KR101060190B1 - Optical Gigabit Ethernet Switch that provides IPMP functionality for HCP network IPTV services - Google Patents

Abstract

The present invention relates to an optical gigabit Ethernet switch providing an IGMP function for an HFC network IPTV service. The present invention relates to a network transceiver for transmitting and receiving a packet by accessing Gigabit Ethernet, and to a voice packet, an image packet, and data in a packet received through the network transceiver. A QoS logic section for distinguishing packets and providing a QoS suitable for the packet, and a memory section for temporarily storing data or packets processed by the QoS logic section, wherein the QoS logic section refers to a predetermined reference by referring to a header of the packet. Packet classifier to classify according to, the packet marker to mark with a data tag promised in advance in the header of the classified packet, to control the rate of traffic so that the packet is not sent out over a preset bandwidth by measuring the amount received A policy and a packet marked when the congestion occurs according to a predetermined rule. And a queue scheduler temporarily storing the memory unit.

Gigabit Ethernet Switch, QoS, Packet

Description

Optical gigabit ethernet switch which provides Internet Group Management protocol function for HFC network IPTV service}

The present invention relates to an optical gigabit Ethernet switch providing an IGMP function for an HFC network IPTV service, and an optical gigabit ethernet switch for enabling triple play service (TPS) of voice, video, and data in an HFC network. L2 Switch).

Hybrid optical coaxial cable (HFC) network was considered as a powerful medium for next-generation network infrastructure with the launch of comprehensive cable broadcasting in 1995, and started the first two-way cable modem pilot service in Korea in 1997 It is a network with a long history of establishing itself as a business flagship. As of the end of April 2002, the optical network is the Internet network used by more than 3 million people, about 35% of the 8.5 million high-speed Internet subscribers in Korea, along with ADSL and VDSL, making Korea the world's No. 1 in broadband distribution. Despite his reputation as a top celebrity, his reputation has been undermined by the publicity of DSL. The general public is familiar with ADSL and VDSL but not with optical network. In fact, the number of high-speed Internet subscribers using optical networks rather than xDSL is significantly higher in foreign countries, and even in the United States, optical users are more than double DSL users. Recently, as the name of FTTH (Fiber To The Home) becomes a reality due to the emergence of Passive Optical Network (PON), the prediction that the mainstream of the next generation network is optical is gaining strength. Interest in Internet access services on the network was behind.

However, looking at the recent trends at home and abroad, it can be seen that the position occupied by xDSL is gradually giving way to PON, and the penetration rate of cable modem is increasing again. In Korea, Powercomm started to enter the high-speed Internet market in September 2005. It excluded xDSL from business and focused marketing as a flagship product of 100Mbps apartment LAN-based Xspeed LAN. Emphasis was placed on driving the subscribers' driving force. However, according to the survey in early 2006, only 5,8787 100Mbps X-Speed optical LAN subscribers of Powercomm's subscribers are compared to 12,3609 subscribers of 10Mbps cable modem X-speed prime subscribers. This is because Powercom concentrates its marketing on 100Mbps optical LAN, giving customers the recognition that it is 'fast' in the market, and provides the service through the cable network that was secured by the existing network lease. It is known because it is accommodated as a fruit. In the case of Hanaro Telecom, a 100Mbps optical LAN service was launched at the end of 2006 using the HFC network of Telest, a Finnish telecommunications equipment company. Hanaro Telecom's 100M-class optical LAN service installs master equipment in the company, modulates 100M Ethernet signals to transmit them to the optical network, and installs slave equipment on the subscriber's side. Is a technology that provides 100M service of bi-directional symmetry by demodulating Ethernet signal by directly connecting with UTP cable. In Japan, the Yano Institute of Economics released a survey on Japan's broadband market in November 2007, on November 22, 2007. The number of broadband contracts in Japan was 3,030,000, an increase of 3.89 million from the previous year. The number of DSLs increased by 4.33 million to about 13.13 million, but the DSL was about 13.33 million, down 720,000 from the previous year. CATV access rose 290,000 year-over-year to about 3.9 million, and FWA fell 0.80,000 year-on-year to about 0.4 million. The number of contracts by line type in 2012 is estimated to be 32.03 million FTTH, approximately 7.75 million xDSL, and approximately 484 million CATV access.

Along with the FTTH market, the main reason for the cable modem market to rebound is due to the fact that xDSL using high frequency is a technical limitation, and in terms of stability, FTTH using ETTH (Ethernet-To-The-Home) or PON, and HFC network It is analyzed that it is gradually giving the seat back to the used cable modem. Taking full advantage of the unmodified PSTN network, which has the most extensive network infrastructure, carriers have been providing services in ADSL, 10 Mbps VDSL, and 50 Mbps VDSL for the past four to five years. The problem with the stability of the solution itself is growing, and the solution to this is far from a long list. The real concern for vendors developing and supplying DSLAM is that it is virtually difficult to resolve, which means that the given environment is too poor to lead the technology to deliver the highest speed services through the lowest quality media (PSTN). Korean and Japanese carriers, which have continued to compete indefinitely, have found it difficult to recognize VDSL as a competitive medium due to the limitations of speed and stability, and they have preferred FTTH using PON as a major medium to replace it. . However, this is the area of the telecommunications carrier that has sufficient optical lines. Otherwise, a cable modem using an optical network is in the spotlight as another alternative. This is because cable modems have come up with solutions that solve the speed limitations that had previously been available to xDSL. DOCSIS 3.0, which can service more than 100Mbps, and its corresponding non-standard cable modem technology, are emerging on the market. have. The largest increase in the number of FTTH subscribers, cable modem subscribers, and decreasing number of xDSL subscribers, both in Korea and Japan, are a testament to the technical limitations of xDSL and the transition of network infrastructure to new media. It is evidence that it has begun. Therefore, it is necessary to develop a device suitable for such a network environment.

