CA2310855A1 - Multiservice optical switch - Google Patents

Abstract

A method and apparatus for building a multiservice switch using a time-slotted optical switch fabric are disclosed. Service-specific line cards transfer blocks of information across the optical switch by transmitting fixed-duration bursts of optical signals. A
modular method is disclosed for expanding the number of cards that can be supported by a time-slotted N input by N output optical switch. Apparatus and methods for building said optical switches using electrooptic wafer beam deflector components are disclosed.
SONET, ATM, and packet transfer services are supported by the service specific line cards.

Description

Multiservice Optical Switch A met d apparatus for building a multiserv' Itch using a time-slotted optical switch fabric disclosed. Service-s~i~'c line cards transfer blocks of information across the optical s h by tr lttmg fixed-duration bursts of optical signals. A
modular method is disc expanding the number of cards that can be supported by a time-slotted N ' ut by N output op ' witch. Apparatus and methods for building said optic~svltches using electrooptic wafer be eflector components are disclosed.
SON-E'I', ATM, and packet transfer services are supporte the service specific line , cards.
Field of Invention The present invention in general relates to high-speed digital crossconnects, packet switches and routers used in telecommunication and computer networks to switch and route information arriving in one or a plurality of input ports to one or a plurality of output ports.
In switches and routers data packets arriving from input ports are buffered and after routing and classification, that is done by processing the packet headers, the packets are queued and scheduled for transfer to individual output ports. The data packets must traverse the switching fabric of the switch or router in order to be be transferred from the input port to the output port. Electronic switching fabrics are becoming a bottleneck as optical fiber transmission increase their transmission by carrying multiple signals of increasing speeds at different wavelengths. A similar problem occurs with digital crossconnects where streams of frame-based information is exchanged between optical fibers.
More specifically, the present invention relates to the construction of high-speed switches, routers and digital crossconnects that can achieve very high aggregate throughput and port count through the use of a time-slotted optical switching fabric.
Discussion of Previous Art Optical transmission technologies have increased dramatically the information-carrying capacity of a single optical fiber. Using wavelength division multiplexing, it is now possible to transmit in a single fiber over 100 wavelengths, each carrying 10 Gigabits per second for an aggregate rate of more than 1 Terabit per second. The purpose of a switch, is to transfer information among attached optical fiber transmission systems.
Future switches must therefore be able to transfer aggregate information rates in the many Terabits per second. Switches that can handle these information rates are extremely difficult to build because of the limited information-carrying capacity of electronic systems [Turner 1998].

Optical switches that transfer information in optical form can avoid the bottleneck inherent in electronic switches. Optical switching fabrics have been introduced that can provide an end-to-end connection across a network using a single wavelength.
Such optical fabrics operate at a relatively high level of granularity, that is, they handle the information stream carried by an entire wavelength. Here we are concerned with optical switches that can transfer optical signals at smaller levels of granularity, for example, 1 microsecond bursts of optical signal. Such switches have been disclosed in Patent Application # xyz, "Time-Slotted Optical Space Switch," by Leon-Garcia et al.
In time-slotted switches, the optical signals that arrive at each input port consist of a sequence of signal bursts, of some fixed duration called a "time slot". Each burst at an input port must be routed independently across the switch. A space switch that can be reconfigured rapidly can be used to build a time slot switch. At periodically recurring time instants, the switch is reconfigured so that it connects input ports to output ports according to some desired permutation arrangement. No valid signal information is allowed into the switch during the reconfiguration time, so a guaxd time is used to separate the bursts of optical signal that traverse the switch. The time slots that arrive at different input ports need to be synchronized so that they enter the switch at the same time, right after the end of the guard time. Time-slotted switches allow the transfer of optical information from any input port to any output port over relatively short time intervals, in the order of microseconds, and in effect support the simultaneous transfer of information among all ports attached to the optical switch.
Time-slotted optical switches are ideal as switching fabrics for high-speed digital crossconnects, routers, and switches. The line cards that are attached to the optical fabric exchange information through the exchange of bursts of optical signals. The transfer of information in optical form allows extremely high volumes of information to be handled in a limited amount of space and at very high speeds. Indeed the transmission capacity of optical switches is so high, that a single electronic line card can only utilize a small fraction of the available capacity. In the present invention we propose the use of WDM
in the optical switch fabric to utilize the large transmission capacity of the switch.
Summary of Invention The present invention discloses methods and apparatus for a multiservice switch in which service-specific line cards exchange information using a common format across a time-slotted optical switch fabric that can provide circuit or on-demand transfer of bursts of optical signals from input ports to output ports. Services that can be supported by the switch include SONET frame crossconnect, ATM cell switching, and IP packet transfer.
The invention is based on a time-slotted space switch disclosed in Patent Application #
xyz, "Time-Slotted Optical Space Switch," by Leon-Garcia et al. The present invention discloses methods for building a hierarchical switch structure in which WDM is used to expand the number of line cards that can be supported by the optical switch fabric and thus to exploit the high information transfer capability of the fabric.

