting the switching matrix : Sending the control packet in advance/not in advance (Toffset)
Releasing the switching matrix: Giving the payload length in advance/use release packet after the burst
One-way reservation (no ACK) or two-way reservation
Compare the low latency of one-way and the guaranteed delivery of two-way
TAG-based OBS (No ACK out-of-band) to achieve both datagram and VC switching
JET signalling (No FDL)
Built-in-offset Toffset=0 (FDL)
Other schemes available like fixed or limited duration and two-way signalling schemes based on RWA algorithms with practical limitation on number of nodes (not discussed).
In here the BCP is in-band (in the same wavelength) with the data so you have to set a

8. Just Enough Time (JET)
The src node S has a burst b to transmit send the BCP or BHc signal in the channel toward the D
The BHC is processed in node 1 and 2 : the node chooses the appropriate wavelength on the outgoing link, reserves the bandwidth on it, sets up the optical switching matrix
S send the burst after offset time T(i) calculated dt x H where dt is the time needed to process the BCP in one node and H number of hops along the path. The BCP will carry the arrival time info along the path as well for each node will process the burst at a delayed time T+dt x H(i). The BCP also has the release time info because it has the length of the burst in the BCP ts + l will be the release time

9. OBS (2)
Requires a careful precomputation of T to avoid possible burst loss by compensating the total latency experienced by BHC.
Research on varying the QoS (CoS) by varying the offset time T.
Limitation: The burst blocking probability related to the number of wavelegths and the traffic load.
The built-in TAG-OBS uses OPS schemes
The built-in optical buffer (FDL) allows the burst to be «queued» for the time it takes to process BHC and set the switching matrix
Usually a low offset time + FDLs are employed throughout the network! (OPS-like OBS)
In here the BCP is in-band (in the same wavelength) with the data so you have to set a

10. OBS Contention Resolution
Contention: Burst requiring the same output, same wavelength at the same time in one node -> use alternative forwarding path
Wavelength domain
The most effective solution because it does not require additional latency while maintaining the shortest path or minimum hop.
Time domain
FDL (FIFO) and all inherited problems of such queueing and FDL size
Space domain
Hot-potato, forward to another output and let the network itself be a buffer.
Out-of-order sequencing
Delay/jitter

11. Optical Switching Fabrics (1)
OBS vs. OPS
Subwavelength granularity
Fast switching speeds (us, ms) vs. nanoseconds in OPS
Other considerations?
Optical Switching fabrics carachteristics
Signal Quality Issues : Crosstalk, Jitter, Chirp, attenuation, OSNR
Configuration Issues: Scalability, blocking/nonblocking, promptness, switching domain, optical transparency, practical implementation
Performance issues: Switching speed, PLD, Insertion loss, level of transparency
Wide-sense non-blocking are those that can realise any new connection according to its routing rules but without changing the active connections (as in blocking happens)
Strict-sense non-blocking are those that do not require rearrangement of complex routing algorithm for each new connection; so that these new connections can use any free path in the switch.
Switching speed=the time it takes the switch to change its state

12. Optical Switching Fabrics (2)
Optical switching technologies for OCS, OBS, and OPS

13. OPS Technologies (1) Categorizing based on the combination of:
Synchronous/Asynchronous pkt switching
Fixed/Variable packet length
Store and forward vs cut-through pkt switching
WDM, TDM and Optical CDM (difficulties in developing multiplexing devices for TDM and CDM)
Synchronous
fixed length
Asynchronous
Variable length

14. Optical Hdr Processing

15. Optical Hdr Techniques
The RF fading as a result of coherent interference between the carrier (payload in the baseband) and the two sidebands, constructive or destructive.

16. Optical Packet Synchronizer/Time Switch
In a system with N time granularity, K ports and W wavelengths there are needed NxKxW modules (Scalability? Complexity compared with the OPS itself?)
One possible solution is shared or loopback buffering.
The CORD project: The signal comes at the input with a specific wv and is splitted to pass through header processing that sees wt time frame it needs to be in so its a delay counter so to speak, then the TWC is set so that the signal corresponds to the appropriate wv for that delay and goes through the SOAgates and to FDL. This is the first stage of the delay lines. Than the signal is coupled once again and sent throuh the second stage of delay through a demux AWG

17. OPS for packet switching (1)
Guard time : longer than the longest transition time but short for efficient switching
Space Switching (KEOPS example Broadcast and select or NxN OXS with SOA)
The guard time between packets should be longer than the longest transition time of the switches in the path
BS all packets are broadcasted to experience all possible delays achievable. he electronic circuit though will select the time (delay) for each input and controll at the wavelength Selector which input we are transmitting

18. OPS for packet switching (2)
Optical Phase Array
Like a phased-array radar, the OPA components select wavelength paths across a 64 × 64 cross-point switch via an optical interference mechanism that operates by changing the waveguides’ refractive indices. 64 non-blocking 1x64 switches, the switching time 30ns.
Wavelength Routing Switching Fabric
KW x KW AWGRs with F shared
recirculating FDLs; Switching in time,
space and wavelength

19. OPS for packet switching (3)
Store and Forward OPS and Optical Buffers
The lack of the optical buffer is the main problem in the OPS research so far mainly because of this switching paradigm.
TCP congestion control algorithm determines the size of the buffer
RTT x (data rate of the link) -> For OPS 25Mb for 10Gbps link
PLR vs systems scalability
Pipelined router architecture
TDM and CDM OPS
Research is less active due to the difficulties of producing ultrafast mux/dmux in optical TDM and optical CDM technologies
OPS using CMOS/RAM
Switch with high speed OE converters parallel to serial -> CMOS RAM -> serial to parallel (Is it still OPS all-optical?)

20. Optical Label Switching (OLS)
DARPA proposal and patent (optical-tag switching) interoperable with MPLambdaS through GMPLS extension. It facilitates interoperability between OCS, OBS and OPS.
Discarding Store-and-forward,
4 classes of labels 40 bits long:
Class A label dst oriented similar to IP hdr ( dst, src, QoS, CoS, optical TTL, exp bits)
Class B = Class A plus TE in the exp bits
Class C for label based forwarding similar to MPLS
Class D for Circuit Switching
Use a unified and pipelined contention resolution scheme in the wavelength, time and space domains
Error-free 101 hop-cascaded OLS router operations have been demonstrated with rapid clock recovery 1ns and guard time 2-3ns

21. Optical Label Switching (OLS)

22. Summary
OPS and OBS research on the combination of the vast optical bandwidth and subwavelength granularity by switching/routing packets and burst in the optical layer.
OBS offers BE with one-way signalling with milli to microseconds switching time
OPS needs faster switching times up to nanoseconds to be effective
Optical Label (Hdr) Processing and switching in times, space and wavelength domains provide the nanosecond speeds.
OLS facilitates the interoperability between OCS, OPS and OBS with less power requirement and reduced complexity??