The present invention is to develop a Gigabit Ethernet Switch (Tigabit Ethernet Switch) to enable a triple play service (TPS) of voice, video, and data in a hybrid fiber cable network (HFC network), more specifically, DOCSIS1 in an optical network DOCSIS 3.0 provides a stabilized packet network solution by installing the QoS (Quality of Service) function required for TPS in the packet switching technology of CMTS (Cable Modem Termination System), which is applied on the basis of .0 / 2.0. To provide a network equipment corresponding to the CMTS.

An optical gigabit Ethernet switch providing an IGMP function for an HFC network IPTV service according to an embodiment of the present invention. A QoS logic section for distinguishing an image packet and a data packet and providing a QoS suitable for the packet, and a memory section for temporarily storing data or packets processed by the QoS logic section, wherein the QoS logic section includes a header of the packet. A packet classifying unit classifying the packet according to a predetermined criterion, a packet marker unit marking a data tag previously promised in the header of the classified packet, and measuring the received amount of the packet so that the traffic is not sent out over a preset bandwidth A policy for controlling the speed of the packet, and a packet marked when the congestion occurs In accordance with a predetermined rule includes a scheduler queue to temporarily store in the memory unit.

A packet network solution according to an embodiment of the present invention can provide a CMTS as a packet processor in an optical network, and a new voice / video service business by implementing a TPS-capable packet switching technology. A model can be provided.

The activation of the new business model can activate local cable operators' high-speed Internet business and provide various services to subscribers using the widespread already installed optical network. In addition, new business models can be created by realizing VoIP over cable networks. In addition, it provides a high performance CMTS, enabling the domestic and overseas markets.

Hereinafter will be described in detail for the practice of the present invention. The contents described herein do not limit the scope of the present invention, but are described by way of example. Accordingly, the content described herein may be modified and applied by those skilled in the art, and such applications are also included in the claims of the present invention. The process or unit described in the present invention may be provided through one chip or board, or may be configured through a combination of multiple chips and boards. That is, the process or unit described in the present invention is a logical unit necessary for implementing the present invention, and is not limited to the physical unit.

1 is a view showing a service overview of the HFC network in which the Ethernet switch of the present invention operates.

The most important element of packet switching in the optical network is the Cable Modem Termination System (CMTS), which allows each cable modem to connect to the upper network. Although it has been largely used for low speed internet service by current DOCSIS 1.0, higher speed DOCSIS 3.0 may require enhanced QoS and network security. Currently, the major semiconductor chips needed to implement DOCSIS 3.0 are being studied.

The widespread use of optical networks already in place in the region is yet another trend along the shoulders of FTTH, as described above, but with data networks in the past when xDSL and cable modems competed for speed. This is quite different from the current data network. In other words, the business strategy of winning at a high speed as in the past does not fit the present reality at all. As IP-TV became active in Europe, the term TPS (Triple Play Service) was created before and after. TPS, which refers to the type of service that provides voice, video, and data in one medium, is based on each traffic class. Through bandwidth control, network service that meets its characteristics is required. That is, voice has the smallest amount of data, but is an interactive service, so it is sensitive to network jitter or delay, so that an appropriate bandwidth should be allocated and guaranteed. For example, broadcasting services such as VoD and IP-TV use end-node's own buffering, which makes them less susceptible to jitter or delay. On the other hand, although the amount of data is quite large, data services do not require high quality. HD-quality high-definition broadcasting service is possible in the current optical network (although there will be a change in SO cable transmission equipment), so there is no problem in watching TV in the future, and data service is provided through a separate frequency band. The communication between the cable modem and the CMTS is through communication, so there is no need to change the service method. Therefore, IP-TV may not be necessary in the optical network, but considering the interactive service characteristics of the IP-TV itself, the cable network may be broadcast through the CMTS and the cable modem in the optical network. Service business is possible. In other words, if the IP-TV service is not provided by the cable broadcasting station, the user will not be able to do home shopping or pay any fees through the TV remote control at home.

A packet switching platform, which is an embodiment of the present invention, accommodates QoS functions to enable such TPS and manages subscribers connected in a serial manner on a cable. In conventional optical LAN, AON (Active Optical Network) switch can be used to distinguish subscribers by port.However, cable modem can be connected by coaxial cable with bridge tap. none. Subscriber management is essential for billing and failover for local SOs to complete data services.

2 is a diagram illustrating a configuration of a gigabit Ethernet switch according to an embodiment of the present invention.

The gigabit Ethernet switch 200 may serve as an aggregation switch in a ME (Metro Access Network). Gigabit Ethernet switch 200 includes the following elements. The gigabit Ethernet port 220 may be configured in a 1RU size of 19 inches rack mountable in order to increase the port density of the port. In addition, more than 24 Gigabit Ethernet Fiber ports can be connected to multiple devices, and the optical interface allows the operator to mount and use the desired optical module (SFP cage). .

It can provide full wire speeds to fully serve as an L2 Gigabit Ethernet Switch.

For the Triple Play Service (TPS), QoS functions such as classification, rate control, and filter by the L2 / L3 / L4 field must be provided. Each QoS provided by the Gigabit Ethernet switch, which is an embodiment of the present invention, will be described later. The capacity of memory should be sufficiently considered to develop and mount the unicast / multicast routing protocol function. In order to act as a CMTS in the HFC network, a DHCP server, relay, and snooping function may be supported for each MAC address.

3 is a diagram illustrating a configuration of a QoS model of a gigabit Ethernet switch implemented by an embodiment of the present invention.

In order to implement high-speed DOCSIS 3.0 in the gigabit Ethernet switch 200 of FIG. 2, various functions must be provided in a quality of service (QoS) for processing a received packet.

