WO2014085972A1 - 基本自路由单元及其半清器、排序单元、网络集线器和组播交换网络构建方法 - Google Patents

基本自路由单元及其半清器、排序单元、网络集线器和组播交换网络构建方法 Download PDF

Info

Publication number
WO2014085972A1
WO2014085972A1 PCT/CN2012/085793 CN2012085793W WO2014085972A1 WO 2014085972 A1 WO2014085972 A1 WO 2014085972A1 CN 2012085793 W CN2012085793 W CN 2012085793W WO 2014085972 A1 WO2014085972 A1 WO 2014085972A1
Authority
WO
WIPO (PCT)
Prior art keywords
output
routing
input
network
multicast
Prior art date
Application number
PCT/CN2012/085793
Other languages
English (en)
French (fr)
Inventor
李挥
李硕彦
尘福兴
朱鸷璞
崔凯
Original Assignee
北京大学深圳研究生院
深圳市利德嘉实业有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 北京大学深圳研究生院, 深圳市利德嘉实业有限公司 filed Critical 北京大学深圳研究生院
Priority to PCT/CN2012/085793 priority Critical patent/WO2014085972A1/zh
Priority to CN201280077284.3A priority patent/CN105556904A/zh
Publication of WO2014085972A1 publication Critical patent/WO2014085972A1/zh
Priority to US14/711,802 priority patent/US20150319080A1/en

Links

Classifications

    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04LTRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L45/00Routing or path finding of packets in data switching networks
    • H04L45/16Multipoint routing
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04LTRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L45/00Routing or path finding of packets in data switching networks
    • H04L45/54Organization of routing tables
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04LTRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L47/00Traffic control in data switching networks
    • H04L47/10Flow control; Congestion control
    • H04L47/12Avoiding congestion; Recovering from congestion

