Date post: | 27-Dec-2015 |
Category: |
Documents |
Upload: | arline-warren |
View: | 217 times |
Download: | 0 times |
Unidirectional rings◦ Most WDM ring net-
works are based on unidirectional fiber ring carrying W wavelengths
◦ Each of the N ring nodes deploys OADM to drop and add traffic
◦ For N = W, each node has dedicated home channel for reception
◦ For N ≥ W, system becomes scalable since number of nodes is independent of number of available wavelengths
Channel vs. receiver collision◦ Each ring node is equipped with
i fixed-tuned transmitters (FT) j tunable transmitters (TT) m fixed-tuned receivers (FR) n tunable receivers (TR)
◦ Node architecture described as FTi-TTj-FRm-TRn, whereby i, j, m, n ≥ 0
◦ Channel collision occurs when a node inserts packet on a given shared wavelength while another packet is passing by
◦ Receiver collision (destination conflict) occurs when a node’s receiver is not tuned to wavelength of arriving packet
◦ Channel & receiver collisions can be mitigated or completely avoided at node architecture and/or medium access control (MAC) protocol level
Categorization◦ WDM ring networks can be categorized with
respect to applied MAC protocol
Slotted ring w/o channel inspection◦ Simple way to avoid channel & receiver collisions is
deployment of TDMA with static bandwidth assignment, whereby time is divided into slots equal to packet transmission time
◦ Typically, slots are of fixed size & arealigned across wave-length channels
◦ Well suited for uni-form regular medium to high traffic loads
◦ Low channel utiliz-ation under bursty & low traffic loads
MAWSON◦ Metropolitan area wavelength switched optical
network (MAWSON) is based on a FTW-FR or alternatively TT-FR node architecture
◦ N=W nodes connected to ring via passive OADMs using fiber Bragg gratings (FBGs) for dropping different home channel at each node => no receiver collisions
◦ With FTW-FR node structure, broadcasting & multicasting achieved by turning on multiple lasers simultaneously
◦ Channel access is arbitrated by deploying so-called Request/Allocation Protocol (RAP)
RAP◦ Time divided into fixed-size slots aligned across
all W wavelengths◦ Each slot further subdivided into header & data
fields◦ Slots dynamically assigned on demand by using
statically assigned N-1 TDMA Request/Allocation (R/A) minislots
◦ Each minislot comprises one request & one allocation field
RAP◦ Node i ready to send variable-size data packets to
node j uses request field of its assigned R/A minislot on j’s home channel to make a request
◦ After receiving request, node j allocates one or more data minislots to node i by using allocation field of its assigned R/A minislot on i’s home channel
◦ After one RTT, node i transmits data packet using allocated data minislots but no more than M data minislots
◦ Benefits of MAWSON & RAP Simple node architecture & protocol (e.g., no carrier sensing
capabilities required) save costs Due to in-band signaling, no separate control channel & control
transceivers needed Completely avoids channel & receiver collisions, achieves good
throughput & fairness, at expense of overhead & delay
Slotted rings w/ channel inspection◦ Most slotted WDM rings avoid channel collisions
by enabling nodes to check status (used/unused) of each slot
◦ Generally, this is achieved by tapping off some power from fiber & delaying slot while status is inspected
◦ Packet can be inserted in slot at unused wavelength
◦ Typically, node maintains separate VOQs, either for each destination or for each wavelength
◦ MAC protocol has to select appropriate VOQ according to given access strategy A priori access strategy
Node selects VOQ prior to inspecting slot status A posteriori access strategy
Node first checks status of slot & then selects VOQ
RINGO◦ Ring optical (RINGO) network uses FTW-FR node
structure◦ Each node has channel inspection capability built with
commercially available components◦ Nodes execute multichannel empty-slot MAC protocol
with a posteriori access strategy Number of wavelengths equal to number of nodes Each node has one FIFO VOQ for each wavelength In tie situations, longest among selected VOQs is chosen
◦ Single bit sufficient to identify status of a given slot => small overhead of empty-slot MAC protocol
◦ No separate control channel & control transceivers required
◦ Variable-size packets can be transmitted without segmentation & reassembly by deploying optical FDLs
SRR◦ Synchronous round robin (SRR) is another
empty-slot MAC protocol for unidirectional WDM ring with fixed-size time slots & destination stripping
◦ Each of the N nodes is equipped with one fixed-tuned receiver & one transmitter tunable across all wavelengths on a per-slot basis (TT-FR)
◦ Each node deploys N-1 separate FIFO VOQs, one for each destination
◦ SRR uses a priori access strategy
SRR node architecture
SRR protocol◦ In SRR, each node cyclically scans its VOQs in a round-robin
manner on a per-slot basis, looking for a packet to transmit◦ First (oldest) packet of selected VOQ is transmitted,