Classifier 310 determines the QoS class by referring to the header of the received packet. The markers (Traffic Marker, Packet Marker, 320) indicates the QoS level in the packet header (802.1p, MPLS EXP, IP TOS / DSCP / IP- Prec).

The Polymer 330 measures the packet inflow rate (bps) so that it does not send more than the predetermined bandwidth, and the Buffer Manager (340) determines how and what packets to discard when congestion occurs. To judge. The queue scheduler 350 decides which packets to send preferentially when congestion occurs.

In the existing best effort network where QoS has not been applied, the method of providing QoS can be divided into a method of increasing network bandwidth and using complex QoS techniques. The former is called the Big Pipe approach or the Big Bandwidth approach, and the latter is called the Controlled Bandwidth approach. The Big Bandwidth method solves the QoS problem by increasing the bandwidth of the network. Increasing the number of links and replacing low-speed equipment with high-speed equipment is a simple solution, but can be expensive. In addition, QoS issues, including congestion, can occur at any time unless the bandwidth is increased indefinitely. Therefore, it is necessary to solve the QoS problem, but it is not the ultimate solution.

The controlled bandwidth method is a method of appropriately controlling traffic while providing a reasonable level of bandwidth using various QoS techniques. To this end, various QoS-related protocols and standards such as DiffServ, IntServ, MPLS, and IEEE 802.1p / Q, and QoS implementation techniques that specify them can be used. QoS enabling technologies include packet classification, mapping, metering, marking / remarking, rate limiting, shaping, and flow. ) Or congestion control. However, this approach also has limitations in terms of utility. Therefore, there is a need to consider both bandwidth increase at a realistic level and efficient and effective QoS implementation techniques simultaneously.

FIG. 4 is a diagram illustrating the structure of QoS in an edge node implemented in a gigabit Ethernet switch implemented by an embodiment of the present invention.

The application of QoS of FIG. 4 is premised on switching from a single service model of best effort to a multi-service model. These different service models are associated with different traffic classes, and they are handled with different QoS profiles. In general, the single service model of Best Effort is divided into multiple service classes at the edge or boundary nodes, and QoS corresponding to each class at the edge node and core node. It is processed with a profile.

5 is a diagram illustrating the structure of QoS in a core node implemented in a gigabit Ethernet switch implemented by an embodiment of the present invention.

To support QoS in a best effort network, all nodes that make up the network must have QoS. If a node in the middle does not support QoS and only the best effort service, no QoS characteristic can be guaranteed for the traffic passing through that node. In this regard, the user equipments must also be able to recognize QoS. However, at present, most user equipment transmits traffic in a best effort manner and separates these traffic at edge nodes, so it is strictly guaranteed to end-to-end QoS. It can be difficult. In addition, all devices must have the same QoS application structure to provide consistent QoS. If there are different QoS application structures, consistent QoS guarantees are not possible. In this case, it is necessary to organically connect traffic classes defined differently between network equipment or service domains having different QoS application structures. This function is achieved through a mapping function to be described later. The core node implementation method distinguished from the edge node implementation method of FIG. 4 is the same as that of FIG. 5.

The classifier described in the preceding drawings performs packet classification to classify packets.

Packet classification (packet classification) literally refers to the function of classifying input packets into several classes according to some criteria. Simplifying the QoS structure by processing packets with the same or similar characteristics together is the main function of packet classification. In other words, if each of the traffic having a variety of QoS characteristics are handled separately, there must be a number of packet processing criteria. This is practically impossible and, if possible, places a heavy load on the node. Therefore, the packet is divided into limited classes. There are some differences in the function of packet classification performed at the edge and core nodes. At the edge node, packets are classified into a limited number of classes by various criteria. Categorizing into a limited number of classes means determining which queue the incoming traffic will be stored on. However, all processing is done in per-flows defined in the access control list until it is actually queued. On the other hand, the core node uses the result classified at the edge node or reflects the change of the class of the packet during other QoS processing. That is, at the edge node, various types of flows are classified into several traffic classes. If the edge node has the same QoS application structure as the edge node, the core node maintains a small number of classes classified at the edge node. . Therefore, packet classification performed at the edge node is much more complicated than that performed at the core node and puts a heavy load on the node.

The classification of packets can be performed using several criteria. Packets can be classified based on the port or interface of the device where the packet is input, address information such as MAC or IP can be used, and field values of the packet header defined in the protocol can be used. Any one of these may be used to classify the packet, or a packet may be classified through a combination of several fields. As in the latter case, classifying packets using various field information of the packet header is called multi-field packet classification (MFPC or MFC). In addition, the packet may be classified based on user information such as the type of service subscribed to by the user or the affiliation of the user. In general, classifying packets using user information requires a user authentication procedure. However, using the user information or performed in the application layer can be variously implemented in software. In order to examine the classification of packets using the information between L2 and L4 among the OSI 7 layers, information between L2 and L4 should be used, and the header structure of frames or packets of L2 to L4, which are commonly used, is applied. can do.

FIG. 6 is a diagram illustrating an L2 frame structure transmitted and received by a gigabit Ethernet switch implemented by an embodiment of the present invention.

The 610 frame is a structure of a commonly used Ethernet frame, the 620 frame corresponds to the IEEE 802.1p / Q frame structure. In addition, the 630 frame structure shows the frame structure of MPLS over Ethernet. First, referring to the 610 frame structure, fields containing information for identifying a packet are the first three fields, that is, a DA MAC (Destination MAC Address) field, an SA MAC (Source MAC Address) field, and a Type field. The DA MAC field and the SA MAC field correspond to unique values of the device, and the Type field indicates the type of packet (protocol) included in the payload field indicated as IP datagram in the figure. There are various ways to classify packets using this information. First, SA MAC information can be used to separately classify frames coming from a specific host. Likewise, DA MAC information can be used to classify frames destined for a particular host. Information in the Type field may be used to separately classify packets using a specific protocol. In order to classify packets belonging to a specific protocol destined for a specific SA MAC to a specific DA MAC, three fields may be combined at a time.