Definitions

  • the present invention relates to communication networks, and more particularly to a basic self-routing unit and its semi-clearer, sorting unit, network hub, and multicast switching network construction method.
  • Network video services are one of the most used services.
  • Video streams have typical characteristics, and their data content tends to be single-point transmission, multi-point reception, that is, multicast (also called multicast).
  • Traffic for today's popular video-based applications such as IPTV, video conferencing, interactive emulation, and multi-party gaming has multicast features. Therefore, in order to improve network performance and effective utilization, multicast communication technology is particularly important.
  • multicast communication technology is particularly important.
  • multiple unicast transmission methods were often used. This multi-unicast strategy not only aggravated the workload of the server, but also increased the traffic in the network.
  • IP multicast technology a forwarding tree of the sender to the receiver is established through the multicast protocol.
  • the sender only needs to send a data content, and the data can be copied through the intermediate routing switch node, and finally forwarded to All recipients can radically reduce the workload of the server and the load on the network.
  • Deering proposed a multicast structure that incorporates multicast functionality into the IP layer of the data network.
  • the multicast experiment network - Mbone was established.
  • the IP multicast protocol standard and deployment method were studied in Mbone.
  • the inter-domain routing protocol system experiment was conducted in two backbone networks (vBNS, Abilene) of Internet2.
  • the implementation of multicast requires the support of the application layer, transport layer, network layer, and data link layer in the network. From the physical location of the implementation, it can be divided into terminals and intermediate routing nodes. From the perspective of implementation, it can be divided into multiple unicast software scheduling and hardware circuit line-speed fan-out copying. But overall, the terminal basically implements various application layers, transmission control layer, and network layer protocol in software mode. Usually, there is no bottle header; the implementation of the routing switching node is the key to multicast performance. First, because the packet may pass through several The routing node of the hop (such as 5 hops, 8 hops) only passes through the one-hop terminal. The second is that the routing node needs to handle the contention of many ports, which is the bottleneck of the network itself.
  • the multicast implementation of the routing switching node can be divided into multiple unicast software scheduling methods and hardware circuit line rate fan-out copying methods.
  • Literature Liu Ying, Xu Wei; “Intemet Multicast Architecture", Science Press, April 2008] Given that the majority of multicasts on the routing switching node are multi-unicast soft multicast, that is, before routing After the multicast packets are copied, they are queued in the source port queue, and then the software is used to implement port pairing in the respective scheduling mode to complete the packet transmission.
  • Cisco Wideilliam R.
  • Figure 1 shows the Knockout switch fabric, Knockout, the sporting knockout.
  • the proposed structure comes from an obvious fact: In the same time slot, the probability of all input cells reaching the same destination port is very small.
  • Figure 1 shows a prototype of a switch fabric based on this theory. Each input port is connected to all of the output interface modules in a fully connected manner. At each output interface, through the address filter, the cell whose destination address is not the output module is filtered out, and then through an N:L knockout hub, the number of cells allowed to the same destination port in the same time slot is The value L, the cell beyond this value is discarded.
  • the overall logic is divided into MGN1 and MGN2.
  • Two levels due to the bus type connection between each level, and the structure of the MGN module CrossBar Point, this structure also has multicast characteristics.
  • cell replication is performed at the appropriate location for each level, and finally reaches the specified destination port.
  • the cell uses self-routing to reach the desired output group.
  • the structure of Figs. 2 and 3 although the cell expansion is effectively reduced by the cluster expansion method, the structure is based on the device complexity of the CrossBar structure ⁇ ( ⁇ 2 ), and the inter-stage bus type.
  • the complete connection has to face the limit of the drive capability of the line device, which makes the large-scale expansion of this type of solution bottleneck.
  • the previous solution is implemented based on the switching structure with its own multicast characteristics.
  • the multi-level interconnection network such as Banyan, since it generally does not have multicast characteristics, most of them use the multicast replication method to achieve Multicast exchange, [TT Lee; "Nonblocking copy networks for multicast packet switching", IEEE Journal on Selected Areas in Communications, vol. 6, no. 9, pp. 1455-1467, Dec. 1988] Copy scheme, [N. McKeown; "The iSLIP scheduling algorithm for input-queued switches," lEEE/ACMTransactions on Networking, vol. 7, no. 2, pp. 188-201, April 1999] using cell feedback
  • the method implements the replication of multicast cells.
  • the scheme adopts a multicast strategy of pre-routing replication, and the key is the design of the pre-replication network.
  • the replication network is used to copy the cells that need to be multicasted to the number of fanouts (the number of destination addresses), and then the final cell exchange is completed by the peer-to-peer unicast switching structure.
  • Figure 4 is an internal representation of Tony Lee [TTLee; "Nonblocking copy networks for multicast packet switching", IEEE Journal on Selected Areas in Communications, vol. 6, no. 9, pp. 1455-1467, Dec. 1988].
  • a non-blocking replication network design is an internal representation of Tony Lee [TTLee; "Nonblocking copy networks for multicast packet switching", IEEE Journal on Selected Areas in Communications, vol. 6, no. 9, pp. 1455-1467, Dec. 1988].
  • the concentrator network and the broadcast banyan network are composed of three parts.
  • the accumulation and network are used to calculate the number of cells that need to be copied. As shown in Figure 4, at the input of the accumulation and network, each cell identifies its fanout size. After the fanout of each input port is calculated by the accumulation and the network, the number of required cells to be accumulated for each port of the time slot can be obtained, because the total number of duplicated packets required in the same time slot may be larger than the output port. The number is large, so it is necessary to reject the port whose first accumulated sum is greater than the number of output ports, and the cells of all ports after the port.
  • the multicast network preferably provides internal non-blocking, self-routing and fixed delay characteristics for the input cells
  • the structure has the following problems: Accumulation and network and virtual address encoders are not suitable for large-scale expansion; when the sum of input packet replication requirements exceeds the number of output ports, an overflow phenomenon occurs, if the replication network does not Providing a retransmission mechanism, packets exceeding the number of ports are discarded, and in the replication network, the probability of being lost due to overflow is increased from top to bottom, resulting in unfair service; the structure performs multicast address translation.
  • the required cache overhead is very large, as pointed out in Turner, JS; "A practical version of Lee's multicast switch architecture", IEEE Transactions on Communications, vol. 41, pp.
  • Each lookup table has a size of 256Mb and a large cache overhead. Therefore, its structure is too complicated, and the component complexity of the hub network and the broadcast eucalyptus network are both 0 (W ⁇ 2 N), which is equivalent to the general self-routing network complexity.
  • Figure 5 shows Sunshine [N. McKeown; "The iSLIP scheduling algorithm for input-queued switches," IEEE/ ACM Transactions on Networking, vol. 7, no. 2, pp. 188-201, April 1999] is another The class is based on the multicast switching structure of the Banyan network. Its overall structural features are similar to those of Starlite swit, and are based on the basic framework of the Batcher-Banyan network. It introduces K parallel parallel Banyan networks to enhance the robustness of the entire structure and enhance the characteristics of anti-burst traffic. In addition, it also introduces feedback to increase multicast capability. This is a scheme of multicast replication for cell feedback. .
  • Each input of the network has an input port controller, but when a multicast cell is found, a special tag enables the cell to be selected by the selector before being sent to the parallel Banyan network. Feedback is returned to the input until all destination addresses of the multicast cell are sent.
  • the complexity of Batcher network 0 (N log 2 N) in the above SunShine structure is itself an extension bottleneck of the whole structure.
  • the scheme of implementing multicast replication through feedback avoids the requirement of copying network, but there is output chaos. In order, the problem of multicast cell delay is increased.
  • Clos Due to Clos [C Clos, "A study of non-blocking switching networks", Bell System Technical Journal, 1953] the historical position of the three-level interconnection network in circuit switching, and its superior non-blocking situation is superior to the traditional CrossBar structure. Component complexity ( Clos is ⁇ ( ⁇ 15 )), which attracted many scholars to apply Clos to packet switching. The difficulty in applying the Clos structure for packet switching is how to perform efficient routing to complete the exchange of cells. This problem can be equivalently converted into the decomposition of the rate matrix or the minimum edge coloring of the bipartite graph [HJ Chao, Bin Liu; Performance Switches and Routers, John Wiley & Sons, April 2007] (P383-387).
  • FIG. 6 presents a multicast based on the structure of Growable packet switch (GPS) [DJ Marchok, CE Rohrs, RM Schafer; "Multicasting in a Growable Packet (ATM) Switch,"INFOCOM'91, pp. 850-858, 1991] Scheduling Algorithm.
  • GPS is similar to the three-level structure of Clos, but the first two levels are non-cached self-routing interconnect networks, and the last level module is a small-scale output switching module. In this type of structure, routing is mainly achieved by specifying the first two levels. Implemented by routing.
  • the technical problem to be solved by the present invention is to provide a multicast implementation technology that facilitates large-scale expansion and does not have a network bottleneck, which is a basic self-routing unit and the above-mentioned shortcomings of the prior art that cannot be expanded on a large scale and have a bottle head. Its semi-cleaner, sorting unit, network hub and multicast switching network construction method.
  • the technical solution adopted by the present invention to solve the technical problem is: construct a basic self-routing unit for multicast, the basic self-routing unit includes two input ports and two output ports, and the input port includes the An input port and a second input port, wherein the output port is a first output port and a second output port, respectively; the signal input by the input port includes a routing signal plus a data attribute and a data content; the routing signal has The routing signal includes a bi-cast signal, a unicast signal, and an idle signal; when the routing signals of the two input ports are a dual-cast signal and an idle signal, respectively, the routing signal is a bi-cast signal.
  • the input port is respectively connected to the first output port and the second output port, the routing signal value outputted by the first output port is a Boolean product of two input routing signals, and the routing signal value output by the second output port is A Boolean sum of two input routing signals. Further, when the routing signals input by the two input ports all point to one of the two output ports, the two input ports are contiguous according to the data priority in the data attribute of the input signal.
  • An output port wherein the two input ports are connected to the output port with a high priority, and the data having a lower priority is connected to the other output port; when the routing signals input by the two input ports are respectively When pointing to different output ports, the input ports are respectively connected to the output port in a cross or parallel connection;
  • the cross connection includes a first input port and a second output port, the second input port and the first output a port connection;
  • the parallel connection includes the first input port being connected to the first output port, and the second input port being connected to the second output port.
  • the routing signal having an algebraic lattice structure comprises constructing a self-routing in-band routing signal table using an algebraic lattice, the algebraic lattice being a distribution lattice.
  • the invention further relates to a semi-clearer comprising the above-mentioned basic self-routing unit, wherein the k 2 sequentially arranged 2 ⁇ 2 tuned sorting units, the 2 ⁇ 2 dual-tone sorting unit comprises two input ends and respectively output
  • the input signal has a small median value to a low output terminal and a larger value to a high output terminal; an output of the nth of the k binomial sorting units is an nth input of the half clearer
  • the other output is the k+n input of the semi-clearer; the low output of the nth of the k dual-tone sorting units is the nth output of the semi-clearer
  • the high output end is the k+n output end of the semi-clearer; the first output end to the kth output end of the semi-clearer output a bi-level sequence a1, the semi-clearer
  • the k+1th output to the 2kth output output a 3 ⁇ 4 modulation sequence a2,
  • the semi-clearer is the basic self-routing unit.
  • the device includes a plurality of 2 X 2 double-tone sorting units, and the output ends of each of the 2 X 2 double-tone sorting units are respectively connected to the input ends of different 2 X 2 double-tone sorting units included in the next stage of the semi-finalizer or Different semi-clearer inputs in the next stage.
  • the invention further relates to a network hub comprising the above self-routing unit, the network hub comprising 2G inputs and 2G outputs, the network hub having 2 GXG arbitrary binary sorting units and the 2
  • the output port is a 1-output group, and the G minimum ordered output ports of the semi-clearer are 0-output groups.
  • an address filtering unit is also serially connected to each output end of the network hub.
  • the invention also relates to a method for constructing a multicast switching network, comprising the following steps:
  • the basic self-routing unit and the semi-clearer, sorting unit, network hub and multicast switching network construction method embodying the invention have the following beneficial effects: since the basic self-routing unit outputs the output signal at the output end respectively The Boolean sum of the input signal and the Boolean product, so that it can not only achieve unicast, but also can sort or filter the signal during multicast; at the same time, the basic unit can also obtain the semi-clearer, the sorting unit and the network hub. In turn, the routing exchange function of the multicast network is implemented. Therefore, it is easy to expand on a large scale and there is no network bottleneck. DRAWINGS
  • FIG. 1 is a schematic structural diagram of a Knockout switching structure in the prior art
  • FIG. 2 is a schematic diagram of a basic switching structure based on MOBAS in the prior art
  • FIG. 3 is a schematic structural view of the MNG module of Figure 2;
  • FIG. 4 is a schematic structural diagram of a Tony Lee replication network in the prior art
  • FIG. 5 is a schematic diagram of another type of multicast switching structure based on Banyan network in the prior art
  • FIG. 6 is a schematic diagram of a Growable packet switch multicast switching structure in the prior art
  • FIG. 8 is a partial state of a basic self-routing unit in a multicast state according to an embodiment of the present invention.
  • FIG. 9 is a state diagram of a distribution grid as a multicast signal table in an embodiment of the present invention.
  • FIG. 10 is a schematic diagram of a sorting network formed by a basic self-routing unit in an embodiment of the present invention
  • FIG. 11 is a schematic structural diagram of a dual tone sorting unit in an embodiment of the present invention.
  • FIG. 12 is a schematic structural diagram of a k-half cleaner according to an embodiment of the present invention.
  • Figure 13 is a schematic diagram showing the state enumeration of the k-half cleaner in the embodiment of the present invention.
  • FIG. 14 is a schematic structural diagram of a bi-directional sorting unit in an embodiment of the present invention.
  • 15 is a schematic structural diagram of an arbitrary binary sequence sorting unit in an embodiment of the present invention.
  • 16 is a schematic structural diagram of a network hub in an embodiment of the present invention
  • 17 is a schematic diagram showing a state of a distribution grid and a non-distributable grid in the embodiment of the present invention
  • FIG. 18 is a schematic diagram of a signal transmission process of an actual network hub in an embodiment of the present invention.
  • Figure 19 is a diagram showing the structure of a network constructed using a network hub in an embodiment of the present invention.
  • FIG. 7 is a fully distributed self-routing model based on group theory multi-level interconnection network [Wei He, Hui Li, Bin-qiang Wang, "Load-Balanced Multipath Self-routing Switching Structure by Concentrators", IEEE Proc. of ICC2008, May, 2008] uses a class called Divide & Conque [S.-Y. .
  • a 2x2 basic self-routing sorting element's data packet sorting exchange process can be implemented using in-band signaling, such as adding two-bit in-band signaling in front of the data packet.
  • the first bit indicates whether there is a data packet in the current time slot, 1 indicates that there is an active valid data packet, 0 indicates a null packet with no valid data currently; when 1 is 1, the second bit 0 indicates the output destination port address of the packet. It is 0 or 1, and it is meaningless when it is 0. Therefore, it is equal to ' ⁇ ', ' ⁇ ', '00' respectively represent valid data packets with output port target 0, valid data packets with output port target of 1, and empty packets.
