EP1754345A1 - Method for selective load compensation - Google Patents

Abstract

The invention relates to a method for redistribution of a traffic load on a switch, in which the following five steps are carried out: read the traffic information from the switch, supply the read traffic information to a processor, carry out an algorithm by the processor whereby the processor provides a precise examination of criteria of the traffic information, determination of whether the results from the precise examination are comparable with a pre-determined result, on a negative result for said comparison the information from the reference table is read and processed, such that, on implementation of the processed reference table, the data load is distributed such that the result of future investigations lie within the range of the pre-determined result and transmission of the processed reference table to the switch for actual implementation thereof.

Description

Procedure for selective load balancing

Background of the Invention

The present invention relates to a method for a selective load balancing according to the preamble of claim 1. And even more particularly it relates to a method in which the means required for implementation are arranged outside a switch, with an effective dynamic load balancing taking place ,

Switches, such as Ethernet LAN (local area network) switches, are known and are used as components of a packet-switched telecommunication network or a packet-switched telecommunication network. These switches cause the data packets to communicate between a large number of network nodes, which are based on a preset algorithm. Data packets arriving in the switch are temporarily stored in a buffer memory or register. The message header is read and an address for the data packet is defined. A reference table with addresses is consulted and a procedure is then derived in order to obtain the best possible connection between the switch and the destination network node of the data packet. A connection between two correspondingly long segments is shown in order to forward the data packets from the switch to the network node through the segments. Ordinary Ethernet-based LANs use an Ethernet frame, which comprises a normal data packet, the fee information of the frame and a special message header, which has the MAC address information about the origin and destination of the data packet. MAC addresses, or Media Access Controll, are physical addresses of a facility like this like the NIC (Network Interface Card) of a computer in the network. The MAC address comprises up to two identical parts, each 6 bytes long. The first three bytes identify the company that made the NIC. The second three bytes are the serial number of the NIC itself. NIC stands for Network Interface Card, which, as the name suggests, enables communication between the subscriber terminal and the network. NIC can also refer to the Ethernet card.

Packet-switched data switching comprises several known methods for forwarding data packets. Each of them depends on the in-circuit hardware and software. Such methods include: forwarding (cut-through), storage and forwarding (store and forward) and the transmission free of fragments (fragment-free). During the pass-through, the switch reads the MAC address as soon as a data packet has been recognized and immediately begins the pass-through or or forwarding after the six bytes have been stored, regardless of the remaining data packets that are still received or received. During storage and forwarding, the entire data packet is stored in a memory, checked for errors and then sent on, based on the MAC address subsequently read. If an error is discovered in this process, the data packet is discarded. Some switches are also known to combine forwarding and storage and forwarding procedures in varying degrees. The procedure "transmission free of fragments" is equivalent to the transmission, with the difference that the first 64 bytes of the data packet are saved before the forwarding takes place. A basis for this procedure is that most errors occur in the first 64 bytes occur, which means for the switch there is an opportunity to find out the errors early without having to accept the delay due to the storage of more than 64 bytes. As a result, the switch collects certain basic information about the data packets to be forwarded, regardless of the type of pre-ringing that is carried out by the switch.

In order to additionally learn something about the information of the data packets, the switches are configured so that they design their own reference table based on learned or received information from a network node arranged in the network. Such learning is automatically initiated without any need for mediation by a network administrator. The establishment of a reference table is commonly referred to as transparent bridging. Transparent bridging comprises five parts: learning, flooding, filtering, sending and aging. In short, learning is a process in which the switch, when it encounters a new MAC address, simply adds the address with the respective network node to its own reference table. Flooding is a process in which the switch searches for a network node whose address is previously unknown to it and in which it then supplies (floods) all the network segments known to it (except for the receiving segment) with the data packet. As a result, the appropriate desired recipient will then acknowledge receipt of the packet. With this acknowledgment, the flooding switch now knows the two network nodes and their location and can thus establish a connection between them and transmit the data packet. In the event that the switch detects two network nodes and determines that they have the same network in common, the switch will update its own reference table and consequently will not interfere in the communication between the two nodes sharing, which is the part of the Filtering was described. Aging is a form of cleaning the switch's memory from addresses that have not been used within a specified period of time. On the basis of what has been described above, the reference table of the switch can be designed and the information of the network node connection is available through the memory allocated in the switch.