In the case of the IEEE 802.1p / Q frame of 620, it can be seen that the TPID, PRI, CFI, and VLAN ID fields exist in addition to the DA MAC, SA MAC, and Type fields. The Priority field allows three bits to classify the packet priority into eight levels. That is, it means that L2 frame can be processed differently according to the value of this field. The VLAN ID field is a VLAN ID value as it is. That is, it is used to distinguish which VLAN a frame belongs to. This means that packets can be processed differently according to the type of VLAN. The values of the Priority field and the VLAN ID field may be combined with the values of the DA MAC, SA MAC, and Type fields to further classify the frames.

630 shows the MPLS over Ethernet frame structure. This frame structure also basically has a 20-bit Label field and a 3-bit Experimental field, in addition to the fields used for Ethernet frames. Compared to the IEEE 802.1p / Q frame structure 620, the Label field corresponds to the VLAN ID field and the Experimental field corresponds to the Priority field. That is, the Label identifies the path of the connection used in MPLS, and the Experimental field is used to indicate the priority of the frame. These may also be used in combination with field values used for other Ethernet.

When comparing the 610 frame structure of FIG. 6 with the 620 and 630 frame structures, the biggest difference is that the 620 and 630 frame structures have fields for indicating the priority of the frames. In practice, the values in this field allow for the provision of differentiated services for pure Ethernet frames. These values are also used for priority mapping with L3 IP packets.

FIG. 7 is a diagram illustrating an L3 frame structure transmitted and received by a Gigabit Ethernet switch implemented by an embodiment of the present invention. Like the L2 MAC frame of FIG. 6, a source IP address and a destination IP address may be used to distinguish a packet. In addition, a type of service (ToS) field and a protocol field are used to distinguish packets. The ToS field is 8 bits, and the IP Precedence field corresponding to the first 3 bits is used to indicate the priority of the packet. Therefore, it can be used by mapping one-to-one with the Priority field or Experimental field of the L2 frame. The Protocol field is used to indicate the type of L4 protocol used for the Data field.

FIG. 8 is a diagram showing the structure of a ToS field of an IP packet header transmitted and received in a Gigabit Ethernet switch and a DS field of DiffServ according to an embodiment of the present invention. The 8-bit ToS field 810 has a 3-bit IP Precedence field of 820 and an information field used for routing of DTRC. This field has been redefined as a DS field in the DiffServ 830. The DS field is composed of a 6-bit DiffServ Code Point (DSCP) field and an unused CU (Currently Unused) field. The first three bits of the DSCP field are used to indicate the class, and the next two bits are used to indicate Drop Precedence. The first three bits are compatible with IP Precedence in the IP ToS field. Thus, priority can be maintained consistent between L2 frames and L3 packets. Most of the packets used in L4 are TCP and UDP packets. They have a Source Port Number and Destination Port Number fields in the packet header and can be used to classify packets. Different L4 protocol packets each have a unique packet structure, and specific field values are used to distinguish different packets belonging to the same protocol. In general, L4 information is used together with L3 protocol information.

9 is a diagram showing the configuration of a traffic conditioner implemented in a gigabit Ethernet switch implemented by an embodiment of the present invention.

Traffic conditioning checks whether the traffic coming from the edge router follows the user's predefined traffic profile, and provides the basis for basic QoS processing or other QoS processing depending on compliance. Say process. The basic QoS processing referred to here may be such as marking a packet that changes the priority of the packet or packet drop through rate limiting. Other QoS processing means that the results of traffic conditioning can be affected during flow control or packet scheduling. These will be described later. A functional module that implements traffic conditioning is called a traffic conditioner, and its configuration depends on the scope of defining the traffic conditioning. When defining traffic conditioning in a broad sense, a traffic conditioner consists of a packet classifier, a meter, a marker, and a dropper or shaper. In a narrow sense, however, a traffic conditioner consists only of meters and markers, and includes only the ability to measure traffic and indicate compliance in the packet accordingly.

9 shows a traffic conditioner consisting only of meters and markers. Meters and markers are distinguished in functional terms, as shown in FIG. 9, but are generally present together. For this reason, a traffic conditioner is configured in a narrow sense. The meter enters the device to measure packet-sorted traffic flow. Typically, the input rate of a traffic flow measures bandwidth or bursts. The meter determines compliance by comparing the traffic profile of the input flow with the pre-appointed / defined traffic profile, and accordingly, the marker performs the necessary marking processing. In general, if a predetermined traffic profile is satisfied, a predetermined range is exceeded, and a predetermined range is exceeded, DiffServ marks packets as Green, Yellow, and Red in each case. Do In the case of Cisco equipment, the terms Committed, Excess, and Violated are used to refer to each case. In the case of DiffServ, single-rate three-color marker (sr-TCM) and two-rate three-color marker (tr-TCM) are used as representative markers or traffic conditioners. The sr-TCM and tr-TCM described in RFC 2697 and 2698 respectively use a double token bucket structure by default. While sr-TCM uses only the committed information rate (CIR) as the token update rate, tr-TCM uses both CIR and peak information rate (PIR) as the token update rate.

FIG. 10 is a diagram illustrating an operation method of sr-TCM implemented in a gigabit Ethernet switch implemented by an embodiment of the present invention.

If the size B of the input packet is smaller than the token counter Tc, it is marked as a green packet.If it is larger than Tc but less than Te, it is marked as a yellow packet, and if it is larger than Te, it is marked as a red packet. It shows the process of becoming.