  • a 2x2 network infrastructure component can implement self-routing according to the linear ordering relationship as follows: 10 ⁇ 00 ⁇ 11, unicast 2x2 basic self-routing component with 2-bit in-band signaling self-routing
  • the control method can be seen in the following table:
  • the component is based on fast knockout [S.-YR Li, Hui Li (Li Hui); "Optimization in the fast knockout algorithm for self-route concentration", IEEE Proceedings of ICC'98, pp. 630-634, Atlanta , Jime, 1998] Hub Construction Algorithm or Bitonic Circular [S.-YRLi (Li Shuoyan); “Algebraic Switching Theory and Broadband Applications", Academic Press, 2001] Half-cleaner [Hui] Li (Li Hui), S.-YR Li (Li Shuoyan); "On the Complexity of Concentrators and Multi-stage Interconnection Networks in Switching System", The Chinese University of Hong Kong, 2000.1 1] , can constitute any 2G-to- G's self-routing group hub.
  • multicast in order to achieve complete integration of multicast and unicast, multicast is required to construct a basic multicast circuit unit by fan-out copying the target at the physical data link layer.
  • the ideal structure of the switch fabric is that its scale can be recursively expanded arbitrarily. There is no bottleneck, that is, the most basic 2x2 component constructs a large-scale switching system that meets the performance index at the lowest cost through various topologies. .
  • the unicast 2x2 basic self-routing component described above does not have the properties required for multicast.
  • a new 2x2 basic self-routing component that satisfies the multicast situation that is, a basic 2x2 sorting unit, is constructed, and a large switching system is obtained by the unit.
  • the routing unit means that there is a definition of packet routing: A valid packet contains the routing signal plus the subsequent load, and the value of this signal is not 0-bound or l-bound, depending on its own target output. At the same time, the null value of the signal is represented as idle.
  • This extended set Q bicast is sorted by the respective parts of the rule 'J 0-bound ⁇ idle ⁇ l-bound and the rule J3 ⁇ 4'J 0-bound ⁇ bicast ⁇ l-bound.
  • the multicast unit performs sorting on the signals belonging to the set Q bieast , but when /d/e encounters b/o ⁇ , the exchange control follows the following rules: output-0 port output signal value 0-bound, output-1 port output signal value 1-bou nd , followed by the same load.
  • the load of the packet can be a through state when passing through the multicast unit, and the cross state or a multicast mode in which one input connects the two outputs.
  • the multicast unit forwards as much of the packet load as possible to its intended destination port.
  • a lattice is a collection of binary operands Boolean sum (“V”) and Boolean product (“ ⁇ ”), and adheres to the following properties:
  • a common lattice is a set of subsets that follow a set of union and intersection operations in a Boolean operation.
  • a ⁇ b aAb a
  • each part can be derived from a partially ordered sequence. For example, if there is a maximum lower bound and a minimum upper bound on any two elements in a partially ordered set (referred to as a partial ordered set), the partial ordered set is called a partial ordered lattice. These two bounds are represented by a Boolean sum and a Boolean product.
  • the partial order lattice has the characteristics of a lattice.
  • each set of partial sets defined by the lattice is naturally a partial order lattice. It seems that the partial order lattice is logically equivalent to the lattice.
  • Figures 8a-8h depict partial ordering of Q bicasts , which together specify the connection of smaller elements at a higher position and larger elements at a lower position. It is obvious that the partial order set Q bicast has the characteristics of a partial order lattice and is therefore called a lattice.
  • the three rules (two collations and one exchange rule) that the multicast unit runs can be unified into the following Boolean rules: that is, the value of the 0 output port is the Boolean product of the two inputs, and the value of the 1 output port is two. Enter the boolean sum.
  • aA(bvc) (aAb)v(aAc) ,
  • this cell is a distribution grid.
  • Q bicast is an example of a single ticket, as shown in Figure 9, which shows a partial set of larger elements connected to the high end and a larger element at the lower end.
  • the partial order set Q bkast has the characteristics of a lattice partial order set and can therefore be regarded as a lattice.
  • the multicast unit is sometimes referred to as a Boolean unit.
  • a basic self-routing unit for multicast is constructed, which includes two input ports and two output ports, and the input ports thereof include a first input port and a second input port.
  • the output ports thereof are the first output port and the second output port (the above 0 output port and 1 output port);
  • the input port input signal includes the routing signal plus the following a data attribute (the data attribute herein includes a data priority field indicating a priority of the data) and a data content;
  • the routing signal has an algebraic structure;
  • the routing signal includes a bi-cast signal, a unicast signal, and An idle signal; when the routing signals of the two input ports are respectively a dual-cast signal and an idle signal, the routing signal is an input port of the dual-cast signal and is respectively connected to the first output port and the second output port,
  • the routing signal value outputted by the first output port is a Boolean product of two input routing signals
  • the routing signal value output by the second output port is A Boolean sum of two input routing signals.
  • the routing signal has an algebraic lattice structure comprising constructing a self-routing in-band routing signal table using an algebraic lattice, the algebraic lattice being a distribution lattice.
  • the hierarchical relationship with the following algebraic structure for sorting is easily obtained by the above description: ordered set ⁇ assigned lattice ⁇ lattice ⁇ partial ordered set.
  • a distribution grid is used to construct a self-routing in-band signal table for multicast sequencing and multicast switching. So far, common switch fabrics use linear ordered sets, especially when the signal value is a digital representation.
  • the conjunction paradigm is based on all five axioms that support the distribution grid.
  • the generalized ordering 0-1 principle is introduced in the literature [S.-Y. R ⁇ i (Li Shuoyan); "Algebraic Switching Theory and Broadband Applications", Academic Press, 2001], which requires the same axioms of the same paradigm.
  • the signal table is usually constructed as a partial ordered set structure.
  • mathematical tools derived from the axioms that govern the partial order or even the partial order lattice are not sufficient for multicast exchange.
  • Figure 10 clearly illustrates the importance of the distribution grid, which shows a 4x4 three-level Boolean sorting network consisting of a 2x2 Boolean cell plus a 1x1 delay between stages.
  • the four input values a, b, c, and d belong to any of the distribution grids.
  • the derivation of the four output expressions linearly increasing from top to bottom in Figure 10 is applied to the distribution law. If there is no distribution law, the uppermost output value can only be expressed as abc((bvc)d) instead of abcd.
  • a 2 x 2 bittone sorter is a combination logic that can sort the two input signals by the size and send them to the output port according to the address information of one bit, which is a comparator.
  • a Bitonic sequence is a sequence consisting of only a few 0's and 1's. Their order may be first incremented and then decremented, or first incremented and then incremented, or monotonically increasing or monotonically decreasing. (Forms like 0...01...10...0 or 1...10...0111, 0...01...1, 1. ⁇ 10 ⁇ 0 are all sequences Double tone sequence.)
  • a k-double tone sequencer is a network that can order a double-tone sequence of length k into a monotonic sequence.
  • the output data is arranged from small to large in the direction of the arrow. That is, the input can be monotonically added monotonically reduced, or monotonically reduced monotonically increasing the length of the k-tone sequence into a length of k monotonically increasing linear 0...01...1 sequence.
  • a k-half-clearer is a hierarchical network that splits a two-tone sequence a of length 2k into two double-tone sequences aa 2 and guarantees a ⁇ a 2 .
  • Operator " ⁇ " is defined as follows:!. For two equal length sequences of dual tone, a and a 2, & if each element in the sequence a 2 or less in the element, there ai ⁇ a 2 Note that not all This relationship exists between the two-tone sequences. This relationship exists only if one of the two sequences to be compared is all 0s or all ones. Such as 000000 ⁇ 001100, 111100 ⁇ 111111.
  • the specific configuration of the k-half cleaner is as shown in FIG. In FIG.
  • the k consecutively arranged 2 ⁇ 2 double-tone sorting units, the 2 ⁇ 2 dual-tone sorting unit includes two input ends and low output ends respectively outputting a smaller value of the input signals thereof. a high output of a larger value; one input of the nth of the k two-tone sorting units is the nth input of the semi-clearer, and the other input is the semi-clearer a k+n input terminal; a low output end of the nth of the k dual tone sorting units is an nth output end of the semi-clearer, and a high output end thereof is a k+n output terminals; the first output end to the kth output end of the semi-clearer output a bi-level sequence a1, and the k+1th output end of the semi-clearer
  • the input sequence is divided into two parts: where "the upper half” represents the upper half sequence a ⁇ .-ak, and the “lower half” represents the lower half sequence.
  • a k+1 ... a 2k "min”, which represents the sequence of ⁇ ( , +k ) output after the comparison of the semi-cleaners, and "max” , which represents the sequence of max(aj, aj +k ) output after comparison.
  • the half-clearer of the g-class scale is gradually reduced by half, and the number of semi-cleaners of each level is increased by a factor of two.
  • Level 1 is 2 (g_1) vertically stacked 2x2 dual-tone sequencers, ie 2x2 sorting Unit; Level 2, pairing the output of level 1 into 2 (g - 2) double-tone sequences input to 2 (2 ) vertically stacked 4x4 dual-tone sequencers; so recursive..., g-level, The output pair of g-1) is paired into a double-tone sequence of length G into a GxG double-tone sequence sequencer.
  • a 2G-to-G table refers to a 2G X 2G sorting switch module that routes the largest G of the 2G input signals to the G output ports with the largest output address, and routes the remaining G to The G output ports of the minimum output address.
  • the G largest sorted output ports can be thought of as "1-output groups” and the remaining n small sorted output ports as "0-output groups”. Since in the self-routing structure, each input port of each time slot may be idle without data, there may be data sent to "1-output group" or "0-output group", so there are at least three cases.
  • any sequencer that can accurately sort 0- 1 binary sequences can also sort sequences consisting of arbitrary numbers. Therefore, the sequencer constructed above can sort the two pieces of information of the above three input packet states, and the sort size is 1-output group, idle, 0-output group.
  • the sort size is 1-output group, idle, 0-output group.
  • the multicast hub theorem refers to the n-to-m hub built by the sorting unit multi-level internetwork, and the sorting unit is replaced by the dual-cast unit.
  • the signal table be n bicast , let's set it, V.
  • the input values are 0-bound, the input values are 1-bound, and the V B input values are bi-cast.
  • the following results are obtained:
  • the upper nm output ports may produce a total of min ⁇ nm,V. +V B ⁇ 0-bound and bi-cast signals.
  • the lower m output ports may generate a total of minim ⁇ + V B ⁇ 1-bound and bi-cast signals.
  • a Boolean network is a multi-level internetwork in which all nodes are Boolean units. Further, when it is built into a hub, the Boolean Hub network. Each Boolean network is equivalent to a multi-level Internet with Boolean elements.
  • the multicast hub theorem can achieve optimal multicast when it is input with an appropriate multicast signal table n bkast Exchange.
  • n bkast Exchange When a Boolean unit is used instead of a sorting unit in a hub network, the theorem holds for a signal table of any one of the lattices or the lattice structure. This raises a question: What essential properties of the n bicast lattice structure lead to the multicast hub theorem? Careful observation reveals that Q bicast is divided into upper ideal ⁇ , ⁇ and lower ideal ⁇ 1, 1 ⁇ in the statement (7), and similarly divided into ⁇ , ⁇ and ⁇ 0, 1 ⁇ in the statement (8). .
  • S is a sub-grid; if XG S, £ ⁇ xAyG S then sub-single S is an upper [ ⁇ ideal; x £ S, y £ ⁇ xVyG S The sub-grid S is a lower ideal.
  • mapping between two cells is said to be a lattice homomorphism if its Boolean operation remains unchanged.
  • the lattice homomorphism from the lattice ⁇ ] ⁇ 2 is equivalent to the lattice ⁇ divided into an upper ideal and a lower ideal.
  • n-to-m is composed of a sorting unit multi-level internetwork
  • all sorting units in the multi-level internetwork are replaced with Boolean units.
  • Boolean Hub Just get an n-to-m Boolean hub network called a Boolean Hub.
  • the arbitrary distribution lattice ⁇ is divided into an upper ideal U and a lower ideal L. Enter u values from U, 0 ⁇ u ⁇ n, after entering the hub network from nu values in L:
  • the upper nm ports output min ⁇ n_m, u ⁇ belongs to the value of U; the lower m port outputs min ⁇ m, n _ u ⁇ is the value of L.
  • the above unit is an n-to-m Boolean hub.
  • each input signal S with ⁇ (5).
  • This converts the n input signal values into a combination of u zeros and nu ones.
  • the upper nm port output is min ⁇ n_m, u ⁇ 0 and the lower n port outputs min ⁇ m,n _ u ⁇ 1. Because ⁇ is a lattice homomorphism, the same signal is substituted between stages.
  • the output value of one of the output ports of the hub is 0 or 1
  • n-to-m hub consisting of a sorting unit multi-level internetwork, all its sorting units are replaced with bicast units, so that when the signal table is n bicast , one of the following statements holds:
  • n-m ports output only 0-bound signals; the following m ports output only 1-bound signals; no port outputs are idle signals; no port outputs are bicast signals.
  • the abbreviated signal values 0-bound, 1-bound, bicast, and idle are 0, 1, B, and I, respectively.
  • V 1 + V 2 ⁇ m + 1 If V 1 + V 2 ⁇ m + 1, the value of a m+1 is not 0 or B, because some monomial in o m+1 is only assigned values of 0 or B among n variables. Variable related.
  • Boolean hub theory can not only route as many signals as possible to the destination output group, but also achieve optimal routing depending on the priority.
  • “Priority, the concept needs an explanation, because the signal table is only assumed to be a non-ordered set of distribution lattices. This definition is suitable for all possible ways to divide the distribution grid into the upper ideal and the lower ideal. Description.
  • ⁇ 0 + , 0-I, ⁇ , 1-, 1, 1+ ⁇ be the distribution lattice in Figure 17(a)
  • the naming rules for the elements in ⁇ are: Superscript '+, to reach the desired The highest priority of the hub's destination address 0 or 1, '-, indicates the lowest priority.
  • the priority of all routing signals through an n-to-m Boolean hub in ⁇ involves all possible divisions of ⁇ into the upper ideal U set and the lower ideal L set. Here you can combine:
  • Similar priority processing can also be applied in the ordered subset ⁇ 0 + , 0-, I, B, 1-, 1, 1+ ⁇ .
  • Fig. 18 shows an actual hub, wherein Fig. 18 (&) shows a hub of 840-3 applying Boolean hub theory, and the signal with the superscript "+" is processed with high priority.
  • Figure 18(b) When the signal table is a non-allocation grid in Figure 17(b), the ideal exchange is not guaranteed.
  • the signal from the distribution lattice ⁇ is propagated through an 8-to-3 Boolean hub network.
  • 0,1, B, and I represent 0-bound, 1-bound, bicast, and idle, respectively.
  • the circle in the shaded portion of the figure indicates the multicast signal, where the bicast and idle signals are replaced by the 0-bound and 1-bound signals.
  • the output of group 0 is the largest of the total of 0-bound and bicast signals
  • the output of output group 1 is the largest of the total number of 1-bound and bicast signals.
  • this route is optimally consistent with (5) and (7).
  • Boolean hub theory can still be statistically applied when high-priority multicast communications account for a small portion of the overall traffic.
  • a highly self-routing partition network with high modularity and low device complexity can be used for the ultra-large-scale line speed group in this study.
  • the broadcast routing switch architecture provides a basic network structure model.
  • a method of constructing a large-scale almost non-blocking switching structure is to apply a "statistical line group" technique to a divide-and-conquer network. Amplify each 2x2 node in the self-routing network into a 2Gx2G node and replace it with a 2G-to-G fast knockout hub. Each link in the network is replaced with a G-line, thus constructing a band.
  • the structure is implemented by replacing each 2x2 node in a 16x16 banyan network with a Boolean hub consisting of 2G-to-G Boolean units.
  • G 8.
  • G should be a large number, so a 2nx2n banyan-type G-line version of the network is built into an N ⁇ almost non-blocking multicast switch.
  • any G can be constructed into a group hub of any size.
  • This embodiment proposes a line-rate-based packet multicast switching structure based on algebraic lattice.
  • the structure has the following characteristics:
  • the multicast switching structure is modular and the component complexity is low, and there is a recursive extension model supported by mathematical theory; Wire-speed fan-out copy of physical data link is completed, low-latency jitter-free; multicast pursues user quality experience (QoE) and quality of service (QoS), and does not pursue complete non-blocking of video; there is no resource bandwidth and computing power
  • QoE user quality experience
  • QoS quality of service
  • the bottle header through the possible access control algorithm to determine the user's request to join the multicast or new access, and abandon the port scheduling per cell slot.