Switches work in a conventional manner on the data link or the second layer of the OSI reference model. Some switches, which are more similar to routers and are known as Layer 3 switches, operate on the third layer of the OSI reference model, i.e. at the network level. The applications that control the network run on the highest layer, the application layer. A general basis for network control applications is the Signaling Network Management Protocol or the Simple Network Management Protocol (SNMP). As a result, most network controlled devices are compatible with the SNMP. For this reason, network control applications interrogate a control agent using a supported management information base (MIB). The question is carried out via SNMP. The MIB is a repository of properties and parameters that are controlled by a network device, such as a switch. Consequently, SNMP communicates with the switch via MIB, receives information from the switch and controls the switch in another way. This information may include an error counter, data load, and so on.

In addition, API (Application Programming Interface) is software that is used to enable communication with the controlled switches of the network. in principle API is a set of standard software of data formats that application programs use to establish contact with network services.

It is the hardware of the switch that recognizes the data packets and defines or executes hash functions and logically defines the assignment to a network node (via a hash value). The procedure mentioned above is based on internal (with reference to the switch) measures (hardware and software). Once implemented, switches remain relatively static and predictable in terms of how they work. In other words, the switches do not behave dynamically with regard to the different data loads and their requirements. As a result, certain ports on a switch can be more heavily loaded by data than other ports on the same switch. This poses a problem of the accumulation of data traffic on the switch, as a result of which the switch becomes slower and thus also the communication between the respective network nodes and ultimately (in different gradations) the overall network performance also decreases.

A known solution to this problem involves introducing some flexibility in the switch hardware. This higher flexibility represents a change to a number of parameters and / or hash functions in order to achieve a more balanced load distribution. The changes can be based on an analysis of the current traffic situations and their requirements. Such restructuring may be possible during ongoing operations. As always, the communication that has already been logically assigned to a special port of the switch cannot be redirected or redirected to another port. Therefore, a solution to switch over or to the U Data sharing required where the switch can operate more dynamically while data load can be redirected or redistributed regardless of the fact that the data loads have already been allocated. The solution must also be flexible to use, so that it can also be used for existing switches, which increases their performance without being forced to replace the switches.

The present invention is directed to solving this problem and specified in the characterizing part of patent claim 1. Refinements are specified in the dependent claims. The present invention is directed to a method in which a processor external to the switch is provided in order to examine the data load in detail and to determine whether the load advantageously resembles a desired distribution. If this is not the case, the switch's reference table is read, processed and then switched back to active, so that after the execution of this process, the processed table redirects the data traffic to other ports, thereby achieving a more advantageous distribution result.

The present invention will be described with reference to the drawings, in which:

Figure 1 is a flow diagram of the method, and Figure 2 is a box diagram of a system for performing the method shown.

FIG. 1 shows a flow diagram with the steps that are carried out by the illustrated invention in order to achieve a desired load balancing within a switch. The method begins at step 100 and continues to step 102, the information of the data load being read out from the switch. As is well known and discussed above, the switches provide multiple sources and several different significant sets of information indicative of the data load. The sources of the information include SNMP, API, network support, network manager, switch storage and so on. The information can be as simple as counting data packets that are routed via a special port, or it can also contain data from a network node or the data about details regarding the nature of the data packets sent. The data is read out for one or more ports of the switch. After receiving the read load distribution of the data traffic, a comparison is made between the read load distribution and the desired load distribution in step 104 and it is determined whether the read load distribution advantageously corresponds to the desired load distribution. The tolerance for this comparison is a matter of designing and applying the process. The illustrated invention is also not intended to be limited to any type of data or data source.