provided current slot was sensed empty◦ If selected VOQ is empty, first packet from longest queue of
remaining VOQs is sent◦ If current slot is occupied, next VOQ is selected for next
transmission attempt in subsequent slot according to round-robin scanning of SRR
◦ In doing so, SRR converges to round-robin TDMA under heavy uniform load conditions when all VOQs are nonempty
SRR performance◦ For uniform traffic, SRR asymptotically achieves a
bandwidth utilization of 100%◦ However, presence of unbalanced traffic leads to
wasted bandwidth due to nonzero probability that a priori access strategy selects wavelength channel whose slot is occupied while leaving free slots unused
◦ A posteriori access strategies avoid this drawback & achieve improved throughput-delay performance at expense of increased complexity
◦ Benefits of SRR Good performance requiring only local VOQ backlog information Destination stripping allows for spatial reuse & increased capacity,
but raises fairness control problems especially under nonuniform traffic
HORNET◦ Hybrid optoelectronic ring network (HORNET) is
unidirectional WDM ring using destination stripping◦ Nodes have a TT-FR structure◦ Similar to SRR, each node uses VOQs, one for each
wavelength, and both a priori & a posteriori access strategies can be deployed
◦ Nodes sense availability of each slot by monitoring subcarrier multiplexed (SCM) tones SCM-based carrier-sensing scheme is more cost-effective than
demultiplexing, separately monitoring, and subsequently multiplexing all wavelengths
Instead of wavelength demultiplexer, photodiode array, and wavelength multiplexer, HORNET channel inspection scheme needs only a single photodiode
CSMA/CA◦ Carrier sense multiple access with collision avoidance
(CSMA/CA) MAC protocols are used in HORNET First CSMA/CA protocol
Multiple different slot sizes according to predominant IP packet size distributions (e.g., 40-, 552-, and 1500-Byte long IP packets) circulate along the ring
Dedicated node controls size & number of slots Second CSMA/CA protocol
Unslotted A node begins to transmit a packet when a wavelength is sensed
idle Packet transmission is aborted when another packet arrives on
same wavelength Incomplete packet is marked by adding jamming signal Aborted transmission is resumed after backoff time
CSMA/CP◦ A more bandwidth-efficient modification of
unslotted CSMA/CA is the so-called carrier sense multiple access with collision preemption (CSMA/CP) MAC protocol In CSMA/CP, variable-size IP packets do not necessarily
have to be transmitted in one single attempt Variable-size IP packets are allowed to be transmitted &
received as fragments by simply interrupting packet transmission
Thus, successfully transmitted parts of original IP packet are not retransmitted => higher channel utilization
A posteriori buffer selection schemes◦ For an empty-slot protocol to be run on HORNET, certain
rules must be given to select buffer or packet whenever more than one wavelength channel carries an empty slot
◦ A posteriori selection processes are computationally more complex than a priori schemes
◦ Possible a posteriori VOQ selection strategies Random selection Longest queue selection Round-robin selection Maximum hop selection Channel-oriented TDMA (C-TDMA) selection
Each VOQ is allocated certain slot within a TDMA frame of size W (number of wavelengths)
◦ Random & round-robin buffer selection schemes provide satisfactory compromise between performance & complexity
FT-TR rings◦ Unidirectional empty-slot WDM ring may also use
source stripping and nodes with one fixed-tuned transmitter & one tunable receiver (FT-TR) => FT-TR rings At each node, packets are buffered in single FIFO transmit
queue In applied source stripping, a sender must not reuse the slot it
just marked empty => simple fairness mechanism that prevents node from starving entire network
However, destination stripping clearly outperforms source stripping in terms of throughput, delay, and packet dropping probability
Receiver collisions can be avoided in several ways Recirculating packets on ring until receiver is free Replacing TR with array of W fixed-tuned receivers Using optical switched delay lines at destination node
Slotted rings with control channel◦ In some slotted rings, status of slots is
transmitted on separate control channel (CC) wavelength
◦ To this end, each node is typically equipped with additional transmitter & receiver, both fixed tuned to CC wavelength
◦ Benefits of separate CC wavelength Enables nodes to exchange control information at high
line rates Eases implementation of enhanced access protocols
with fairness control & QoS support
Bidirectional HORNET◦ Original unidirectional TT-FR HORNET ring architecture
can be extended to slotted bidirectional ring whereby SCM is replaced with CC wavelength, one for each direction
◦ CC conveys wavelength availability information that allows nodes to “see” one slot into the future
◦ On each ring, every node deploys one fast tunable transmitter & one fixed-tuned receiver for data, and one transceiver fixed-tuned to CC => CC-FT2-TT2-FR4 system