Landscapes

  • Engineering & Computer Science (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Data Exchanges In Wide-Area Networks (AREA)

Abstract

本发明涉及一种用于组播的基本自路由单元,所述基本自路由单元包括两个输入端口和两个输出端口,所述输入端口输入的信号包括路由信号加上其后的数据优先级控制段和数据内容;所述路由信号具有代数格的结构;所述路由信号包括双播信号、单播信号和空闲信号;当所述两个输入端口的路由信号分别为双播信号和空闲信号时,所述第一输出端口输出的值为两个输入的布尔积,所述第二输出端口输出的值为两个输入的布尔和。本发明还涉及使用上述基本自路由单元构建的网络集线器及其组成的网络的方法。实施本发明的基本自路由单元及其半清器、排序单元、网络集线器和组播交换网络构建方法,具有以下有益效果:便于大规模扩展、不存在网络瓶颈。

Description

基本自路由单元及其半清器、 排序单元、 网络集线器和组播交换网络构建方法 技术领域
本发明涉及通讯网络, 更具体地说, 涉及一种基本自路由单元及其半清器、 排序单 元、 网络集线器和组播交换网络构建方法。
背景技术
近年来, 网络服务在各行各业得到极大的普及。 网络视频服务是使用最多的服务之 一。视频流有典型的特性,其数据内容趋向于单点发送,多点接收,即组播(也叫多播)。 如今流行的基于视频应用的 IPTV、 视频会议、 交互式仿真、 多方游戏等应用的流量都具 有组播特性。 所以为了能够提升网络性能和有效利用率, 组播通信技术显得尤为重要。 互联网早期为了实现组播的特性常采用的是多次单播发送的方式, 这种多次单播的策略 不仅加剧了服务器的工作量, 同时增大网络中的流量。 如果采用 IP组播技术, 通过组播 协议建立起发送方至多接收方的一棵转发树, 发送方只需要发送一份数据内容, 就能通 过中间的路由交换节点进行数据的复制, 最终转发至所有的接受方, 就能从根本上降低 服务器的工作量和网络的负载。 1988年, Deering提出了将组播功能结合到数据网 IP层 的组播结构。 1992年, 组播实验网 -Mbone建立, 在之后的 5年内 IP组播的协议标准和 部署方法在 Mbone中进行研究实验。 1997年, 域间路由协议体系实验在 Internet2的两 个骨干网 (vBNS, Abilene ) 中进行。 1999年, 开始采用 MBGP/MSDP和 PIM-SM协议体 系实现域间组播路由。 2003年, 提出特定源组播协议(SSM:source specified multicast ) 的体系结构。 目前, 较完整的组播协议体系已经形成, IP组播的研究工作逐渐集中于流 量控制和拥塞控制研究, 无线组播研究和大规模高效组播研究。
虽然组播的实现需要网络中包括应用层,传输层,网络层,与数据链路层等的支持。 从实现的物理位置看可以分为终端和中间路由节点, 从实现的方式来看可以分为多次单 播软件方式调度和硬件电路线速扇出拷贝方式。 但整体看, 终端基本是以软件方式实现 各种应用层, 传输控制层, 网络层协议, 通常不存在瓶頭; 而路由交换节点的实现是组 播性能的关键, 一是因为分组可能通过若干跳 (如 5跳, 8跳) 的路由节点但只经过一 跳终端, 二是路由节点需要处理许多端口的争用, 本身是网络的瓶颈。
路由交换节点的组播实现方式可以分为多次单播软件调度方式和硬件电路线速扇 出拷贝方式。 文献 [刘莹, 徐恪; 《Intemet组播体系结构》 , 科学出版社, 2008年 4月] 给 出目前路由交换节点上组播多数是多次单播的软组播, 即在路由前对组播包先进行复制 后各自在源端口队列排队, 再通过软件实现端口配对各自调度方式完成包传送。 目前 Cisco [William R. Parkhurst; 《Cisco multicast routing and switching》, McGraw-Hill, 1999]、 华为公司等绝大多数路由器都实现了 PIM-SM组播路由协议和 IGMP协议, 具有支持组播 业务的能力。 但软组播实时性差而不能保证质量, 这种方法使得发送至不同用户端的组 播分组数据包, 在同一节点经历不同的时延, 使得接收方的数据产生抖动, 组播性能下 降。
图 1示出了 Knockout交换结构, Knockout即体育淘汰赛之意, 该类结构的提出来源 于一个显而易见的事实: 在同一时隙中,所有输入信元到达同一目的端口的概率非常小。 图 1是基于该理论设计的交换结构原型, 每个输入端口采用完全连接的方式连接到所有 的输出接口模块。 在每个输出接口, 通过地址过滤器, 过滤掉目的地址非该输出模块的 信元, 而后通过一个 N:L的 knockout集线器使得在同一时隙中, 最多允许到同一目的端 口的信元数量为定值 L , 超出该值的信元则被丢弃。 文献 [Y.-S.Yeh,M.G.HIuchyj, A. S.Acampora; "The knockout switch:Asimple,modular architecture for high-performance packet switching " IEEE Journal on SelectedAreas inCommunications, vol. 5, no. 8, pp. 1274-1283, Oct. 1987]中指出,对于独立同分布的单播流量, L=12便可使得信元的丢失率 低于 10- 1Q, 且与交换结构的规模无关。
由于每个输入端以广播的方式连接到所有的输出模块, 这种结构具有组播特性。 文 献 [H.丄 Chao, B. S. Choe; "Design and analysis of a large-scale multicast output buffered ATM switch " IEEE/ACM Trans. Networking, vol. 3, pp. 126-138, Apr.1995] [ H. J. Chao, B. S. Choe, J. S. Park, N. Uzun; "Design and implementation of Abacus switch: a scalable multicast ATM switch " IEEE J. Selected Areas Commun., vol. 15, no. 5, pp. 830-843, 1997] [ K. Y. Eng, M. G. Hluchyj, Y. S. Yeh; "Multicast and Broadcast Services in a Knockout Packet Switch," INFOCOM'88, pp.29-34, 1988]提出了基于该结构的组播设计方案, 图 2为 [H. J. Chao, B. S. Choe; "Design and analysis of a large-scale multicast output buffered ATM switch," IEEE/ACM Trans. Networking, vol. 3, pp. 126-138, Apr.1995]中的 MOBAS(Multicast Output Buffer ATM Switch)的基本结构: 该结构由输入 /输出端口控制器 (IPC/OPC) , 组播集群网絡 (MGN,图 3)以及组播地址翻译表 (MTT)三部分构成,整体逻辑上分为 MGN1与 MGN2的两 级, 由于每级之间总线型的连接方式, 以及 MGN模块 CrossBar的结构特点, 该结构同 样具有组播特性。 组播信元进行交换时, 在每级适当位置进行信元复制, 最终到达指定 的目的端口。 信元在每一个 MGN模块中, 采用自路由方式到达其所希望到达的输出群 组。 在图 2、 3的结构中, 虽然通过集群扩展的方式有效地减低了信元争用的丢失率, 但 该结构是基于 CrossBar结构 Θ(Ν2)的器件复杂度, 及级间总线式的完全连接所不得不面对 线路器件驱动能力的极限问题, 使得该类方案的大规模扩展存在瓶颈。
前面的方案是基于自身具备组播特性的交换结构而实现的, 而基于 Banyan这一类多 级互联网络, 由于其一般不具备组播特性, 所以大多都采用的是组播复制的方法来实现 組播交换 , [T. T. Lee; "Nonblocking copy networks for multicast packet switching", IEEE Journal onSelected Areas in Communications, vol. 6, no. 9, pp.1455-1467, Dec. 1988] 采用的是先复制后路由的复制方案, [N. McKeown; "The iSLIP scheduling algorithm for input-queued switches," lEEE/ACMTransactions on Networking, vol. 7, no. 2, pp. 188-201, April 1999]则采用信元反馈的方式实现组播信元的复制。
该方案采用路由前复制的组播策略, 其关键在于前级复制网络的设计。 复制网络用 于将需要组播的信元复制为其扇出 (目的地址的数量) 的个数, 再由后级的点对点的单 播交换结构完成最终信元的交换。 图 4是 Tony Lee[T.T.Lee;"Nonblocking copy networks for multicast packet switching", IEEE Journal on Selected Areas in Communications, vol. 6, no. 9, pp.1455-1467, Dec. 1988]提出的一种内部无阻塞的复制网络的设计方案。
Tony Lee的复制网洛由累力 P和网络 (running sum adder network) , 集线器网络
(concentrator network)以及广播榕才对网络 (broadcast banyan network)三大部分组成。 累加 和网络用于计算需要复制信元的数量。 如图 4所示, 在累加和网络的输入端处, 每一个 信元都标识了其扇出的大小。 每个输入端口的扇出通过累加和网络计算后, 可以得到该 时隙每个端口累计所需要复制信元的个数, 由于在同一个时隙中需要组播分组的复制总 和可能比输出端口的数量要大, 故需要拒绝第一个累加和的值大于输出端口数目的端口, 及该端口之后的所有端口的信元。 成功通过累加和网络的信元, 通过虚拟地址编码器 ( dummy address encoder )编码, 得到每个信元输出地址范围的编码, 然后通过集线器 网络汇聚在彼此相邻的端口中,以满足广播榕树网络内部无阻塞的条件如文献 [T. T. Lee; "Nonblocking copy networks for multicast packet switching", IEEE Journal on Selected Areas in Communications, vol. 6, no. 9, pp.1455— 1467, Dec. 1988]。 最后使用布尔间 隔分割算法( boolean interval splitting algorithm ) , 把分组在网络节点合适位置处进行复 制, 以便实现复制网络的功能, 而信元通过组播地址转换表, 将组播地址转换为相应的 单播地址, 再由后级的单播网络完成最终的交换。
虽然该组播网络较好地为输入信元提供了内部无阻塞, 自路由且有固定时延的特性, 但该结构存在如下问题: 累加和网絡及虚拟地址编码器不适合大规模扩展; 当输入分 组复制要求的总和超出了输出端口数量时, 满溢 (overflow)的现象便会发生, 若复制网络 不提供重传的机制, 超出端口数量的分组便会被丢弃, 且在该复制网络中, 因满溢而丟 失的概率将由上至下递增, 造成服务的不公平; 该结构进行组播地址转换所需要的緩存 开销'非常庞大, 文南史 [Turner, J.S; "A practical version of Lee's multicast switch architecture", IEEE Transactions on Communications, vol.41, pp.1166 - 1169, Aug 1993]中也指出了其每个 查找表的大小为 256Mb, 緩存开销大。 故其结构过于复杂, 集线器网络和广播榕树网络 的元件复杂度均为 0(W^2N) , 与一般自路由网络复杂度等价。
文献 [Turner, J.S; "A practical version of Lee's multicast switch architecture", IEEE Transactions on Communications, vol.41, pp.1166 - 1169, Aug 1993]中提出减小组播转换表 的方法, 但它进一步增加了复制网络的复杂性, 使得难以扩展到大规模的交换结构的设 计中。
图 5示出的 Sunshine [N. McKeown; "The iSLIP scheduling algorithm for input-queued switches," IEEE/ ACM Transactions on Networking, vol. 7, no. 2, pp. 188-201, April 1999]是 另一类基于 Banyan网络的组播交换结构, 其整个结构特点与 Starlite swit相类似, 均基于 Batcher-Banyan网絡的基础框架。 它通过引入 K个并行的 Banyan网絡增强了整个结构的 健壮性以及增强了抗突发流量的特性, 此外还通过引入反馈来增加组播能力, 这是一种 信元反馈的组播复制的方案。 该网络的每一个输入端有一个输入端口控制器 (input port controller ) , 但发现组播信元时, 通过特殊的标记, 使得该信元能够在送入并行 Banyan 网络前被选择器(Selector )反馈回到输入端, 直到组播信元的所有目的地址均被送到为 止。 上述 SunShine结构中 Batcher 网络 0(N log2 N)的复杂性, 本身就是整个结构的扩展瓶 颈, 而通过反馈来实现组播复制的方案虽然避免了对复制网络的要求, 但却存在着输出 乱序, 增大了组播信元时延等问题。