If the read data advantageously resemble the desired load distribution (step 106), the method returns to the beginning (step 108). However, if the data read out does not resemble the desired load distribution (step 110), the reference table is read and examined in detail (step 112). A determination is made as to how the reference table can be processed so that the data traffic currently handled by the switch can be redistributed among the ports of the switch in order to achieve the desired distribution of the traffic load. Because the present method for every type of desired load distribution for every type of switch, the processor-equipped ports and means for data storage is suitable, the external processor and its programmability must include a high degree of flexibility. The present processor and programming are therefore a matter of design. After processing, the reference table is sent back to the switch for implementation (step 116). The method then returns to the beginning after a preselected period of time (step 118) and the above steps are repeated.

Figure 2 describes a box diagram of a system for performing the inventive method. As shown, a switch 10 comprises four components: a switch table 20, a hash function 22, a processor interface 24 and a register 26. The work tasks of these individual components have already been described above. A brief description for repetition: The switch table 20 shows the port of the switch via which packets for a specific network node are forwarded. The hash function 22 is a processing of information according to special criteria. The processor interfaces 24 enable communication with the processor 16 via a connection 14. The register 26 is a temporary memory which is used for receiving, holding and transferring data. The processor 16 includes software 18 for processing algorithms. Numerous segments 12 enable communication with the switch 10, each of them with connection or subscriber interfaces. As applied with this present invention, the algorithms can be selected so that their execution results in a desired distribution of the data load. The design of the algorithms is known to the expert experienced in his field.

While the algorithm is used here for simply counting the packets, other criteria and implementations stanchions are provided. For example: comparing the static hashes of the switch; a last bit of the srcMAC, which is zero or one, and a hash on the port; the number of frames transported from the register per port; other functions of SNMP functions that provide detailed information about the traffic load situation, qualitatively separated from the pure count, such as sorting by business customer, by private customer, the amount of connection interfaces, etc.

While the present invention has been described in the above embodiment, it will be apparent to those skilled in the art that other embodiments can be found without departing from the essence of this invention, such as embodiments without link aggregation and hash function to determine the load distribution on logically separated connection interfaces.

LIST OF REFERENCE NUMBERS

10 switch 12 segment 14 switch-processor connection 16 processor 18 software for processing algorithms 100 start 102 read out data 104 query load balancing 106 query answer is "yes" 108 return to start 109 query answer is "no" 110 read out reference table 112 edit the reference table 114, replacing the original reference table with the edited reference table 116, returning to the beginning 118 end

Claims

claims 1. A method for redistributing a traffic load on a switch, characterized by the steps: ^■ reading the traffic ^information from the switch; ^■ routing the read traffic information to a processor; ^■ Execution of an algorithm by the processor, o- in which the algorithm initiates a precise examination of criteria of the traffic information; ^■ Determine whether the results from the exact examination are comparable to a predetermined result; ^{In the} event of a negative result of this comparison, the information in the reference table is read and processed, so that when the processed reference table is implemented, the data load is redistributed such that the result of future investigations is within the scope of the predetermined result; and ^■ forwarding the edited reference table to the switch for its actual implementation. 2. The method of claim 1, wherein one of the criteria of the traffic information is the number of data packets that are processed by at least one port of the switch. 3. The method of claim 1 or 2, wherein one of the criteria of the traffic information is the number of frames transported per port. 4. The method according to any one of claims 1 to 3, wherein one of the criteria of the traffic information are considerations of the quality of service. 5. The method according to any one of claims 1 to 4, further comprising the steps: ^■ Examining the data load carried by the switch with at least one SNMP function and API; and ■ communicating the results of the examination to the processor for precise analysis using the algorithm. 6. The method according to any one of claims 1 to 5, comprising the steps of analyzing the register of the switch and communicating the results of the analysis to the processor for further analysis by means of the algorithm. 7. The method of claim 1 to 6, further comprising the steps of periodically repeating the steps of reading the traffic information, determining whether the traffic load is balanced, determining a changed reference table for the switch and replacing the current reference table with the changed reference table for a predetermined period.