◦ Benefits of CC-based bidirectional dual-fiber HORNET Preserves advantages of original unidirectional HORNET (e.g.,
scalability & cost-effectiveness) Provides improved fault tolerance against node/fiber failures &
survivability
◦ Bidirectional HORNET deploys so-called segmentation and reassembly on demand (SAR-OD) access protocol
SAR-OD◦ SAR-OD supports efficient transport of variable-
size packets by reducing number of segmentation & reassembly operations Packet transmission from a given VOQ starts in an empty
slot If packet is larger than a single slot, packet transmission
continues until it is complete or following slot is occupied Packet is segmented only if required to avoid channel
collision Segmented packet is marked incomplete Transmission of remaining packet segment(s) continues in
next empty slot(s) on corresponding wavelength SAR-OD reduces segmentation/reassembly overhead by
approximately 15% compared to approach where all packets larger than one slot are segmented irrespective of state of successive slots
Segmentation/reassembly◦ Segmentation & reassembly of variable-size packets can
be completely avoided in CC-based slotted WDM rings◦ To achieve this, each node of unidirectional HORNET
ring uses additional transmitter fixed-tuned to the node’s drop wavelength => CC-FT2-TT-FR2 system
◦ Additional transmitter used to forward dropped packets destined to downstream nodes sharing same drop wavelength
◦ Furthermore, each node is equipped with two VOQs for each wavelength, one for short packets & one for long packets
◦ Nodes deploy MAC protocol based on reservation frames
Reservation frames◦ Ring is subdivided into multiple reservation
frames with frame size equal to largest possible packet length
◦ In these frames, multiple consecutive slots are reserved to transmit long packets without segmentation
◦ Single reservation control packet containing all reservations circulates on CC
◦ Each node maintains table in which reservations of all nodes are stored
◦ When control packet passes, a node updates its table & is allowed to make a reservation
◦ Additional fixed-tuned transmitter forwards packets concurrently with transmitting long packets
◦ Short packets fitting into one slot are accommodated by means of immediate access of empty & unreserved slots
Wavelength stacking◦ Recall that wavelength stacking was used in PSR
networks◦ Wavelength stacking/unstacking allows a node to
simultaneously send & receive multiple packets in one slot using only one transceiver
◦ Wavelength stacking can be used to transmit multiple packets in one slot of CC-based slotted unidirectional WDM ring
◦ Wavelength stacking Time is divided into slots of duration Tp
Each node is equipped with one fast-tunable transmitter & one photodiode
Node starts transmission W time slots before its scheduled time slot, where W denotes number of wavelengths
Wavelength stacking
Virtual circles with DWADMs◦ In unidirectional slotted ring WDM networks, each
node may deploy a dynamic wavelength add-drop multiplexer (DWADM)
◦ As opposed to tunable transmitter & receiver, the input & output wavelengths of a DWADM must be the same
◦ As a consequence, a given node receiving on λ i must transmit on same wavelength λi => virtual circles
◦ DWADMs expected to be less expensive than tunable transceivers
◦ However, wavelength utilization expected to be smaller than in TT-TR systems where TT & TR can be tuned to any arbitrary wavelength independently
Virtual circles with DWADMs
Virtual circles with DWADMs◦ Virtual circles can be changed dynamically
according to varying traffic demands◦ Operation
W data wavelength channels & a separate TDMA control wavelength channel
Nodes exchange (1) transmission requests and (2) acknowledgments over control wavelength channel
W+1 wavelengths divided into three periodically recurring cycles In first cycle, a control packet sent by a server node
collects transmission requests from all nodes In second cycle, server node sends wavelength
assignments/acknowledgments back to nodes In third cycle, each node with assigned wavelength tunes
its DWADM appropriately & starts data transmission
Multitoken rings◦ Slotted WDM ring networks have several advantages
Easy synchronization of nodes even at high data rates High channel utilization Low access delay Simple access schemes
◦ However, variable-size packets are difficult to handle & explicit fairness control is needed
◦ In contrast, variable-size packets can be transported in reasonably fair manner in (asynchronous) token rings Access controlled by means of token circulating around the
ring Each node can hold token for a certain period of time during
which the node can send packets Due to limited token holding time fairness is achieved
MTIT◦ Multitoken interarrival time (MTIT) is a token-
based access protocol for source-stripping unidirectional WDM ring with CC-FTW+1-FRW+1 node structure
MTIT◦ CC used for access control & ring management◦ Channel access regulated by multitoken approach