由于 Clos[C Clos, "A study of non-blocking switching networks", Bell System Technical Journal, 1953]三级互连网络在电路交换的历史地位, 以及其在严格无阻塞情况下优于传 统 CrossBar结构的元件复杂性( Clos为 Θ (Ν1 5) ), 其吸引了不少学者对 Clos应用于分组交 换的研究。应用 Clos结构进行分组交换的难点在于如何进行有效的路由选择来完成信元 的交换,该问题可等效的转换为速率矩阵的分解或二分图的最少边染色 [H. J. Chao, Bin Liu; 《High Performance Switches and Routers》 , John Wiley & Sons, April 2007](P383-387)。 与 单播流量的情形一致, Clos结构应用于组播交换的关键点同样是路由的选择与信元的调 度。 图 6基于 Growable packet switch(GPS)[ D. J. Marchok, C. E. Rohrs, R. M. Schafer; "Multicasting in a Growable Packet (ATM) Switch," INFOCOM'91, pp.850-858, 1991]结构提出 了一种组播调度算法。 GPS与 Clos的三级结构类似,但前两级为无緩存自路由互连网络, 最后一级模块为小规模的输出交换模块, 在这一类结构中实现路由主要是通过指定前两 级的路由而实现的。
文献 [D. J. Marchok, C. E. Rohrs, R. M. Schafer; "Multicasting in a Growable Packet (ATM) Switch," INFOCOM'91, pp.850-858, 1991]引入了一种路径分配向量, 通过在第一级模块中 以一种环形的方式对中间模块的路径进行预留来实现组播路径的分配, 该算法类似 [B. Bingham, H. Bussey; "Reservation-based contention resolution mechanism for Batcher-Banyan packet switches", Electron Lett. 24 (13), pp 772-773, 1988]所述的一种基于 Banyan网络的解决输出争用的算法。 虽然这种算法不是中央集中式的调度, 不存在中央 处理能力的瓶颈, 但其整个路径决策需要第一级所有模块依次参与, 假定第一级模块的 数量为 k, 其算法的时间复杂度达 Θ(Κ) , 这使得该结构不能够大规模扩展。
许多基于 Clos的组播路径选择算法相继提出, 2010年最新文献[Jastrzebski, A.,Kubale, M.; "Rearrangeability in multicast Clos networks is NP-complete"; 2nd International
Conference on Information Technology (ICIT), pp.183-186, Aug. 2010]证明对 Clos网络进 行组播路径匹配是 NP完全问题, 这进一步表明了 Clos组播调度算法的复杂度。
综上所述, 现有技术中均存在不能大规模扩展、 存在瓶颈的问题。
发明内容
本发明要解决的技术问题在于, 针对现有技术的上述不能大规模扩展、 存在瓶頭的 缺陷, 提供一种便于大规模扩展、 不存在网络瓶颈的组播实现技术, 即基本自路由单元 及其半清器、 排序单元、 网络集线器和组播交换网络构建方法。
本发明解决其技术问题所釆用的技术方案是: 构造一种用于组播的基本自路由单元, 所述基本自路由单元包括两个输入端口和两个输出端口, 所述输入端口包括第一输入端 口和第二输入端口, 所述输出端口分别为第一输出端口和第二输出端口; 所述输入端口 输入的信号包括路由信号加上其后数据属性和数据内容; 所述路由信号具有代数格的结 构; 所述路由信号包括双播信号、 单播信号和空闲信号; 当所述两个输入端口的路由信 号分别为双播信号和空闲信号时, 所述路由信号为双播信号的输入端口分别与所述第一 输出端口和第二输出端口连接, 所述第一输出端口输出的路由信号值为两个输入路由信 号的布尔积, 所述第二输出端口输出的路由信号值为两个输入路由信号的布尔和。 更进一步地, 当所述两个输入端口输入的路由信号均指向两个输出端口中的一个时, 所述两个输入端口依据其输入信号的数据属性中的数据优先级争用所述被指向的输出 端口, 所述两个输入端口中数据优先级高的与所述输出端口连接, 数据优先级低的与所 述的另一输出端口连接; 当所述两个输入端口输入的路由信号分别指向不同的输出端口 时, 所述输入端口分别与所述输出端口交叉或平行连接; 所述交叉连接包括第一输入端 口与第二输出端口连接, 所述第二输入端口与所述第一输出端口连接; 所述平行连接包 括所述第一输入端口与第一输出端口连接, 所述第二输入端口与所述第二输出端口连接。
更进一步地, 所述路由信号具有代数格的结构包括使用代数格构建自路由带内路由 信号表, 所述代数格是分配格。
本发明还涉及一种使用上述基本自路由单元构成的半清器, 所述 k个依次排列的 2 X 2默调排序单元,所述 2 X 2双调排序单元包括两个输入端和分别输出其输入信号中值 较小的到低输出端和值较大的到高输出端; 所述 k个双调排序单元中的第 n个的一个输 出端为所述半清器的第 n个输入端, 其另一个输出端为所述半清器的第 k+n个输入端; 所述 k个双调排序单元中的第 n个的低输出端为所述半清器的第 n个输出端, 其高输出 端为所述半清器的第 k+n个输出端; 所述半清器的第 1个输出端至第 k个输出端输出一 个双调序列 al ,所述半清器的第 k+1个输出端至第 2k个输出端输出一个 ¾调序列 a2, al < a2; 其中, k为正整数, n=l , 2 , k;所述 2 χ 2双调排序单元为所述基本自路由单 元。
更进一步地, 当 k=l时, 所述半清器为所述基本自路由单元。
本发明还涉及一种使用上述自路由单元构成的双调排序单元, 所述双调排序单元包 括 G个输入端和 G个输出端, 所述 G = 2g ,g为正整数; 所述双调排序单元包括 g级, 其 中, 第 m级包括 2"1-1个 k=G/2m的半清器, 其中, m=l,2, .. .,g; 所述每级半清器包括多 个 2 X 2双调排序单元, 每个 2 X 2双调排序单元的输出端分别连接在其下一级半清器所 包括的不同 2 X 2双调排序单元的输入端或其下一级中不同的半清器输入端。
本发明还涉及一种使用上述自路由单元构成的任意二元序列排序单元, 所述任意二 元排序单元包括 G输入端和 G个输出端, 所述 G=2g, g为正整数; 所述任意二元排序 单元包括 g级双调排序单元, 其中, 第 p级包括 2g-p个 G=2P的 G X G双调排序单元; 所 述双调排序单元按照其所在的级依次连接。
本发明还涉及一种使用上述自路由单元构成的网络集线器,所述网络集线器包括 2G 个输入端和 2G个输出端,所述网络集线器有 2个 G X G任意二元排序单元和与所述 2个 G x G任意二元排序单元的输出连接的 k=G的半清器组成, 所述半清器的 G个最大排序 输出端口为 1-输出群组, 所述半清器的 G个最小排序输出端口为 0-输出群组。
更进一步地, 所述网络集线器的每个输出端上还串接有地址过滤单元。
本发明还涉及一种构建组播交换网络的方法, 包括如下步骤:
A ) 构建一个使用版图复杂性最优的分治网络结构的自路由结构网络, 所述自 路由结构网络包括多个 2 X 2路由单元及其间的连线;
B )使用如权利要求 9所述网络集线器代替所述 2 X 2路由单元, 并将每条所述 连线使用 G根线的线束代替;
C )得到具有 M个输出群组, 每群组包含 G个输出端口的 N X N组播交换网络; 其中, N表示所述组播交换网络的输入 /输出总线数, N=MG。
实施本发明的基本自路由单元及其半清器、 排序单元、 网络集线器和组播交换网络 构建方法, 具有以下有益效果: 由于其基本自路由单元在出现组播信号时分别在其输出 端输出输入信号的布尔和以及布尔积, 使得其不仅能够实现单播, 也可以在组播时实现 信号的排序或筛选; 同时, 该还可以由该基本单元得到半清器、 排序单元以及网絡集线 器, 进而实现组播网络的路由交换功能。 所以, 其便于大规模扩展、 不存在网络瓶颈。 附图说明
图 1是现有技术中基于 Knockout交换结构的结构示意图;
图 2是现有技术中基于 MOBAS的基本交换结构示意图;
图 3是图 2中 MNG模块的结构示意图;
图 4是现有技术中 Tony Lee复制网络的结构示意图;
图 5是现有技术中另一类基于 Banyan网络的组播交换结构示意图;
图 6是现有技术中基于 Growable packet switch组播交换结构示意图;
图 7是本发明实施例中基本自路由单元在单播状态时的部分状态;
图 8是本发明实施例中基本自路由单元在组播状态时的部分状态;
图 9是本发明实施例中作为组播信号表的分配格状态图;
图 10是本发明实施例中由基本自路由单元构成的排序网络示意图;
图 11是本发明实施例中双调排序单元的结构示意图;
图 12是本发明实施例中 k半清器的结构示意图;
图 13是本发明实施例中 k半清器的状态枚举示意图;
图 14是本发明实施例中双调排序单元的结构示意图;
图 15是本发明实施例中任意二元序列排序单元的结构示意图;
图 16是本发明实施例中网络集线器的结构示意图; 图 17是本发明实施例中分配格和不可分配格的状态示意图;
图 18是本发明实施例中一种实际的网络集线器的信号传输过程示意图;
图 19是本发明实施例中使用网络集线器构成的网络结构示意图。
具体实施方式
下面将结合附图对本发明实施例作进一步说明。
如图 7所示, 在本发明基本自路由单元及其半清器、 排序单元、 网络集线器和组播 交换网络构建方法实施例中, 正是以图 7所示的网络为基础, 通过设置或创建一种新的 基本自路由单元, 进而由这种基本自路由单元得到在网络中使用的其他器件或单元, 从 而得到本实施例中的组播交换网络。 图 7是基于群论多级互连网络完全分布式自路由模 型 [Wei He, Hui Li (李挥) , Bin-qiang Wang (汪斌强) ; "Load-Balanced Multipath Self-routing Switching Structure by Concentrators", IEEE Proc. of ICC2008, May ,2008]使 用了一类称为分治网络( Divide & Conque ) [S.-Y. . Li(李硕彦), Hui Li(李挥); "Layout complexity of bit-permuting exchange in multi-stage interconnection networks", book chapter in《 Switching Networks: Recent Advances》, Kluwer Academic Publishers, Boston, USA, pp.259-276, 2001] , 其具有高度模块化, 最低的器件复杂度 (Mog2N ) 的特点。 如下表 1是 N = 64, 256, 1024, 4096时的模型描述, 其中 ":,,表示一级 2x2单元, 而两级 间连线用一位置换群元素表示,如 (6 3) (5 2) (4 1),数字表示源和目标二进制地址的下标, 如 S 2S3S4, ^(12(13(14。 图 7表示 N=64时的模型。 (注: N为任何 2n规模的网络都有其对应 的分治网络) 下表描述了几个不同规模分治网络的置换群模型:
64x64: [ id: (6 5): (6 5 4): (6 3)(5 2)(4 1): (6 5): (6 5 4): id ];
(如图 3.1所示 )
256x256: [ id :(8 7): (8 6)(7 5) :(8 7):(8 5)(7 3)(6 2)(5 1) :(8 7): (8
6)(7 5) : (8 7) : id ];
1024 1024: [id :(10 9):(10 9 8) :(6 9 7 10 8): (10 9 ) :(10 5)(9 4)(8 3)(7
2) (6 1): (10 9): (10 9 8) :(6 9 7 10 8): (10 9 ): id ];
4096x4096: [ id : (12 11): (12 11 10): (12 9) (11 8) (10 7):
(12 11) : (12 11 10) : (12 6)(11 5)(10 4) (9 3)(8
2)(7 1 ) : (12 11): (12 11 10): (12 9) (11 8) (10 7):
(12 11): (12 11 10): id]; 在分治网络中, 一个 2x2基本自路由排序元件的数据分组排序交换过程可以使用 带内信令来实现, 如在数据分组的前面加上两位带内信令。 第一位 ^舌跃比特表示当 前时隙是否有数据分组, 1表示有活跃的有效数据分组, 0表示当前没有有效数据的空 分组; 当 为 1时第二位 0表示分组的输出目标端口地址是 0或 1, 为 0时 没意义。 故 等于' ΙΟ' , 'ΙΙ', '00' 分别代表输出端口目标为 0的有效数据分组、 输出 端口目标为 1的有效数据分组和空分组。
因此, 在单播时,一个 2x2网络基础元件可以按如下规定的线性排序关系来实现自 路由: 10<00<11 , 单播 2x2基本自路由元件以 2位带内信令自路由交换的具体控制方 式可参见下表:
Figure imgf000011_0001
该元件根据 fast knockout[S.-Y. R. Li (李硕彦), Hui Li (李挥); "Optimization in the fast knockout algorithm for self-route concentration", IEEE Proceedings of ICC'98, pp. 630-634, Atlanta, Jime,1998]集线器构造算法或 bitonic circular [S.-Y. R.Li (李硕彦); 《 Algebraic Switching Theory and Broadband Applications》, Academic Press, 2001] 再酉己合 一级半清器( half-cleaner ) [Hui Li (李挥), S.-Y. R. Li (李硕彦); 《On the Complexity of Concentrators and Multi-stage Interconnection Networks in Switching System》 , The Chinese University of Hong Kong, 2000.1 1] , 可以组成任意 2G-to-G的自路由群组集线器。 在本实施例中, 为了实现组播与单播完全融合, 组播以线速扇出拷贝在物理数据链 路层完成的目标, 需要构建基本组播电路单元。 一般来讲, 交换结构的理想的结构是其 规模可以任意递归扩展, 不存在各种瓶颈, 就是由最基本的 2x2元件通过各种拓朴结构 以最低的代价构造符合性能指标的大型交换系统,。
