WO2008119241A1

WO2008119241A1 - A method for controlling the message channels of the main-secondary multi-processor system

Info

Publication number: WO2008119241A1
Application number: PCT/CN2008/000530
Authority: WO
Inventors: Qizhi Liu; Fang Xu
Original assignee: Maipu Communication Technology Co., Ltd.
Priority date: 2007-03-30
Filing date: 2008-03-18
Publication date: 2008-10-09
Also published as: CN101277253B; CN101277253A

Abstract

A controlling of message channels in router is related. Aiming at the disadvantages that the resources of the processor are exhausted and the low priority queues are starved in present main-secondary multi-processor system, a method for controlling the message channels between the main processors and secondary processors of the main-secondary multi-processor system is provided. The scheme of this invention is a method for controlling the message channels of the main-secondary multi-processor system, the characters are: switching the system states based on the channel queue spare ratio q; adjusting the system strategy under the different system states, and initiatively discarding the message before the channel congested. The beneficial effect of the method by this invention is that the congestion is avoided in mechanism. The low speed interface can not link-break or interface-passless caused by discarding the keepalive packet when the high speed interface heavy-pours packets. The messages which need to assure the priority can be ensured when a mass of messages are coming. The phenomenon that other general messages are submerged is greatly mitigated when the attack messages come suddenly.

Description

Instruction manual

Master-slave multiprocessor system message channel control method

The present invention relates to a router technology employing a master-slave multiprocessor in a communication system, and more particularly to control of a message channel in a router.

Background technique

In a router with a master-slave multiprocessor, the slave processor is generally used for message forwarding, and the main processor performs control and special message processing. In the packet channel from the processor to the main processor, a large number of packets need to be transmitted, such as link control packets, local routing protocol packets, and IP forwarding packets lacking fast forwarding records. Special messages that are strictly checked by the firewall function, and so on. If the message channel is not handled well, congestion may occur and the system may not operate properly. For example, due to the high-speed interface, a large number of packets on the low-speed interface, such as keepalives, are lost. Due to malicious attacks, a large number of packets flood the main processor, causing the main processor to fail to work properly and routing. When a change occurs, the forwarding entry needs to be refreshed, causing a large number of packets to be blocked on the channel and other normally sent packets to be lost. Since the data channel itself has the characteristics of a queue, generally, in the management of the channel, a manner of discarding redundant messages through the queue is adopted.

One method is to adopt a priority queue method, that is, classify packets according to type, and then enter queues of different priority levels, and schedule them according to queue priority. This method can ensure that important packets are enqueued, but may appear. The low priority queue is starved to death.

Another method is to poll the scheduling according to the receiving interface into different queues. This method ensures that the packets of the low-speed interface are not squeezed by the high-speed interface. However, if there are many low-speed interfaces, the polling itself is more resource-consuming. The performance of the high-speed interface is degraded. At the same time, the packets that need to be guaranteed by the same interface may be obscured by other large packets.

The above methods all rely on the queuing mechanism to achieve natural discarding, so there is a problem that the processor resources may be exhausted, that is, when the amount of packets far exceeds the processing power of the processor.

Recognize When scheduling tasks, it may consume 100% of the processor resources, while other low-priority tasks are delayed from scheduling problems.

Summary of the invention

The technical problem to be solved by the present invention is to provide a message channel control between a master-slave multiprocessor master-slave processor, which is disadvantageous in that the prior art is prone to exhaustion of processor resources and low-priority queues are starved to death. method.

The present invention solves the technical problem, and adopts a technical solution, which is a message channel control method of a master-slave multiprocessor system, which is characterized in that: according to a channel queue idle rate q, system state switching is performed; in different system states, Adjust the system policy to discard packets before the channel is congested.

Specifically, the following steps are included:

a. Set the threshold of the idle rate q in the system state machine: ql>q2>q3>q4; b. In the initial stage, the message forwarding is not restricted, the system is defined as a free state; c when the system is in the free state, check The current idle rate of the queue. When the idle rate is q<q2, the system is defined as a critical state;

d. After the system enters the critical state, continue to check the queue idle rate:

When the idle rate gradually becomes higher and q^ql, the system returns to the free state; or

When the idle rate continues to decrease and q q3, the system is defined as the limit state, and the first adjustment period is calculated, and the following steps are performed;

e. After the system enters the restricted state, the incoming packets are reviewed and the queue idle rate changes are checked during the determined adjustment period:

If the queue idle rate is gradually getting higher and q q2, then the system returns to the free state, or; the idle rate continues to drop rapidly, then enter the following steps:

f. According to the type of the current active interface, the system determines a default receiving ratio, and adjusts the default ratio according to the network condition captured by the critical state as the initial receiving ratio; at the end of each subsequent period, adjusts the proportional value according to the network traffic change. ;

g. Each interface adjusts its state according to the assigned receiving ratio and the condition of receiving the message;

h. Each interface processes the packet according to its different status: If the idle state is lower than q4, the interface is in the discard state and the packets are discarded.

