EP1547324A2 - Method for partially maintaining packet sequences in connectionless packet switching with alternative routing - Google Patents

Abstract

In order improve maintenance of packet sequences in a connectionless network comprising several routers enabling the packets of a flow to be transmitted via several pathways, a packet arriving in a router is transmitted on a route in the same way as a packet which is already located in the router and belongs to the same flow or an incoming packet in the router, which arrives in the router within a time window and which is opened by a packet belonging to the same flow, is transmitted on the same route as the packet opening said window.

Description

description

Procedure for the partial maintenance of the packet sequence in connectionless packet switching with alternative routing

The subject of the application relates to a method for maintaining the packet sequence in connectionless packet switching with alternative routing for a network with multiple routers.

Networks with connectionless packet switching (e.g. today's Internet) generally have no way of maintaining the order of packets during transport through the network, i.e. to offer the same order at the exit of the network as at the entrance if, within the network, for example, the route to a destination can be individually selected for each packet for load distribution.

Errors in the packet order can, for example, unnecessarily reduce the throughput of connections if this is regulated by the protocol TCP (Transmission Control Protocol).

In order not to overload the network and to achieve a fair distribution of the total bit rate over many connections, a TCP sender reduces its transmission rate (by reducing the transmission window) after detecting a packet loss. In practice, a packet order reversed in the network also leads to repeated confirmations with the same order number, so that TCP also reduces the rate here.

Alternative routing at the packet level, i. H. Traffic distribution of packets in a flow is calculated from the above Reasons usually not used today. To the above Avoiding problems is, for example, with MPLS (Multi Protocol Label Switching) alternative routing on the level (aggregated)

Flows proposed, ie all packets belonging to the same connection or between the same pair of network nodes exchanged are sent through the network in the same way. For this, however, corresponding connection information must be stored in each network node, e.g. B. by configuring paths (static), or by first establishing a path for each connection (dynamic, but connected with a lot of effort and therefore not necessarily scalable for large networks). The number of flows to be saved depends very much on the duration of the flows and can become very large for long flows with little traffic.

In addition, devices can be used at the network output to restore the packet sequence, which, however, is a non-trivial task in IP (Internet Protocol) networks, since IP packets generally do not have a sequence number that can be used for such purposes. The "Identity ^> " field in the packet header uniquely identifies a packet, but is not necessarily increased by 1 within each TCP connection or each UDP (User Datagram Protocol) association. To evaluate the specified octet-level TCP "Sequence Num over ^Λ, must be the packet header further evaluated, as this number increases by a TCP segment to the next by the number of bytes in the segment. Since the segments can also carry different amounts of useful information within a connection, a resequencing device cannot know how many packets are still missing between two other received packets if their sequence numbers do not connect to one another. In addition, if a resequencing device were used, packet losses would cause a delay in the playout of the packets and would thus override the “Fast Retransmit * mechanism of TCP, which would cause the bandwidth regulation in TCP to be downgraded significantly and therefore not an advantage compared to delivery out of sequence would bring.

The invention has for its object to provide a method that in a network with multiple path options Reduced performance degradation due to package overhauls.

The object is achieved by the features of claim 1 or claim 4.

The frequency of packet overhauls, in particular for high-bit-rate connections, is greatly reduced by the invention. The frequency of exchanges in the packet order is determined by the above. technical features decreased. The maximum number of flow entries in the flow table FT is predetermined by the number of packets to be stored in the router. The restriction to packets stored in the router thus greatly reduces the amount of status information in the router compared to solutions such as MPLS or IP switching, which must maintain a status for each existing flow. In addition, in contrast to MPLS or IP switching, no signaling is required between the network nodes, so that there is no unnecessary delay, especially with short flows. The limitation to the short lifespan of the status information also has the advantage that the flexibility of alternative routing for load distribution in the network is guaranteed, so that a compromise can be reached between absolute compliance with the packet sequence and optimal load distribution. Connections that send at a high rate and of which at least one packet is always cached in the router will not be swapped. Connections in which a packet is seldom sent will also have no problems if the runtime differences between the different paths chosen in the network are small compared to the time between two packets. The described solution is therefore particularly advantageous for connections in which data is sent in batches (e.g. World Wide Web).

Advantageous developments of the invention are specified in the subclaims. The invention is explained in more detail below as an exemplary embodiment to the extent necessary for understanding with reference to figures. Show:

1 shows a simplified representation of an IP network,

2 shows a schematic representation of an IP router,

Fig. 3 is a schematic representation of an inventive

4 and 4 a schematic representation of the content of the flow

Table FT.

In the figures, the same designations denote the same elements.

A simple network is shown in FIG. 1, in which two terminal devices E1 and E2 are connected to one another via several paths, the routers Rl to R5 being intended to carry out a connectionless packet switching between the links (lines) L0, Ll, L2, ... L7 , FIG. 2 shows part of the IP router R1 as it is constructed according to the prior art for a transmission direction (from L0 to L1 and L2). When a packet arrives, it is classified, the destination IP address is read and the next router on the way to the destination is determined from the routing table RT for this address. The routing table receives current routing information from the routing protocol processor RP, which exchanges accessibility information with other routers via a routing protocol. In general, the "shortest path is ^Λ, that is the (after a predetermined metric) shortest route to the destination as the only way into the routing table RT entered.

In the case of load balancing on several alternative routes, the routing table is expanded and contains, in addition to the next node on the shortest path, further next nodes for further permissible paths to the destination. With every arriving end packet, a permissible output path to the destination can now be selected on the basis of a load distribution ^{algorithm Λ} , to which the packet is then forwarded.