显然前面所述的单播 2x2基本自路由元件, 不具备组播所要求的性质。 为此, 在本 实施例中, 构建一种满足組播情况的新的 2x2基本自路由元件, 也就是一个基本的 2x2 排序单元, 并由该单元得到一个大型的交换系统。
该排序单元在被用于 2x2单播交换结构时, 两个输出标记为 0和 1。 而路由单元意 味着对分组路由具有定义: 一个有效的分组包含了路由信号加上其后的负载, 并且这个 信号的值不是 0-bound就是 l-bound, 依赖于其本身的目标输出。 同时信号的空值表示 为 idle。 在本实施例中, 路由单元其对应的自路由带内信号表为 Qraute = {0 _ bound, 1 _ bound, idle}; 即, 0-bound =10; l-bound=ll, idle=00。 故原来 10<00<11的线性排序等于 如下规定的顺序: 0-bound<idle<l-bound。
进行线性排序操作而把尽可能多地有效信号转发到他们预定的输出端口。 当出现两 个 0-bound 分组或两个 1-bound分组的输出争用, 直通 /交叉状态的选择依赖于分组优 先级而确定。
而要支持线速组播的 2x2的排序单元, 必须具备下表功能的单元, 其中部分状态如 图 7a-图 7c所示, 而这些图表示了单播 2x2基本自路由排序元件 处于平行, 交叉, 争 用的状态。可见当输入组播,这里就表示负载投递的目的为 0和 1两个端口的双播 bicast 信号, 与空闲 idle信号相遇时, 能够线速扇出拷贝 2个分组同时送到输出端口 0和 1是 新增核心功能。 在本文中, "Λ"代表布尔代数的析取, "V"代表布尔代数的合取。 支持组 播以线速扇出的 2x2组播排序单元的功能如下表: 组播交换排序 0输入端的输入信令信息
单元连接状态 0 1 B I
1输入端 0 争用输出口 0 交叉 交叉 交叉
的输入 1 平行 争用输出口 1 平行 平行
信令信
B 平行 交叉 交叉 /平行 组播拷贝 息
I 平行 交叉 组播拷贝 交叉 /平行 可见支持组播的自路由带内信号表必须扩展为: Qbicast = Qroute U {bicast}。
这个扩展集合 Qbicast由规贝 'J 0-bound<idle<l-bound和规 J¾'J 0-bound<bicast<l-bound各 自部分排序。 组播单元对属于集合 Qbieast的信号执行排序, 但当 /d/e遇到 b/o ^时交换控 制遵循如下规则: output-0端口输出信号值 0-bound , output-1端口输出信号值 1-bou nd , 其后都跟随着相同的负载。
这样, 分组的负载在通过组播单元时可以是直通状态, 交叉状态或者是一个输入端 连接两个输出端的组播方式。 按这种方式下, 组播单元将尽可能多的分组负载转发到其 预定目标端口。
为了能够把上述排序规则 (2个)与上述交换规则统一起来, 信号表 Qbieast必须被赋 予代数格的结构。格是一个有二元操作数布尔和 ("V" )和布尔积 ("Λ" ), 且遵守如下性质的 集合:
1) 交换律: aAb = bAa, aVb = bVa
2) 结合律: aA(bAc) = (aAb)Ac, aV(bVc) = (aVb)Vc
3) 幂等律: aAa = a = aVa
4) 吸收律: aA(aVb) = a = aV(aAb)
常见的格是遵守布尔操作中并集和交集运算的集合的一组子集。 根据规则: a≤ b aAb = a , 可知: 每一个格都可以导出一个部分有序序列。 如杲一个部分有序集合(简称 偏序集) 中任意两元素存在一个最大下界和一个最小上界则称这个偏序集为偏序格。 用 布尔和与布尔积表示这两个界限, 偏序格具有格的特性。 另一方面, 每一个被格定义的 偏序集自然为偏序格。 如此看来偏序格逻辑上等同于格。
根据上述排序规则, 图 8a-图 8h描述了 Qbicast的部分排序, 它们共同规定了高处的较 小元素和低处的较大元素的连接。很明显偏序集 Qbicast具有偏序格的特性,因此被称为格。 至此组播单元运行的 3个规则 ( 2个排序规则和 1个交换规则) 能够被统一于以下布尔 规则: 即 0输出端口的值为两个输入的布尔积, 1输出端口的值为两个输入的布尔和。
如果一个格满足分配律:
aA(bvc) = (aAb)v(aAc) ,
aV(bAc) = (aVb)A(aVc)
则此格为分配格。 Qbicast就是一个筒单的例子,见图 9, 其表示连接高端的较小元素与低端 的较大元素的偏序集。 偏序集 Qbkast具有格偏序集的特性, 因此可以被看作格。 实际上一 个分配各可以作为组播单元的信号表。 因此有时也把该组播单元称为布尔单元。 综上所述, 在本实施例中, 构建了一种用于组播的基本自路由单元, 其包括两个 输入端口和两个输出端口, 其输入端口包括第一输入端口和第二输入端口 (上述 0输入 端口和 1输入端口), 其输出端口分别为第一输出端口和第二输出端口 (上述 0输出端 口和 1 输出端口); 其输入端口输入的信号包括路由信号加上其后的数据属性 (此处所 讲数据属性包括了表示该数据的优先级的数据优先级字段) 和数据内容; 所述路由信号 具有代数格的结构; 所述路由信号包括双播信号、 单播信号和空闲信号; 当所述两个输 入端口的路由信号分别为双播信号和空闲信号时, 所述路由信号为双播信号的输入端口 分别与所述第一输出端口和第二输出端口连接, 所述第一输出端口输出的路由信号值为 两个输入路由信号的布尔积, 所述第二输出端口输出的路由信号值为两个输入路由信号 的布尔和。 当所述两个输入端口输入的路由信号均指向两个输出端口中的一个时, 所述 两个输入端口依据其输入信号的数据属性中的数据优先级争用所述被指向的输出端口, 所述两个输入端口中数据优先级高的与所述输出端口连接, 数据优先级低的与所述的另 一输出端口连接; 当所述两个输入端口输入的路由信号分别指向不同的输出端口时, 所 述输入端口分别与所述输出端口交叉或平行连接; 所述交叉连接包括第一输入端口与第 二输出端口连接, 所述第二输入端口与所述第一输出端口连接; 所述平行连接包括所述 第一输入端口与第一输出端口连接, 所述第二输入端口与所述第二输出端口连接。 在本 实施例中, 所述路由信号具有代数格的结构包括使用代数格构建自路由带内路由信号表, 所述代数格是分配格。
在本实施例中, 通过上述描述容易得出对于排序有以下代数结构的层次关系: 有序 集 分配格 → 格 → 偏序集。 在本实施例中, 采用分配格来构建组播排序和组播交 换的自路由带内信号表。 目前为止, 常用交换结构使用的都是线性有序集, 尤其当信号 值为数字表示的时候更是如此。
对所有带内信号进行完全排序的要求不仅限制了结构的应用范围, 并且更糟的是完 全阻碍了组播操作。 从基于分配格的 2x2组播单元与原来 2x2单播单元的比较可以清楚 的看点这一点。
注意分配律的重要性,分配格可以对特殊的应用进行定制,比如,附加服务质量( QoS ) 特征的组播。举例来说,合取范式基于支持分配格的全部五个公理。 而一般化的排序 0-1 原理在文献 [S.-Y. R丄 i (李硕彦); 《Algebraic Switching Theory and Broadband Applications》 , Academic Press, 2001]中有介绍, 需要合取范式相同的公理基础。 在另一方面, 当所有的 原始信号值不是完全有序的时候, 信号表通常被构建成偏序集结构。 然而用支配偏序集 或甚至偏序格的公理衍生出来的数学工具对组播交换是不够的。
图 10清楚说明了分配格的重要性, 它展示了一个由 2x2布尔排序器( Boolean cell ) 加上级间同步的 1x1延时器组成的 4x4三级布尔排序网络。 四个输入值 a, b, c和 d属 于任意的分配格。 图 10 中从上到下线性增大的 4个输出表达式的导出应用到分配律, 若没有分配律最上部输出值只能表示为 abc((bvc)d)而不是 abcd。
分配律对布尔集线器定理同样重要。更多证据和直觉表明,在构建组播交换结构中, 代数排序的结构层次关系中, 从器件实现的成本复杂性及应用可能性灵活性看, 分配格 是自然适合的完美选择。
在本实施例中, 还包括由上述基本自路由单元得到的半清器、 排序单元和网络集线 器, 这些网络单元都是由上述基本自路由单元通过不同拓朴递归而得, 具体描述如下: 一般来讲, 一个 2 x 2双调排序单元(bitonic sorter)是能根据一位的地址信息, 把 两个输入端信号按大小自路由排序后送输出端口的组合逻辑电路, 即为比较器。 而一个 双调 ( Bitonic )序列是一个仅由若干个 0和 1组成的序列。 它们的排列顺序可能是先递增 后递减, 或者先递减后递增, 也可以是单调递增或者单调递减。 (形如 0...01...10...0或 1...10...0111, 0...01...1, 1.··10···0的序列都是双调序列。 )
一个 k-双调排序器是能将一个长度为 k的双调序列排序为单调序列的网络。 如图 11 所示, 输出端数据按箭头方向从小到大排列。 即是能把输入为单调增接单调减, 或单调 减接单调增总长度为 k的双调序列排序成长度为 k单调增线性 0...01...1序列。
一个 k-半清器是这样一级网络, 它能将一个长度为 2k的双调序列 a, 拆分为 2个 双调序列 a a2, 并且能保证 a^a2。 运算符"≤ "定义如下: 对于两个等长双调序列, a!和 a2,如果 &序列中的每个元素均小于等于 a2中的元素,则有 ai≤a2. 注意并非所有的双调 序列之间都存在这种关系。 只有两个待比较序列中有一个是全 0或全 1时, 才存在此关 系。 如 000000≤001100, 111100≤111111。 在本实施例中, k-半清器具体的构造如图 12所 示。 在该图 12中, 所述 k个依次排列的 2 X 2双调排序单元, 所述 2 x 2双调排序单元 包括两个输入端和分别输出其输入信号中较小值的低输出端和较大值的高输出端; 所述 k个双调排序单元中的第 n个的一个输入端为所述半清器的第 n个输入端, 其另一个输 入端为所述半清器的第 k+n个输入端; 所述 k个双调排序单元中的第 n个的低输出端为 所述半清器的第 n个输出端, 其高输出端为所述半清器的第 k+n个输出端; 所述半清器 的第 1个输出端至第 k个输出端输出一个双调序列 al,所述半清器的第 k+1个输出端至 第 2k个输出端输出一个双调序列 a2,al a2; 其中, k为正整数, n=l,2, k;所述 2 X 2默调排序单元为所述基本自路由单元。
由定义长度为 2k的双调序列 ai...a2k, 它只有两种情况 θΨθ"1,或 l^l"1 , i+j+m = 2k, i, j和 m为 0到 2k的数。 由对称性,证明 θΨ(Γ就可以了。如图 13a-图 13d所示, 根据 i, j, m, 的大小, 决定了把 2k序列等分两半的情况, 有图 13中 a/b/c/d四种情况, 其构造性枚举证明一起示出。 首先将输入序列分为 2 部分: 其中"上半"表示上半序列 a^.-ak,"下半"表示下半序列 ak+1...a2k, "min",表示半清器比较后输出的 ηώι( , +k)序列, "max" , 表示比较后输出的 max(aj, aj+k)序列。
任意两个 0,1组成的序列, 本身均构成一个长度为 2的双调序列。 故 K= l的半清器就 是一个上面规定的 2x2双调排序单元 ( bitonic sorter ) , 其输出已经是线性排序。
根据上面定义及半清器结构, 可用如下方法递归构造输入规模为0 = 2g双调序列排 序器。 第 1级是 1个 k = G/2的半清器, 其输出是 2个 的默调序列, 其中有一个 已经是完全相同的 0或 1序列; 第 2级是 2个(G/2) (G/2) 双调序列排序器。 而每个 (G/2) X (G/2)双调排序器的第 1级是一个 k = G/4的半清器,其输出是 2个 G/4 = 2 ( g- 1 )的双调序列, 其中有一个已经是完全相同的 0或 1序列,故第 2级共输出 4个 G/4 双调序列;第 3级 4个 (G/4) X (G/4)双调序列排序器; ...如此递归, 第 g级是 2(g- 1〕个 2x2双调序列排序器, 即是 k=l 的半清器。
故所以共用了 g级规模逐级减半的半清器, 每级的半清器的数量是按倍递增。
如图 14a是一般递归构造结构, 图 14b具体显示了一个 G = 4的双调排序器可以通过一个 2- 半清器和两个 2 X 2排序单元构造出来。 图 14c表示了 G = 8双调排序器由 1个 4-半清器和 2 个 G = 4双调排序器构成。 因此, 在本实施例中, 双调排序单元包括 G个输入端和 G个输 出端,所述 G = 2g,g为正整数;所述双调排序单元包括 g级,其中,第 m级包括 2"1-1个 k=G/2m 的半清器, m=l,2, ...,g; 所述每级半清器包括多个 2x2双调排序单元, 每个 2x2双调 排序单元的输出端分别连接在其下一级半清器所包括的不同 2 X 2双调排序单元的输入 端或其下一级中不同的半清器输入端。
当输入的 G = 2g个信号是任意的 0-1二元序列时, 可以用双调排序器递归构造: 第 1 级是 2(g_1)个垂直堆叠的 2x2双调排序器, 即 2x2排序单元; 第 2级, 将第 1级的输出配对 为 2(g2)个双调序列输入给 2 ( 2)个垂直堆叠的 4x4双调排序器; 如此递归…, 第 g级, 第 (g- 1 )级的输出配对为 1个长度为 G的双调序列输入给 1个 GxG的双调序列排序器。 图 15 (a)表示 G = 23的任意 0-1二元序列排序器递归构造过程, 图 15 (b) 则显示了一个 G = 8的任意 0-1二元序列排序器的详细构造。 在本实施例中, 任意二元排序单元包括 G输 入端和 G个输出端, 所述 G=2g, g为正整数; 所述任意二元排序单元包括 g级双调排序单 元, 其中, 第 p级包括 2g-p个 G=2P的 G x G双调排序单元; 所述双调排序单元按照其所在的 级依次连接。