Rate limiting and shaping are two representative techniques for controlling the bandwidth of traffic. Both rate limiting and shaping are implemented using Ricky buckets or token buckets. The only difference is that rate limiting does not use buffering and shaping uses buffering.

FIG. 11 illustrates a rate limit operation implemented in a gigabit Ethernet switch implemented by an embodiment of the present invention.

Also known as policing, rate limiting is a technique that literally limits the speed (bandwidth) of certain traffic. In order to limit the speed, it is basically used with a meter and a marker because the speed of the incoming traffic must be known. As shown in FIG. 11, the operation of the rate limiting discards all traffic coming in at or above the target traffic rate (1110-> 1120). In other words, drop excessive traffic without buffering.

The rate limiter is known to be located at the input portion of the device, i.e., in front of the switching fabric, but may be present in the output module at the back of the switch fabric or in both the input and output modules. However, since the shaper to be described later is mainly present in the output module, the limiter is located in an input module that does not use a buffer to avoid duplication with the shaper. Right limiting can be distinguished by several criteria. First, flow-based rate limiting and port-based rate limiting can be classified based on the target to which the rate limiting is applied. In other words, flow-based rate limiting limits the speed / bandwidth of traffic for each application flow, and port-based rate limiting limits the speed / bandwidth of traffic on a per port or interface basis.

Depending on how the rate limiting is applied, it may be divided into soft rate limiting and hard rate limiting. Hard rate limiting is a rate limiting method described so far, in which the incoming traffic does not follow the promised traffic profile, i.e., if it comes in above the promised bandwidth, it is discarded unconditionally. Soft-Rate Limiting is when the equipment is not in a congested state, exceeding the promised traffic profile within a certain range, marking down the low priority for the incoming traffic and then passing on to the next process. Packets marked with low priority may be discarded while competing with other traffic on the output side. Soft rate limiting can be provided by the service provider at the service level.

12 is a diagram illustrating a shaping operation implemented in a gigabit Ethernet switch according to an embodiment of the present invention. Shaping, like rate limiting, is a technique that limits the speed / bandwidth of any traffic. The difference from rate limiting is that it uses buffering.

By using buffering, traffic coming in above the target rate is temporarily stored and serviced later (1210-> 1220). The operating characteristics of this shaping are well described in FIG. This shaping has the advantage of being able to accommodate burst traffic to some extent and reducing packet loss rate and improving overall throughput compared to rate limiting. However, buffering can add additional delay, so using large buffers can affect real-time application traffic. Because shaping uses buffering, it is implemented where the buffer exists. It can be implemented on both the input and output sides, but is typically implemented on the output side where the buffer is located. This can also be understood as avoiding duplication with the implementation of rate limiting, which performs functions such as shaping on the non-buffered side.

One embodiment of the present invention enables flow control and congestion control.

Congestion control is a technique for solving congestion caused by input traffic or traffic concentrating on a specific port beyond the network equipment's ability to handle, i.e., the device continues to be overloaded. As a method for solving congestion, an access control, a scheduling method, and a flow control method may be used. Access control is a method of limiting the amount of traffic input to the equipment, and scheduling is a method of efficiently serving the traffic input to the equipment. Flow control refers to a set of techniques that allow a data source to match its transmission rate to those currently available in the network and receiver.

Flow control solves congestion problems by allowing a data source or a data source to reduce its transmission rate. The flow control may be divided into a method of predicting congestion and allowing flow control to be performed before congestion occurs, and a method of performing flow control after confirming congestion has occurred. In this case, the former is called proactive, and the latter is called responsive or reactive. Either way, the data source should basically have the ability to adjust the transmission rate. Much of the traffic on the Internet is TCP traffic, and the TCP server has this ability to adjust its transmission rate, allowing flow control. Therefore, we will look at the operation of TCP first, and then we will look at the flow control techniques RED and WRED using TCP's operating characteristics. TCP's operation operates in two parts: slow start and congestion avoidance. When the TCP starter's window size (CWND) is less than the threshold, the slow start is sent every time an ACK is received for a packet that was sent before a timeout occurred for the transmitted packet. The TCP sender doubles its transmission rate. Congestion avoidance refers to increasing the transmission rate by 1 MSS (maximum segment size) when the window size of the TCP sender is larger than the threshold. If a timeout occurs, the threshold is reduced to half the current window size and the window size is reset to 1 MSS.

When the TCP sender receives an ACK before the timeout

if CWND ≤ Threshold-> CWND = 2 * CWND

if CWND> Threshold-> CWND = CWND + 1

On the other hand, when the TCP sender does not receive an ACK before the timeout

Threshold = CWND / 2

CWND = 1 MSS

Figure 13 shows the operation of TCP implemented in a gigabit Ethernet switch according to an embodiment of the present invention. The section indicated by 1 corresponds to the slow start, and the section indicated by 2 is the congestion avoidance section. To examine the TCP-based flow control scenario, multiple TCP flows can use a buffer. This will be described in FIG. 14.

14 is a diagram illustrating a scenario in which multiple TCP flows implemented in a gigabit Ethernet switch using one buffer according to an embodiment of the present invention.

In this case, the queue size increases rapidly, and a buffer full occurs and an overflow occurs. Then, all TCP flows will lose all packets arriving after the buffer pool, and TCP senders will not receive an ACK for the lost packets until a timeout occurs. As a result, all TCP senders will slow down their transmissions and reduce queue sizes at a high rate. However, slow start causes the queue size to increase again and overflow, and the same process is repeated for all TCP flows. This phenomenon is called Global Synchronization. As a result of the sudden fluctuation of the traffic volume, the network equipment becomes unstable as well as performance. This global synchronization problem can be solved by selecting an arbitrary TCP flow in a fair way to reduce the transmission speed. One such method and the most representative one is the congestion control method by random early detection / discard (RED).