Each channel is associated with one token that circulates among nodes on CC & regulates channel access
Token holding time controlled by target token interarrival time (TTIT) Token interarrival time (TIAT) defined as time elapsed between two
consecutive token arrivals at a given node Upon token arrival, node is allowed to hold token for TTIT – TIAT When token holding time is up, node must release token as soon as
current packet transmission is completed (or earlier if no more packets are left for transmission)
Node may simultaneously transmit on distinct channels if two or more tokens are concurrently held at node
MTIT◦ MTIT avoids receiver collisions & allows each node for
simultaneously using multiple data wavelength channels
◦ MTIT achieves low access delay due to the fact that a node may grab a token more frequently than in conventional token rings where a node has to wait for one RTT for the next token
◦ MTIT is able to self-adjust relative positions of tokens & maintain even distribution of them => low variance of token interarrival time & consistent channel access delay in support of high-priority traffic
◦ At the downside, capacity of MTIT expected to be smaller than that of destination-stripping ring networks
Meshed rings◦ In unidirectional source-stripping WDM rings,
capacity is limited by aggregate capacity of all wavelengths
◦ Capacity can be increased by means of destination stripping & resultant spatial wavelength reuse For uniform traffic, mean distance between source &
destination is half the ring circumference => two simultaneous transmissions on each wavelength => capacity 200% larger than that of unidirectional source-stripping rings
◦ In bidirectional rings with shortest path routing, mean distance between source & destination is one quarter of ring circumference => capacity increased by 400% on each directional ring compared to unidirectional source-stripping => total capacity increase of 800%
◦ Capacity further increased in so-called meshed rings
SMARTNet◦ Scalable multichannel adaptable ring terabit network
(SMARTNet) based on a bidirectional slotted ring network with shortest path routing & destination stripping
◦ Each node connected to both rings, for each deploying a FTW-FRW structure
◦ All wavelengths are divided into fixed-size slots whose length is equal to transmission time of fixed-size packet & header for indicating slot status
◦ Medium access governed by means of empty-slot protocol◦ In addition to N ring nodes, K equally spaced wavelength
routers, each with four pairs of input/output ports, are deployed to provide short-cut bidirectional links (chords)
◦ For uniform traffic, SMARTNet (K=6, M=2) increases capacity by 720% compared to unidirectional source-stripping rings
SMARTNet◦ Chords provide
short-cuts to the two M-th neighboring routers
◦ Routers r(k+M) mod K and r(k-M) mod K are said to be the M-th neighboring routers of router rk, where k = 0, 1, …, K-1
SMARTNet◦ Each wavelength router characterized by a wavelength
routing matrix that determines to which output port each wavelength from a given input port is routed
◦ Wavelength routing matrix chosen such that average distance between each source- destination pair is minimized with a minimum number of required wave lengths
Fairness control◦ In unidirectional WDM rings, each wavelength can
be considered a unidirectional bus terminating at a certain destination
◦ In an empty-slot access protocol, upstream nodes have a better-than-average chance to receive an empty slot while downstream nodes have a worse-than-average chance => starvation & fairness issues
MMR◦ Multi-MetaRing (MMR) fairness algorithm can be
superimposed to SRR in order to enforce fairness◦ MMR algorithm adapts a mechanism originally
proposed for MetaRing high-speed electronic MAN In MetaRing, fairness is achieved by circulation of control
message, termed SAT (standing for SATisfied) Nodes are assigned a quota/credit (maximum number of
packets) to be transmitted between two SAT visits SAT is delayed at each unSATisfied node until either the node’s
packet buffer is empty or number of permitted packet transmissions is achieved
Each SATisfied node forwards the SAT on the ring
MMR-SS vs. MMR-MS◦ MMR Single SAT (MMR-SS)
A single SAT regulates transmissions of all nodes on all wavelength channels
Each node can transmit at most K packets to each destination since the last SAT visit
Each SATisfied node forwards the SAT to the upstream node => SAT logically rotates in the opposite direction with respect to data (but physical propagation is co-directional)
◦ MMR Multiple SAT (MMR-MS) One SAT is used for each wavelength Similar to MMR-SS, each SAT circulates together with data
packets & is addressed to node upstream of node that emits the SAT
MMR-MS represents better extension of MetaRing fairness control scheme to WDM ring
M-ATMR◦ M-ATMR is an extension of asynchronous transfer mode
ring (ATMR) fairness protocol to WDM ring In M-ATMR, each node gets certain number of transmission credits for