一个 2G-to-G桌 是指一个 2G X 2G的排序交换模块, 它将 2G个输入信号中最 大的 G个路由到具有最大输出地址的 G个输出端口, 并将其余的 G个路由到具有最小 输出地址的 G个输出端口。 所述网络集线器包括 2G个输入端和 2G个输出端, 所述网 絡集线器有 2个 G X G任意二元排序单元和与所述 2个 G X G任意二元排序单元的输出 连接的 k=G的半清器组成, 所述半清器的 G个最大排序输出端口为 1-输出群組, 所述 半清器的 G个最小排序输出端口为 0-输出群组;所述网络集线器的每个输出端上还串接 有地址过滤-单元。
一个 2G-to-G集线器的结构, 就是用 2个 G X G任意 0-1二元序列排序器网络后接 一级 k = G半清器。 自然地, 可以把 G个最大排序输出端口看作 " 1-输出群组", 其余 n 个小的排序输出端口作为 "0-输出群组"。 由于在自路由交换结构中, 每个时隙每个输入 端口可能空闲没数据, 也可能有数据发到 " 1-输出群组" 或 "0-输出群组", 故至少有三 种情况。
尽管前面一直讨论的是对 0- 1二元序列的排序器。 但根据 '0-1,定理可知, 任何能对 0- 1二元序列准确排序的排序器也能对由任意数字組成的序列排序。 因此, 上面构造的排 序器可以对上述三种输入分组状态的两位信息排序, 排序大小按 1-输出群组, 空闲, 0- 输出群组。 这样, 在每时隙中如果输入到某群组的分组个数超过 G时, 就会被错误地自 路由到另外一个群组地址, 为此在每条群组输出线的后面对分组地址进行识别, 阻塞掉 目标地址被误路由的分组, 剩下正确路由的分组, 最多 G个。 图 16就是 G=4的 2G-to-G自 路由群组集线器实例。
对于组播而言,组播集线器定理为以通过排序单元多级互联网络构建的 n-to-m的集 线器为参考, 将其中的排序单元用双播单元代替。 使信号表为 nbicast, 不妨设, V。个输入 值是 0-bound , 个输入值为 1-bound , 以及 VB个输入值为双播。 可得到以下结果: 上部 n-m个输出端口最大可能产生共 min{n-m,V。+VB}个 0-bound和双播信号。下部 m个输出端口 最大可能产生共 minim ^ + VB}个 1-bound和双播信号。
而布尔网络是一个多级互联网络, 其中所有的节点都是布尔单元。 更进一步, 当其 构建成集线器时布尔集线器网络。 每一个布尔网络等同于有布尔单元组成的多级互联网 络。
组播集线器定理在有一个适当的组播信号表 nbkast时为输入时,能够达到最优的组播 交换。 当用布尔单元代替集线器网络中的排序单元时, 对任意一个格或分配格结构的信 号表, 该定理也成立。 由此引发一个疑问: nbicast格结构的什么本质属性引出了组播集 线器定理? 仔细观察发现 Qbicast在声明 (7) 中分为上部理想 {Ο,Β}和下部理想 {1,1}, 在声 明 (8) 中类似的分为 {Ι,Β}和 {0,1}。
格 Ω的非空子集 S 如果满足条件 xe S, ye S → xAye S, xVye S则 S为子格; 如果 XG S, £Ω→ xAyG S 则子格 S是一个上 [^理想; ^口果 x £ S, y £ Ω→ xVyG S则子格 S是一 个下部理想。
两个格之间的映射如果其布尔运算保持不变的话则称映射为格同态。
如从格 Ω^]Ω2的格同态等同于格 Ω划分为一个上部理想和一个下部理想。
从格 Ω映射到 Ω2的一个同态 μ包含 Ω的划分 上部理想 μ- O)和下部理想 μ- ΐ)» 相 反地将格 Ω的划分为上部理想 U和下部理想 L, 若对 S 6 U, μ(5) = 0和 s £ L, μ(5) = 1, 可以 得到 μ是格 Ω到 Ω2同态。
针对一个 n-to-m由排序单元多级互联网络构成的集线器, 用布尔单元替换多级互联 网络中所有的排序单元。 就得到一个 n-to-m布尔集线器网络称为布尔集线器。 将任意的 分配格 Ω划分为上部理想 U和下部理想 L。 从 U中输入 u个值, 0≤u≤n, 从 L中 n-u个值 进入集线器网络之后: 上部 n-m个端口输出 min{n_m,u}属于 U的值; 下部 m格端口输出 min{m,n _ u}属于 L的值。
根据布尔集线器的 0-1原理可知上述的单元为一个 n-to-m布尔集线器。 令 μ表示 s G U, μ(5) = 0和 s e L, μ(5) = 1中 Ω到 Ω2的同态。 用 μ(5)替换每一个输入信号 S。 这就将 n个输入信号值转变为 u个 0的和 n-u个 1的组合。 从集线器网络的性质可知, 上部 n-m端口 输出为 min{n_m,u} 个 0和下部 n端口输出 min{m,n _ u}个 1。 因为 μ是一个格同态, 级间有 同样的信号替代。 当替代后集线器的一个输出端口输出值 0或 1时, 则替代前其输出值分 别属于 μ- O) = U或 μ- ^l) = L。
针对由排序单元多级互联网络组成的 n-to-m集线器, 将其所有的排序单元用 bicast单 元来替换, 从而当信号表是 nbicast时, 以下声明中有一项成立:
上面的 n-m 个端口输出仅为 0-bound信号; 下面的 m个端口输出仅为 1-bound信号; 没有端口输出为 idle信号; 没有端口输出为 bicast信号。
缩写信号值 0-bound, 1-bound, bicast和 idle分别为 0, 1, B和 I。 bicast单元组成的多 级互联网络的精确输入组合为: Vo 个 0, 个 1, VB个 B, ¼个 I。 由于信号表为的 特殊性, 多级互联网络中的 bicast单元将进一步被布尔单元替换。 因此布尔集线器定理得 以应用, 保证了布尔集线网络构成。 现在, 将划分为上部理想 U = {0, B}和下部理想 L = {1,1}。 为了评估输入 n-元组的 n个 变量的初始对称多项式 om+1, 贝' J:
(1)如果 ^ + ^ ≥ 01 + 1, 可得 om+1的值不是 1就是 I, 因为 om+1中的一些单项式只与 n个变量中赋值为 0或 B的值的变量有关。
(2)如果 V1 +V2≤ m + 1, 可得 am+1的值不是 0就是 B, 因为 om+1中的一些单项式只与 n个变量中赋值为 0或 B的值的变量有关。
另一种情况是,将理想的分为上部分 U = {Ι,Β}和下部分 L = {0,1} 0同理可得以下结论:
(3 )如果 ≥m+l, 可得 am+1的值是 1或
( 4 )如果 +V2 < m+ l, 可得 am+1的值是 0或 I。
假设(2) 和(4)成立, am+1唯一可能的值就是 0, 可知上面的 n-m 个端口输出仅 为 0-boimd信号成立。
对称地, 假设(1)和(3 )成立, am+1唯一可能的值就是 1, 可知下面的 m个端口输 出仅为 1 -bound信号成立。
殳设 (2) 和( 3 )成立, am+1唯一可能的值就是 B, 因此 am≥ om+1 = B。 上部分 w-W 个端口输出不可能为 I。 因此下部分 m个端口输出也不可能为 I。 所以没有端口输出为 idle 信号可以成立。
由对称的特性可得, 在( 1) (4)成立的情况下 £设下没有端口输出为 bicast信号也 是可以成立的。
所以, 0-bound和 bicast信号总数在多级互联网络被逐级保留, 1 -bound和 bicast信号 总数亦然。 因此上述针对由排序单元多级互联网络组成的 n-to-m集线器的 4个声明中任意 一个成立应该可以说是布尔组播集线器原理的一个详细版本。 实际上这个版本表明一个 更一般的事实: 布尔集线器理论不仅仅可以把尽可能多的信号路由到目的输出群组, 而 且可以依据优先级的不同而达到最佳路由方式。 在这里"优先,,概念需要一个解释一下, 因为信号表只是假设为分配格的非有序集合。 这个定义是适合各种可能的方式去划分分 配格为上部理想和下部理想。 通过以下例子来说明。
图 17 (a)示出了附加优先级组播信号表的分配格; 而图 17 (b)表示的格不是可分配 的, 因为 (ΙνΒ+)Λΐ = 1≠1 = (ΐΛΐ)ν(Β+Λΐ)。 如果让 Ω = {0+, 0- I, Β, 1-, 1, 1+}为在图 17(a)中的 分配格, Ω中元素的命名规则为: 上标 '+,为到达期望集线器的目的地址 0或 1的最高优先 级, '-,表示最低优先级。 Ω中所有通过一个 n-to-m布尔集线器的路由信号的优先级涉及了 所有可能的对 Ω的划分为上部理想 U集合和下部的理想 L集合。 这里可以组合:
• U = {0+}; L = {0— ,Ι,Β,Ι— ,1,1+} • U = {0+,0- ,1}; L = {B,l- ,1,1+}
• U = {0+,0— ,B}; L = {1,1— ,1,1+}
• U = {0+,0— ,Ι,Β,Ι— }; L =
• U = {0+,0— ,Ι,Β,Ι— ,1}; L =
从布尔集线器定理可以看出, 在路由到输出群组 0的时候 U中任一个元素的优先级 都比 L中任一个元素优先级高, 相反路由输出群组 1也是如此。 应用这个理论到上面所列 出的 5个中划分方法可以得到以下结论:
( 5 )在有序子集 {0+, 0- I, B, 1- 1, 1+}中的信号中去往输出群組 0的较小的元素¾武予较 高优先级, 然而去往输出群组 1的信号被赋予较高的优先级。 布尔集线器的路由最佳性 是和这个优先级的策略相一致的。
( 6 ) 类似的优先级处理同样可以应用在有序子集 {0+, 0-, I, B, 1-, 1, 1+}中。
(7) 同时, B和 I这两个信号在路由向任意输出群组时被赋予同样的优先级, 这就意 味着, B和 I不能同时出现在彼此对立的输出群组中。
图 18示出了一种实际的集线器, 其中, 图 18(&)示出了应用布尔集线器理论的840-3 的集线器, 有上标为 '+,的信号是具有高优先权处理的。 图 18(b)当信号表是图 17(b)中非 分配格的时候, 理想的交换是得不到保障的。 如图 18(a)实例证明来自分配格 Ω的信号通 过一个 8-to-3布尔集线器网络的传播过程。 0,1, B, I分别表示 0-bound, 1-bound, bicast 和 idle。 图中阴影部分的圓圏表示发生组播的信号, 在这里 bicast和 idle信号被 0-bound和 1-bound信号所替换。 这样输出群组 0的输出为 0-bound和 bicast信号总量中的最大的那个, 然而输出群组 1的输出则为 1-bound和 bicast信号总量中最大的那个。 此外, 这个路由最优 是和(5)和(7)相一致的。
例如: 图 18 ( b )例证了图 17(b)中非-分配格 Ω = {0+, 0-, I, B, 1-, 1, 1+}中信号值通过同一 个 840-3布尔集线网络的传播过程。 因为在这个格中 Β+ΛΙ = 0+ 和 B+Vl = l+,一个高优 先级的 bicast信号在一个布尔单元中遇到一个 idle信号输出高优先级 0-bound和 1-bound信 号。 结杲却只有一个有效信号出现在输出群组 0。 是不是输入信号已经以另外一种方式 ——高优先级的 bicast信号在 1级时遇到一个 idle信号发生了交换呢? 这也本应该在输出 群组 0有两个有效信号。 这个次优的路由结果反应了在定义布尔交换, 集线器或排序器 中分配格的信号表是必不可少的。
尽管在实际中应用中甚至有时候非分配格被用作信号表, 当高优先级组播通信在整 个通信量中占^艮小一部分时布尔集线器理论依旧能够被统计应用。
高度模块化、 器件复杂度低的完全自路由的分治网络可为本研究中超大规模线速组 播路由交换结构提供基本的网络结构模型。 为了应用于大规模组播交换, 一种构建大规 模几乎无阻塞交换结构的方法是把 "统计的线组 "技术应用到分治网络中。 将自路由分治 网络中的每一个 2x2节点放大为 2Gx2G的节点,并用 2G-to-G的 fast knockout 集线器替换, 网络中每一根连接线用一束 G根线替换, 从而构造了一个带有统计复用特点的多路径自 路由结构, 每 G根的输出线共享一个 G-bit的地址。 由于通信的波动和突发性引发分组的 丢失率随 G的值的增大成指数级减小。 如图 19就是一个 N=128, M=16 , G=8的多路径路 由交换结构。该结构是将 16x16的 banyan网络中每一个 2x2节点替换为 2G-to-G的有布尔单 元组成的布尔集线器来实现的。 这里 G=8,在实际应用中 G本应该为一个大的数, 所以一 个 2nx2n banyan-type的 G-line版本的网络构建成了一个 N Ν的几乎无阻塞的组播交换器。 根据 fast knockout集线器构造算法或 bitonic circula再配合一级半清器(half-cleaner ) , 可 构造任意 G为任意大小的群组集线器。 在本实施例中, 上述网络结构得到的过程如下: 构建一个使用版图复杂性最优的分治网络结构的自路由结构网络, 所述自路由结构网络 包括多个 2 2路由单元及其间的连线; 使用上述网络集线器逐一代替所述 2 2路由单元, 并将每条所述连线使用 G根线的线束代替; 得到具有 M个输出群组, 每群组包含 G个输出 端口的 N x N组播交换网络; 其中 G表示为群组, 将 G根线组成的一束记为一个群组, 这 个值一般比较大, M表示群组数, N表示交换结构的输入 /输出的总线数 N=M*G。
在图 19中 N=128 , M = 16 , G = 8 , 将集线器与可路由网络相结合构造的 "多路径自 路由结樹'^ ASC。 一般的, 设 N=2n , N = M G, M=2m,G=2g, 先构造一个 M x M的 可路由网络 (通常选择版图复杂性最优的分治网络)。 然后将网络中各级 2 2路由单元替 换为 2G-to-G自路由布尔组播集线器。 这样就建立了一个拥有 M个输出群组, 每群组包含 G个输出端口的 N X N组播交换网络,如图 21。
本实施例提出了一种基于代数格线速分组组播交换结构, 该结构具有以下特点: 组 播交换结构模块化且元件复杂度低, 有由数学理论支持的可递归扩展模型; 组播以物理 数据链路的线速扇出拷贝完成, 低时延无抖动; 组播追求用户质量体验(QoE )及服务 质量 (QoS ) , 对视频不追求完全无阻塞; 不存在资源带宽及计算处理能力的瓶頭, 通 过可能的接入允许控制算法来决定用户加入组播或新接入的请求, 而放弃每信元时隙的 端口调度。
以上所述实施例仅表达了本发明的几种实施方式, 其描述较为具体和详细, 但并不 能因此而理解为对本发明专利范围的限制。 应当指出的是, 对于本领域的普通技术人员 来说, 在不脱离本发明构思的前提下, 还可以做出若干变形和改进, 这些都属于本发明 的保护范围。 因此, 本发明专利的保护范围应以所附权利要求为准。