FIG. 15 is a view illustrating a process of controlling congestion by applying RED implemented in a gigabit Ethernet switch according to an embodiment of the present invention.

RED, also called Random Early Detection or Random Early Discard, is a representative congestion control technique using TCP behavior. As the name suggests, it is a way to cause certain TCP flows to reduce transmission speed by dropping packets in a random way before congestion occurs. The basic operation is to apply two different drop probabilities to the three intervals with two thresholds TH_min and TH_max in the queue. This process is shown in FIG. That is, if the average queue size is smaller than TH_min (1510), no packets are discarded and all are accepted. If the queue size is larger than TH_min but smaller than TH_max (1520), the packet is discarded with a certain probability according to the queue size. If the queue size is larger than TH_max (1530), all incoming packets are discarded. In other words, as the degree of congestion increases, it is a method of reducing the amount of incoming traffic by discarding many packets.

1510 shows a case where all packets arriving when the queue length is smaller than TH_min are accepted. 1520 shows a case where the length of the queue is between TH_min and TH_max. The 1530 discards all packets arriving when the length of the queue is greater than TH_max. In the case of 1520 and 1530, the ACK for the discarded packet does not arrive at the source, thereby reducing the transmission speed and achieving the purpose of flow / congestion control.

The probability of dropping a packet in each section divided by TH_min and TH_max is created as a Drop Probability Function or a probability table, and the drop probability values are summarized for the possible queue sizes.

FIG. 16 shows an example of the RED packet drop probability function used in FIG. 15. K is the maximum cue size and Pmax is the value that determines the slope between TH_min and TH_max. In general, if the value of TH_max is too small, packet drops may occur frequently and seriously affect the overall performance.Throwing all incoming packets above TH_max is the same as the result of the buffer overflow. It will be set to the same value as or close to the maximum size K. In addition, if the Pmax value is too small, the frequency of dropping packets may be low, and congestion control effects may not be properly exhibited. Therefore, when using RED, it is important to set the TH_min, TH_max, and Pmax values appropriately.

17 shows an example of a WRED packet drop probability function implemented in a gigabit Ethernet switch according to an embodiment of the present invention. Weighted RED refers to congestion control by applying a RED function with different characteristics (profile or weight) to one or several different class traffics. The RED function having different characteristics is a RED packet drop probability function having different TH_min, TH_max, and max p values. When WRED is applied to the same class traffic, different RED functions are applied to packets of different priority belonging to the same class. For example, in the AFx class of DiffServ, three different drop precedences exist in the same class, and different RED functions may be applied to each drop precedence. The graph of FIG. 17 shows an example of such a WRED packet drop probability function. This graph shows that different RED functions can be applied to packets marked green, yellow, and red for certain class traffic of DiffServ. As shown in the figure, the more aggressive packet drop probability function is applied to low priority packets or classes, and the conservative packet drop probability function is applied to high priority packets or classes. The figure shows the case where TH_max for C1 with low priority is set larger than TH_min for C2, but it may give the condition that TH_max for C1 must be less than or equal to TH_min for C2.

18 shows an example of applying WRED to one class using a single queue implemented in a gigabit Ethernet switch according to an embodiment of the present invention. Although packets belonging to the same class have different drop priorities, different RED drop probability functions apply. In FIG. 18, the packet is filled up to just below TH_max, yellow, and the new yellow packet is discarded unconditionally because it exceeds the TH_max, yellow threshold. If a green packet arrives instead of a yellow packet, it will be discarded with a certain probability since it is between TH_min, green and TH_max, green.

19 is a diagram illustrating the operation of queuing implemented in a gigabit Ethernet switch according to an embodiment of the present invention.

Queuing occurs when a packet arrives beyond what a network device can handle (service), or when there are packets destined for the same destination at the same time. In other words, queuing is the process of storing packets that cannot be processed at one time in a buffer for later service. Scheduling is a general term for the manner of serving the packets stored in the buffer. Scheduling algorithms must be simple, efficient and fair. Queuing and scheduling may be considered separately, but in general, the scheduling method is determined by the queuing method. Here we will consider splitting the case for using a single FIFO queue on a particular output port or using multiple FIFO queues. Even if a single or multiple FIFO queues are assumed, the packets actually arrived are stored in shared memory regardless of the arrival order, and only the memory information where the packets are stored is stored as a FIFO-type virtual queue. Are managed. The queuing and scheduling schemes discussed here will be limited to FIFO queuing, priority queuing, fair queuing, and weighted fair queuing.

FIFO queuing refers to using a single FIFO queue. Unlike general mapping of more than one class to a single queue, FIFO queuing stores all classes of traffic in a single queue. The scheduling method used for FIFO queuing is a method of first-serving a first packet, regardless of the class or priority of the packet. Corresponds to the queuing and scheduling structure used in traditional Internet networks with only the best effort service model. Because there is no segmentation of traffic, equipment that uses FIFO queuing does not require class-specific features such as packet classification or marking.

The advantage of FIFO queuing is that it is simple to implement and the behavior of the FIFO queue is predictable. That is, the order of the packets is maintained, and the maximum delay of the packets is determined by the maximum size of the queue. The disadvantage of FIFO queuing is that it is impossible to provide a differentiated service because there is no class distinction. It is also unsuitable for burst traffic services, and is advantageous for UDP traffic over TCP when congestion occurs. In other words, congestion when TCP and UDP traffic is mixed

If this occurs, the TCP sender will reduce the transmission rate by the flow control algorithm, but the UDP sender will continue to send traffic, taking up the bandwidth generated by TCP's flow control, and eventually the TCP traffic will be difficult to service.

20 is a diagram illustrating operation of priority queuing implemented in a gigabit Ethernet switch according to an embodiment of the present invention.