each destination A node gets into inactive state when it has used all its credits or has
nothing to send For credit reset, each active node overwrites so-called busy address
field in header of every incoming slot with its own address A node receiving a slot with its own busy address assumes that all
other nodes are inactive Last active node generates a reset immediately after its own
transmission Reset causes all nodes to set their credits to predefined values
DQBR◦ Distributed queue bidirectional ring (DQBR) fairness
protocol is adaptation of DQDB for CC-based HORNET In each CC frame, request bit stream of length W follows the
wavelength-availability information A node receiving a packet in VOQ w notifies upstream nodes by
setting bit w in request bit stream in the CC that travels upstream with respect to packet direction
Upon reception, each upstream nodes increments request counter (RC) of wavelength w
Each time a packet arrives at VOQ w, the node stamps value in RC w onto packet & then clears RC w
Stamp is called wait counter (WC) After reaching the head of VOQ, packet must allow frame
availabilities on wavelength w to pass by as indicated by WC WC is decremented for each availability passing by node Packet can be transmitted when WC equals zero
QoS support◦ Many applications (e.g., multimedia traffic) require quality-
of-service (QoS) with respect to throughput, delay, and jitter◦ For QoS support, networks typically provide different service
classes such as CBR or VBR In general, traffic with stringent throughput, delay, and jitter requirements
is supported by means of circuit switching via resource reservation => guaranteed QoS
To efficiently provide QoS to bursty traffic, network nodes process & forward packets with different priorities while benefitting from statistical multiplexing => statistical QoS
SR3
◦ Synchronous round robin with reservations (SR3) is derived from SRR & MMR protocols
◦ SR3 allows nodes to reserve slots & thereby achieve stronger control on access delays In SR3, time is divided into successive reservation frames Each reservation frame comprises P SRR frames Each node can reserve at most one slot per destination per SRR frame SAT messages used to broadcast reservation information
Each SAT contains reservation field (SAT-RF) which is subdivided into N-1 subfields
Each subfield is assigned to a particular node for reservations
SR3
◦ If node i needs to reserve 1 ≤ h ≤ P slots per reservation frame on wavelength channel j, it waits until it receives j-SAT
◦ Node i then forwards reservation request by setting the i-th SAT-RF subfield to value h
◦ When node i receives j-SAT again, all network nodes are aware of the request of node i & reservation becomes effective
◦ Benefits of SR3 Guarantees throughput-fair access to each node Unreserved bandwidth can be shared by best-effort traffic For multiclass traffic, SR3 achieves very good separation of different traffic
classes
Connection-oriented QoS support◦ To enable connection-oriented QoS support in packet-
switched WDM ring for real-time services, ring is divided into so-called connection frames
◦ Real-time connections are established by reserving equally spaced slots within successive connection frames
◦ Best-effort traffic is supported by using unreserved & empty slots
◦ Pros & cons QoS approach is able to meet delay requirements almost deterministically However, it allows for reserving only one fixed-size slot (i.e., only fixed-size
packets are supported)
VOQ-based QoS support◦ In addition to W normal VOQs, each of the N ring nodes has W
real-time VOQs◦ Packets in real-time VOQs are transmitted via connections in
equally spaced reserved slots◦ On each wavelength, ring is subdivided into frames each
consisting of N/W slots, one per destination receiving on that wavelength
◦ A single reservation slot carries connection set-up field & connection termination field, each consisting of N bits sent on a subcarrier
◦ Connection set-up & termination fields are used by a given node to make & release reservations, respectively
◦ Each node keeps track of reservations by maintaining table that is updated when reservation slot passes node
MTIT – QoS with lightpaths◦ MTIT protocol can be extended to support not only packet
switching but also circuit switching with guaranteed QoS Solution allows for all-optical transmission of packets with source
stripping & circuits via tell-and-go establishment of point-to-point lightpaths with destination stripping
In the latter case, on-off switches at both source & destination nodes of corresponding lightpath are set in off state => spatial wavelength reuse
For all active lightpaths, each node maintains so-called local lightpath table (LLT) that is updated when token passes
So-called token lightpath table (TLT) is sent in each token to broadcast changes of lightpath deployment on wavelength associated with token
Each token has add & delete lists for lightpath set-up & tear-down A source node holding a token sets up & tears down a lightpath by
making an entry in the add list & delete list, respectively