According to the invention, it is proposed to keep a table of flow or connection information (hereinafter referred to only as flow information FI) in the router, which is temporarily stored in the router (i.e. in one of the queues Q1, Q2 or Q3 or just in the switching network) mediated) package saves the selected route. Leaves

(in this embodiment) the packet the router, then the information is deleted again. If a new packet with the same FI arrives in the router, it is forwarded on the same path as the last packet with the same FI.

The decision as to which of the alternative ways a packet is forwarded is therefore only made again if no packet with the same FI as a newly arrived packet is already in the router. This greatly reduces the frequency of packet overhauls for high bit rate connections.

The corresponding router in FIG. 3 contains, in addition to the components of the router from FIG. 2, a flow table FT in which the selected next hop is stored for all packets that are still in the router and have already been classified. For each newly arriving package, the affiliation to one of the flows in the FT is checked. If a packet of the corresponding flow is recorded in the FT, the same selection of the next hop is also made for the new packet. If not

Packet of the same flow is recorded in the FT, a next hop is selected for this packet based on the rules of the alternative routing and the load distribution, the packet is forwarded in the direction of this next node and the flow information is stored in the FT together with the selected next hop , Figure 4 shows an example of what such an FT can look like. The FT contains i for each flow, from which itself Packets are in an outgoing queue of the router, the number ni of packets in the queues, the flow identification information (source IP, destination IP, source port, dest. Port, protocol) and the next hop selected for this flow. The packet counter for each flow is incremented by 1 with each incoming packet for this flow and decreased by 1 with each packet outgoing from the router for this flow. If the counter reaches 0, the entry is deleted from the table.

Further configurations:

1. The principle can be applied to each queue individually, to a subset or to all buffers in a device if an IP router uses, for example, input and output buffers or a combination of such buffers with a central buffer. The following alternatives are possible: a) separate FT and separate packet count. In this case, the FT only refers to the queue, at the exit of which the decision to forward the route is made. Any output buffers arranged behind it and the packets in them no longer have any influence on the route decision for new packets. b) shared FT and separate packet count. In this case, the FT contains one packet counter per queue, which is updated each time a packet enters and exits the queue. The forwarding decision is saved via flow. c) shared FT and shared packet count. The FT is structured according to FIG. 4, where ni relates to the sum of all packets of flow i in all the queues under consideration.

The forwarding decision is also influenced by the packets of a flow that have already passed the decision point. This option b) / c) is preferable to a). 2. IP routers usually have one output queue per output link, whereby the output link can be a physical network connection or a logical channel within a physical connection (e.g. an ATM-VP (Asynchronous Transfer Mode-Virtual Path) or -VC (Virtual Channel ), a wavelength or an STM (Synchronous Transport Module) channel). Backbone routers usually have exactly one next IP router assigned to each of these channels. In local networks, however, an output channel can also reach several next IP routers if the channel is a shared edium (Ethernet or similar), for example. In this case, there are options to enter either the output ^channel or - as indicated in FIG. 4 - the "next hop ^λλ for a flow in the flow table FT. The latter option seems more sensible; for other reasons (eg internal structure of the router) it may be necessary to enter the output channel as a replacement for the "next hop".

3. The description of FIG. 3 provides for flows to be transferred from the flow table as soon as there is no longer a corresponding packet in the router. Alternatively, it can also be provided that the time table for the last packet arrival is stored in the flow table FT in FIG. 4 instead of the number ni of packets in a flow. The entries are then periodically or after a time limit after arrival from the

Table is carried out if the time at which the last packet in a flow was observed was at least a certain time ago.

4. If a router handles several traffic classes, the method can be used for all or only for some of the traffic classes.

5. A time limit according to option 3 can be set adaptively depending on other parameters. In particular, parameters that determine the distribution of traffic (e.g. the frequency for choosing an alternative route) come into question.

Claims

claims 1. Method for maintaining the packet sequence in connectionless packet switching with alternative routing for a network with several routers, accordingly a route is selected for a packet arriving in a router, - the selected route is saved in the router for the associated flow of the package, - the package is forwarded to the selected route, - a newly arriving packet that belongs to the same flow, is forwarded to the saved route and - In the event that there is no associated packet in the router for a flow, the selected route is deleted in the router. 2. The method according to claim 1, characterized in that the criterion of whether there is a packet for a flow in the router, the number of packets associated with the flow in a queue of the router is evaluated. 3. The method according to claim 2, characterized in that the criterion of whether there is a packet for a flow in the router, the number of packets belonging to the flow is evaluated in exactly one queue of the router. 4. Method for maintaining the packet sequence in connectionless packet switching with alternative routing for a network with several routers accordingly for a packet arriving in a router, the time of arrival is stored in the router, for the packet arriving in the router a route is selected, the selected route is saved in the router for the associated flow of the packet, - the packet is forwarded to the selected route, the arrival time is updated for the associated flow of a newly arriving packet - A newly arriving package, for the flow of which a route has been saved, is forwarded to the saved route and, in the event that a predetermined period of time has elapsed since the arrival of the last package of a flow, the saved data for the flow are deleted. 5. The method according to claim 4, characterized in that the storages for the individual flows are deleted periodically. 6. The method according to claim 4 or 5, characterized in that the predetermined period of time is set in accordance with a criterion determining the traffic distribution. 7. The method according to any one of the preceding claims, characterized in that the storage for the individual flows takes place in a flow table (FT).