Priority queuing, also known as strict priority queuing, uses multiple FIFO queues. Since multiple FIFO queues are used, each queue is mapped to a different traffic class. Scheduling is very simple when using priority queuing. That is, packets stored in the low priority queue become service after all the packets stored in the high priority queue are serviced. If a packet is entered in the high priority queue while the packet stored in the low priority queue is being serviced, the low priority queue stops the service for a while and newly serves the newly arrived packet in the high priority queue. The advantage of the priority queuing method is that it can provide differentiated services in a simple way. Thus, it is possible to support real-time applications. However, when a packet is continuously input to a high priority queue, there is a problem that starvation (famine) phenomenon that a packet stored in a low priority queue is not served. This problem can be solved using WFQ, which will be explained later. In addition, when several traffics are stored in one and the same queue, a problem arises in that bursting traffic overwhelms other flows.

21 is a diagram illustrating operation of process queuing implemented in a gigabit Ethernet switch according to an embodiment of the present invention.

Fair queuing uses multiple FIFO queues, such as priority queuing, so that each queue is mapped to one or more traffic classes. In priority queuing, low priority queues experience starvation because high priority queues have absolute service priority for low priority queues as long as they have packets, but in fair queuing all queues Because of the same priority, no stabilization occurs. However, since this scheme considers only fairness at the service level without considering the characteristics of traffic, it is impossible to provide a differentiated service at the scheduling level.

22A, 22B, 22C, and 22D illustrate operations of W fair queuing implemented in a gigabit Ethernet switch according to an embodiment of the present invention.

Weight Fair Queuing (WFQ) was developed to solve the phenomenon of stabilization in priority queuing and at the same time, to solve the phenomenon of failing to provide differentiated services in fair queuing. To this end, each queue is assigned a weight, and scheduling is performed to be proportional to the assigned weight. 22A, 22B, 22C, and 22D show examples of applying WFQ to three types of queues. The case of assigning a weight of 50% to the High Priority queue, 25% to the Medium and Low Priority queues, respectively. 22B and 22C, since packets exist in all queues, it can be seen that the service is performed at a ratio of 50:25:25 in proportion to the weight value. However, in FIG. 22D, since no packet exists in the high priority queue, the weight assigned to the high priority queue is allocated equally to the remaining queues or proportional to the weight value assigned to each queue. It shows the service at 50:50 ratio. Because WFQ aims to solve the stabilization that occurs in Priority queuing, it can set bandwidth that must be guaranteed rather than simply assigning weight values.

If it is necessary to guarantee 25% of the total bandwidth for the high priority queue and 15% of the total bandwidth for the medium priority queue, the following is the case. In this case, the above bandwidth is allocated to each queue first, and the remaining 60% of the bandwidth is allocated to each queue according to the weight value. If the weights are allocated at a ratio of 50:25:25, respectively, the High queue is 30, which is 50% of the remaining bandwidth, and the Medium and Low Priority queues are 25% of the remaining bandwidth. You will be assigned 15 for. As a result, the High Priority queue will be served by 25% guaranteed and 30% allocated according to the weight, that is, 55% of the total bandwidth. Medium queues are guaranteed 15% and allocated 15% by weight, so they receive 30% of the total bandwidth. Low queues are serviced only by 15% allocated by weight. That is, the weight values of the WFQ are applied to maximize the network resources.

A gigabit Ethernet switch as a packet switching platform implemented by an embodiment of the present invention is a platform capable of providing TPS in an optical network and providing a service in a 100 Mbps cable modem network with a management function of CMTS. It can be utilized. To this end, we will classify voice and data traffic like the QoS features described above, and finally provide a TPS network solution using marking, queuing, and scheduling techniques.

In order to provide a function of managing a cable modem, a DHCP vendor option and a boot file may be stored as a configuration download function for the cable modem. You can also use your own DHCP server to automatically assign the modem's IP address and manage the ARP table.

The Gigabit Ethernet switch implemented by an embodiment of the present invention is a packet switching equipment capable of triple play service (TPS) in an optical network, and snooping voice protocols (SIP, H.323). This allows you to create a dynamic QoS Rule table. .

In addition, it provides management functions such as DHCP relay of the current CMTS to operate as a service generation platform.

As shown in FIG. 2, the main components include a classifier, a marker, a polymer, a buffer manager, and a queue scheduler according to the QoS model. We looked in detail. In particular, TPS can verify differentiated service models by traffic.

The gigabit Ethernet switch implemented by one embodiment of the present invention provides various functions as a network function. Switching capacity is provided to simultaneously handle line speed traffic at each port through L2 bridging. Transparent LAN provides more than 8K MAC address table.

Static routing can be provided using more than 2K ARP tables and more than 8K routing tables. It can also support one or more default gateways.

Meanwhile, over IEEE 802.1q VLAN, more than 24 port based VLANs and more than 4K tagged VLANs (TID = 1 to 4094) can be implemented simultaneously. IEEE 802.1w RSTP provides single port loop detection. And multicasting control for IP-TV via IGMP snoop. A fast-leave function can be implemented to reduce the zapping time.

DHCP server provides bootp function and DHCP relay can selectively relay IP pool by MAC address. It also controls the inflow of broadcast packets (Broadcast Storm control), and provides multicast filtering to prevent illegal VOD servers of subscribers. The MAC count limit is set to limit the number of simultaneous Internet connections. COS marking level can be controlled by COS marking and VLAN, and COS-DSCP remarking enables mutual remarking between COS and DSCP (or TOS).

Meanwhile, anti-virus measures can be taken to block the influx of worm viruses and differentiated scheduling of voice, video, and data can be provided to provide TPS (Triple Play Service). have.