2 Brief Description of the Dra~rings Figure 1 Time-slotted space switch Figure 2 Timing in time-slotted space switch operation Figure 3 TDM sequence of interconnection matrices & switch configurations Figure 4 Sequence of interconnection matrices for unbalanced and nonuniform traffic matrix Figure 5 On-demand operation Figure 6 Switch with port expansion Figure 7 Hierarchical switch with port expansion Figure 8 A megapacket structure Figure 9 Megapacket queueing Figure 10 Stream transfer from line card to port card Figure 11 Sub-Megapacket transfer from line card to port card Detailed Description of the Invention Figure 1 shows an NxN time-slotted optical switch. The switch accepts burst of optical signals at specified instants of time and transfers said bursts to desired distinct output ports. The optical switch has an associated switch fabric control unit that accepts the requests for a given set of transfers of bursts of optical signals from given inputs to specific outputs. The switch fabric control executes an algorithm that is specific to the particular internal structure of the optical switch in order to determine what internal configuration is required to realize the requested set of connections. The switch fabric control then distributes digital control signals to the units in the switch specifying their internal configuration. The desired set of connections becomes available when the modules complete the reconfiguration specified by the control signals.
The operation of said NxN switches must operate in cycles of the form shown in Figure 2. Prior to the beginning of each reconfiguration interval, the fabric switch control determines the internal configuration of the switch fabric that is required to achieve the desired transfer. During the reconfiguration interval, the control signals are distributed to the internal switch components which are then reconfigured as directed. At the end of the reconfiguration interval, the NxN space switch is ready to transfer the bursts of optical signals.
The configuration of said NxN switch during cycle t is specified by a matrix P(t) _ ~p;~(t)}, where p;~(t) is equal to 1 if input i is connected to output j, and is equal to zero otherwise. P(t) has the property that each row has exactly one non-zero value, and each column has exactly one non-zero value. The sequence P( 1 ), P(2), P(3), P(4), . . .
represent the sequence of interconnection patterns provided by the NxN switch.
The number of times the ijth component equals 1 in the sequence P(1), P(2), ..., P(k) represents the number of time slots allocated to connection ij in k consecutive cycles.
Hence the allocation of transmission opportunities ("bandwidth") among input-output pairs is determined by the sequence of configuration matrices.
The sequence of interconnection patterns P(1), P(2), P(3), P(4), ... can be selected to meet the bandwidth requirements of the traffic that traverses the switch. In the case where the same level of traffic flows between every input and output port and where the traffic flows are relatively steady, a suitable sequence consists of a repetitive interconnection pattern P(1), P(2), ..., P(N-1), P(N), P(1), P(2), ..., P(N-1), P(N), ...
that provides every input-output pair with 1 transmission opportunity per repetition cycle.
We refer to said repetitive pattern as a Time-Division Multiplexing (TDM) schedule.
Figure 3shows an example of such a repetitive pattern for a 4x4 switch: A
repetitive pattern of 4 permutation matrices and the associated switch configurations are shown.
In the case where different levels of traffic flow between different input and output ports and where the traffic flows are relatively steady, a suitable schedule consists of a repetitive interconnection pattern P(1), P(2), ..., P(N-1), P(N), P(1), P(2), ..., P(N-1), P(N), . . . that provides an input-output pair with a number of transmission opportunities per repetition cycle that is proportional to the relative traffic flow of the input-output pair.
Figure 4 shows an example of a traffic matrix for a 4x4 switch and a corresponding repetitive interconnection pattern that satisfies the traffic demand. The ijth entry in the traffic matrix is the proportion of time information is available for transfer from input port i to output port j. The "x" in the permutation matrices in the figure denote "don't cares" for connections in the switch that have not been assigned. Various algorithms are available for synthesizing an repetitive interconnection pattern for a given traffic matrix [Algoxxx].
The interconnnection pattern can be modified over time to track variations in traffic levels and to deal with temporary surges in traffic. By keeping a running average of the traffic flow between each input-output pair, the variation in the traffic matrix can be tracked and adjustments in the interconnection pattern can be made. These adjustments may consist of small changes in the permutation matrices or in the repetitive pattern itself through the addition or deletion of one or more permutation matrices. Surges in traffic can be monitored through the backlog of information at the input to the switch. "Don't cares" in the permutation matrices can be set to help reduce the backlog for certain input-output pairs.
The interconnection pattern of the time-slotted optical can also be computed dynamically according to a protocol where the fabric control accepts requests for packet transfers from the line cards and the executes an algorithm which determines which line cards are to be granted permission to transmit in the next cycle. Algorithms to arbitrate among competing requests from line cards have been studied extensively in the context of input-buffered switches [McKeown 199x]. For time-slots of duration in the order of microseconds, real-time implementations of the request/grant algorithms are possible for moderate size switches.
A combination of pre-allocated and on-demand assignment of transmission opportunities is also possible. A repetitive pattern of the form P(1), P(2), ..., P(N-1), P(N), P(1), P(2), ..., P(N-1), P(N), ... can be used where a subset of cycles are pre-allocated and where certain cycles are designated for request/grant operation. The processing load associated with the real-time operation of the request/grant algorithm is lessened by spacing the request/grant cycles evenly in the repetition pattern.
Figure 6 shows the use of WDM multiplexers and demultiplexers to concentrate multiple optical signals that occupy non-overlapping wavelengths into a single optical signal that can be switched across the NxN optical switch. The structure of the switch constrains all components of the composite signal to be switched to the same output port.
Each additional wavelength in the composite signal increases the transmission-carrying capability (measured in bits) in each time-slot. The transmission-carrying capability of the overall switch increases accordingly.
Time-Slotted Optical Switch & Line Cards TDM operation On-demand operation Hybrid operation Hierarchical Switch Port Expander Individual wavelengths Impact on Scheduling Megapackets Stream transfer from line card to port card Megapkt from Port to Port Megapkt from Port to Line Card Megapkt in the line card Megapkt to Port Megapkt to Line Card VL and composable megapackets Service-specific Line Cards Label-switched packets ATM cells SONET frames References [Turner 1998] J. Turner and N. Yamanaka, "Architectural Choices in Large Scale ATM
Switches," IEICE Trans. Commun. Vol. E81-B, No. 2, February 1998.

Claims