Claims

权利要求书
1、 一种用于组播的基本自路由单元, 其特征在于, 所述基本自路由单元包括两个 输入端口和两个输出端口, 所述输入端口包括第一输入端口和第二输入端口, 所述输出 端口分别为第一输出端口和第二输出端口; 所述输入端口输入的信号包括路由信号加上 其后数据属性和数据内容; 所述路由信号具有代数格的结构; 所述路由信号包括双播信 号、 单播信号和空闲信号; 当所述两个输入端口的路由信号分别为双播信号和空闲信号 时, 所述路由信号为双播信号的输入端口分别与所述第一输出端口和第二输出端口连接, 所述第一输出端口输出的路由信号值为两个输入路由信号的布尔积, 所述第二输出端口 输出的路由信号值为两个输入路由信号的布尔和。
2、 根据权利要求 1 所述的用于组播的基本自路由单元, 其特征在于, 当所述两个 输入端口输入的路由信号均指向两个输出端口中的一个时, 所述两个输入端口依据其输 入信号的数据属性中的数据优先级争用所述被指向的输出端口, 所述两个输入端口中数 据优先级高的与所述输出端口连接, 数据优先级低的与所述的另一输出端口连接; 当所 述两个输入端口输入的路由信号分别指向不同的输出端口时, 所述输入端口分别与所述 输出端口交叉或平行连接; 所述交叉连接包括第一输入端口与第二输出端口连接, 所述 第二输入端口与所述第一输出端口连接; 所述平行连接包括所述第一输入端口与第一输 出端口连接, 所述第二输入端口与所述第二输出端口连接。
3、 根据权利要求 2 所述的用于组播的基本自路由单元, 其特征在于, 所述路由信 号具有代数格的结构包括使用代数格构建自路由带内路由信号表, 所述代数格是分配格。
4、 一种由如权利要求 3所述的基本自路由单元构成的半清器, 其特征在于, 所述 k 个依次排列的 2 X 2双调排序单元, 所述 2 2双调排序单元包括两个输入端和分别输出 其输入信号中值较小的到低输出端和值较大的到高输出端; 所述 k个双调排序单元中的 第 n个的一个输出端为所述半清器的第 n个输入端, 其另一个输出端为所述半清器的第 k+n个输入端; 所述 k个双调排序单元中的第 n个的低输出端为所述半清器的第 n个输 出端, 其高输出端为所述半清器的第 k+n个输出端; 所述半清器的第 1个输出端至第 k 个输出端输出一个双调序列 al , 所述半清器的第 k+1个输出端至第 2k个输出端输出一 个双调序列 a2, al < a2; 其中, k为正整数, n=l, 2, ..., k;所述 2 χ 2双调排序单元为 所述基本自路由单元。
5、 根据权利要求 4所述的半清器, 其特征在于, 当 k=l 时, 所述半清器为所述基 本自路由单元。
6、 一种由如权利要求 5 所述的半清器构成的双调排序单元, 其特征在于, 所述双 调排序单元包括 G个输入端和 G个输出端, 所述 G = 2 ,g为正整数; 所述双调排序单元 包括 g级, 其中, 第 m级包括 2111-1个 k=G/2m的半清器, 其中, m=l,2 , ...,g; 所述每级 半清器包括多个 2 X 2双调排序单元, 每个 2 X 2双调排序单元的输出端分别连接在其下 一级半清器所包括的不同 2 x 2 双调排序单元的输入端或其下一级中不同的半清器输入 端。
7、 一种由如权利要求 6 所述的双调排序单元构成的任意二元系列排序单元, 其特 征在于, 所述任意二元排序单元包括 G输入端和 G个输出端, 所述 G=2g, g为正整数; 所述任意二元排序单元包括 g级双调排序单元, 其中, 第 p级包括 28·ρ个 G=2P的 G x G 双调排序单元; 所述双调排序单元按照其所在的级依次连接。
8、 一种由如权利要求 7 所述的任意二元系列排序单元构成的网络集线器, 其特征 在于, 所述网络集线器包括 2G个输入端和 2G个输出端, 所述网络集线器有 2个 G x G 任意二元排序单元和与所述 2个 G X G任意二元排序单元的输出连接的 k=G的半清器组 成, 所述半清器的 G个最大排序输出端口为 1-输出群组, 所述半清器的 G个最小排序 输出端口为 0-输出群组。
9、 根据权利要求 8 所述的网络集线器, 其特征在于, 所述网络集线器的每个输出 端上还串接有地址过滤单元。
10、 一种使用如权利要求 9所述的网络集线器构建組播交换网络的方法, 其特征在 于, 包括如下步骤:
A ) 构建一个使用版图复杂性最优的分治网络结构的自路由结构网络, 所述自 路由结构网络包括多个 2 X 2路由单元及其间的连线;
B )使用如权利要求 9所述网络集线器代替所述 1 X 2路由单元, 并将每条所述 连线使用 G根线的线束代替;
C )得到具有 M个输出群组, 每群组包含 G个输出端口的 N X N组播交换网絡; 其中, N表示所述组播交换网络的输入 /输出总线数, N=MG。
PCT/CN2012/085793 2012-12-04 2012-12-04 基本自路由单元及其半清器、排序单元、网络集线器和组播交换网络构建方法 WO2014085972A1 (zh)

Priority Applications (3)

Application Number Priority Date Filing Date Title
PCT/CN2012/085793 WO2014085972A1 (zh) 2012-12-04 2012-12-04 基本自路由单元及其半清器、排序单元、网络集线器和组播交换网络构建方法
CN201280077284.3A CN105556904A (zh) 2012-12-04 2012-12-04 基本自路由单元及其半清器、排序单元、网络集线器和组播交换网络构建方法
US14/711,802 US20150319080A1 (en) 2012-12-04 2015-05-14 Basic self-routing unit and method for building its half-cleaners, sorters, network concentrators and multicast switching network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2012/085793 WO2014085972A1 (zh) 2012-12-04 2012-12-04 基本自路由单元及其半清器、排序单元、网络集线器和组播交换网络构建方法

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US14/711,802 Continuation-In-Part US20150319080A1 (en) 2012-12-04 2015-05-14 Basic self-routing unit and method for building its half-cleaners, sorters, network concentrators and multicast switching network

Publications (1)

Publication Number Publication Date
WO2014085972A1 true WO2014085972A1 (zh) 2014-06-12

Family

ID=50882739

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2012/085793 WO2014085972A1 (zh) 2012-12-04 2012-12-04 基本自路由单元及其半清器、排序单元、网络集线器和组播交换网络构建方法

Country Status (3)

Country Link
US (1) US20150319080A1 (zh)
CN (1) CN105556904A (zh)
WO (1) WO2014085972A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104917701A (zh) * 2015-06-03 2015-09-16 香港中文大学深圳研究院 具有网络编码功能的代数交换系统及其voq调度算法

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108536612B (zh) * 2017-03-03 2021-12-21 西部数据技术公司 确定用于非易失性存储器中的地址映射的控制状态
CN114785746B (zh) * 2022-04-19 2023-06-16 厦门大学 一种用于单组播混传的banyan网络

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101141660A (zh) * 2006-09-05 2008-03-12 北京大学深圳研究生院 自路由交换集线器及其方法
CN101141374A (zh) * 2006-09-05 2008-03-12 北京大学深圳研究生院 自路由集线器以分治网络构成交换结构的方法
CN101350779A (zh) * 2008-08-26 2009-01-21 北京大学深圳研究生院 基于自路由集线器的电路式分组交换方法
CN101394335A (zh) * 2007-09-20 2009-03-25 中兴通讯股份有限公司 一种同步数字交叉连接的自路由的方法及系统

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7106728B1 (en) * 2000-05-01 2006-09-12 Industrial Technology Research Institute Switching by multistage interconnection of concentrators
US6591285B1 (en) * 2000-06-16 2003-07-08 Shuo-Yen Robert Li Running-sum adder networks determined by recursive construction of multi-stage networks
US7970279B2 (en) * 2002-11-05 2011-06-28 Lightfleet Corporation N-way serial-channel interconnect
WO2009015689A1 (en) * 2007-07-31 2009-02-05 Telefonaktiebolaget Lm Ericsson (Publ) All optical batcher banyan switch, batcher switch, banyan switch and contention manager

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101141660A (zh) * 2006-09-05 2008-03-12 北京大学深圳研究生院 自路由交换集线器及其方法
CN101141374A (zh) * 2006-09-05 2008-03-12 北京大学深圳研究生院 自路由集线器以分治网络构成交换结构的方法
CN101394335A (zh) * 2007-09-20 2009-03-25 中兴通讯股份有限公司 一种同步数字交叉连接的自路由的方法及系统
CN101350779A (zh) * 2008-08-26 2009-01-21 北京大学深圳研究生院 基于自路由集线器的电路式分组交换方法

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104917701A (zh) * 2015-06-03 2015-09-16 香港中文大学深圳研究院 具有网络编码功能的代数交换系统及其voq调度算法
CN104917701B (zh) * 2015-06-03 2020-04-07 香港中文大学深圳研究院 具有网络编码功能的代数交换系统及其voq调度算法

Also Published As

Publication number Publication date
CN105556904A (zh) 2016-05-04
US20150319080A1 (en) 2015-11-05

Similar Documents

Publication Publication Date Title
WO2013123654A1 (zh) 一种ip组播数据的分层分发方法、系统
JPH0637796A (ja) パケット交換機の改良型マルチキャスト機構
JPH0637797A (ja) パケット交換網の予約リング機構
US20030231588A1 (en) Method and apparatus for multicast and unicast scheduling
Liew et al. Principles of Broadband Switching and Networking
Guo et al. Multicast ATM switches: Survey and performance evaluation
WO2014085972A1 (zh) 基本自路由单元及其半清器、排序单元、网络集线器和组播交换网络构建方法
Sun et al. Multicast scheduling in buffered crossbar switches with multiple input queues
Mir A survey of data multicast techniques, architectures, and algorithms
Hassen et al. A multi-stage packet-switch based on NoC fabrics for data center networks
Kannan et al. MSXmin: a modular multicast ATM packet switch with low delay and hardware complexity
Zhang et al. Performance analysis of k-fold multicast networks
Li et al. Multi-path Self-routing Switching Structure; by Interconnection of Multistage Sorting Concentrators
Chen et al. Multicast Switching Fabric Based on Network Coding and Algebraic Switching Theory
Senin et al. Efficient multicast support in buffered crossbars using networks on chip
Xu et al. Improving of Load-Balanced Self-routing switch by network code
Chen et al. PPCN: A high-performance copy network for large scale ATM switching systems
Mir et al. Efficient architectures and algorithms for multicasting data in computer communication networks
Chen et al. LB-MSNC: A load-balanced multicast switching fabric with network coding
Chen et al. PPCN: A High-Performance Copy Network for Large Scale ATM Switching Systems
Kim et al. Topology-aware fanout set division scheme for QoS-guaranteed multicast transmission
Mhamdi et al. Multicast Traffic Scheduling Based On High-Speed Crossbar Switches
Guo et al. Design issues for multicast ATM switches
Tripathi et al. New multicasting algorithm for a modified-cyclic-Banyan based ATM switch
Mir An efficient switching fabric for next-generation large-scale computer networking

Legal Events

Date Code Title Description
WWE Wipo information: entry into national phase

Ref document number: 201280077284.3

Country of ref document: CN

121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 12889515

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 12889515

Country of ref document: EP

Kind code of ref document: A1