It also supports CLI (Command Line Interpreter) function to provide management function and can provide remote access by Telnet. It also provides a console port to support the RS-232C console for outbound management. In addition, the device's settings can be saved and restored when the power is turned off and on. In addition, it provides an 8-level log level system log to record the events that occurred in the system. In addition, it can provide login password function and remote software upgrade function for security when CLI access.

The specific contents for the practice of the present invention have been presented. This includes the various examples and implementations related to the Gigabit Ethernet switch for providing the TPS of the HFC network, but the scope of the present invention is not limited to this because of this. The scope of the invention is determined only by the claims.

The present invention can provide an IPTV service and a VoIP service in a hybrid fiber cable network (HFC). Looking more closely, it provides Gigabit Ethernet Switch, which enables triple play service (TPS) of voice, video, and data, and CMTS (Cable) which is currently applied based on DOCSIS 1.0 / 2.0 in optical networks. Modem Termination System's packet switching technology can provide a stabilized packet network solution with the QoS function required for TPS.

1 is a view showing a service overview of the HFC network in which the Ethernet switch of the present invention operates.

2 is a diagram illustrating a configuration of a gigabit Ethernet switch according to an embodiment of the present invention.

3 is a diagram illustrating a configuration of a QoS model of a gigabit Ethernet switch implemented by an embodiment of the present invention.

FIG. 4 is a diagram illustrating the structure of QoS in an edge node implemented in a gigabit Ethernet switch implemented by an embodiment of the present invention.

5 is a diagram illustrating the structure of QoS in a core node implemented in a gigabit Ethernet switch implemented by an embodiment of the present invention.

FIG. 6 is a diagram illustrating an L2 frame structure transmitted and received by a gigabit Ethernet switch implemented by an embodiment of the present invention.

FIG. 7 is a diagram illustrating an L3 frame structure transmitted and received by a Gigabit Ethernet switch implemented by an embodiment of the present invention.

FIG. 8 is a diagram showing the structure of a ToS field of an IP packet header transmitted and received in a Gigabit Ethernet switch and a DS field of DiffServ according to an embodiment of the present invention.

9 is a diagram showing the configuration of a traffic conditioner implemented in a gigabit Ethernet switch implemented by an embodiment of the present invention.

FIG. 10 is a diagram illustrating an operation method of sr-TCM implemented in a gigabit Ethernet switch implemented by an embodiment of the present invention.

FIG. 11 illustrates a rate limit operation implemented in a gigabit Ethernet switch implemented by an embodiment of the present invention.

12 is a diagram illustrating a shaping operation implemented in a gigabit Ethernet switch according to an exemplary embodiment of the present invention.

Figure 13 shows the operation of TCP implemented in a gigabit Ethernet switch according to an embodiment of the present invention.

14 is a diagram illustrating a scenario in which multiple TCP flows implemented in a gigabit Ethernet switch using one buffer according to an embodiment of the present invention.

FIG. 15 is a view illustrating a process of controlling congestion by applying RED implemented in a gigabit Ethernet switch according to an embodiment of the present invention.

FIG. 16 shows an example of the RED packet drop probability function used in FIG. 15.

17 shows an example of a WRED packet drop probability function implemented in a gigabit Ethernet switch according to an embodiment of the present invention.

18 shows an example of applying WRED to one class using a single queue implemented in a gigabit Ethernet switch according to an embodiment of the present invention.

19 is a diagram illustrating the operation of queuing implemented in a gigabit Ethernet switch according to an embodiment of the present invention.

20 is a diagram illustrating operation of priority queuing implemented in a gigabit Ethernet switch according to an embodiment of the present invention.

21 is a diagram illustrating operation of process queuing implemented in a gigabit Ethernet switch according to an embodiment of the present invention.

22A, 22B, 22C, and 22D illustrate operations of W fair queuing implemented in a gigabit Ethernet switch according to an embodiment of the present invention.

Claims (8)

A network transceiver for transmitting and receiving a packet by accessing Gigabit Ethernet; A QoS logic unit for distinguishing a voice packet, an image packet, and a data packet from the packet received through the network transceiver to provide a QoS suitable for the packet; And A memory unit for temporarily storing data or packets processed by the QoS logic unit; The QoS logic unit A packet classifier configured to classify the packet by using information of two layers (L2) to four layers (L4) among OSI 7 layers; A packet marker unit marking a header of the classified packet with a data tag previously promised; A polymerizer for controlling the speed of the traffic so as to measure an amount of the packet received so that the packet is not sent out over a preset bandwidth; And A queue scheduler that temporarily stores the marked packets when the congestion occurs, according to a predetermined rule, The QoS logic unit Provides a rate limiting for discarding all incoming traffic over a preset rate or a shaping function for temporarily storing and transmitting the incoming traffic over a preset rate, and a flow control for controlling the transmission rate of the first terminal transmitting the packet ( Optical Gigabit Ethernet Switch provides IGMP function for HFC network IPTV service that provides flow control function and provides congestion control by dropping packets in a random way when more packets are introduced . The method of claim 1 And the QoS logic unit classifies the received packet into a voice packet and a data packet to manage traffic of each packet separately, and provides an IGMP function for an HFC network IPTV service. The method of claim 1, The packet classifier uses optical identification information of a device to which a packet is input or MAC or IP address information of a device that transmits and receives a packet as a criterion for classifying a packet. Ethernet switch. The method of claim 1, The packet classification unit uses a field value of a packet header defined in a protocol for transmitting and receiving a packet as a criterion for classifying a packet. An optical gigabit Ethernet switch providing an IGMP function for an IPTV service of an HFC network. delete delete delete The method of claim 1, The QoS logic unit queues the voice packet differently according to each rank by setting the voice packet as the highest priority packet, the video packet as the middle priority packet, and the data packet as the lower priority packet. Optical Gigabit Ethernet Switch that provides IGMP functionality for HFC network IPTV services.

Images