Specifically, the idle rate q is defined as: the ratio of the idle length of the current queue to the total length of the queue.

The beneficial effects of the present invention are to avoid the occurrence of congestion from the mechanism. When the high-speed interface is heavily filled, the low-speed interface will not lose the keep-alive packet, causing the link to be disconnected; ensuring that the queue does not continue to be full, effectively ensuring that the processor resources of the main processor are scheduled. Exclusively; When a large number of packets arrive, the packets that need to be guaranteed are guaranteed. When the attack packets are too strong, the phenomenon that other normal messages are annihilated is greatly alleviated.

detailed description

The technical solution of the present invention will be described in detail below with reference to the embodiments.

The present invention judges the current state of the system according to the current idle rate of the queue, intelligently adjusts to different policies in different states, and discards some packets that need to be discarded in advance, thereby avoiding congestion from the mechanism, instead of happening after the occurrence. Naturally discarded.

First, according to the idle rate of the current queue, the system is divided into three states: free state, critical state, and restricted state. Set the threshold for the idle rate q in the system state machine: ql

>q2>q3>q4

The three system states can be switched according to the queue idle rate q. The idle rate of the queue refers to the ratio of the idle length of the current queue to the total length of the queue. The system is initially in a free state, and packet forwarding is not restricted. The state is switched according to the change of the queue idle rate. The main steps are as follows:

1. When the system is in the free state, check the current idle rate of the queue. When the idle rate q < q2, the system switches from the free state to the critical state;

2. After the system enters the critical state, check the queue idle rate: When the idle rate is gradually higher and 9 91, the system switches back to the free state; when the idle rate continues to decrease and q q3, the first adjustment period is calculated, the system switches To the restricted state, proceed to step 3;

3. After the system enters the restricted state, it reviews the incoming packets and determines the changes in the queue idle rate during the determined adjustment period: If the queue idle rate gradually becomes higher and q q2, the system switches back to the free state, and proceeds to step 1;

Or, the idle rate continues to drop rapidly, and the system determines how to process the message through a certain control strategy.

Further, in the restricted state, each interface also adjusts the proportion of the next cycle according to the number of received packets in each cycle to determine a data. Therefore, the interface also has three states: a free state, a restricted state, and a discarded state. The free state is generally for an interface with a particularly small current traffic. The discard state always occurs when the system is particularly urgent, and the restricted state means that the packet received by the interface needs to be further classified, such as a message. Type, connection to which the message belongs, or special indication from the firewall.

Example

In an asymmetric dual-processor routing system, a slave processor implements fast packet forwarding, and a master processor implements system control and special message processing. In such a system, congestion may occur in the message channel from the processor to the main processor, and the present invention can solve this problem relatively well.

The present invention does not rely on other auxiliary tasks to complete, all actions are triggered by a processor message enqueue request, so all actions are performed on the slave processor.

In the system state machine, set the threshold of the idle rate q ql>q2>q3 >q4:

The threshold ql indicates that the idle rate is high;

The threshold value q2 indicates an idle rate start alarm;

The threshold q3 indicates that the idle rate is low;

Idle rate <q3, the queue is no idle;

The idle rate is as low as q4, and some interfaces start to drop packets.

System state machine works:

Free state:

In the beginning, the system is in a free state, the message is not subject to any restrictions, and each enqueue request can be satisfied. At the same time, the current idle rate q of the queue is checked, and one of the two actions A and B is triggered according to the size of the idle rate q:

A: The idle rate is lower than the threshold q2, and the alarm is started; Action: Initialize the default proportion of each interface to receive packets, as well as other required statistics, and the state transitions to a critical state.

B: The idle rate q is still relatively high and continues to be in a free state.

Critical state:

The critical state triggers one of three actions C, D, and E according to the size of the idle rate q: C: After a period of time, the idle rate rises and rises above the threshold ql, indicating that the accumulation of the queue is only an instantaneous burst. The movement has been slowed or ended, and returned to a free state.

D: The idle rate is between the threshold ql and the threshold q3, and there is no significant drop or increase action: Keep it in a critical state and continue to observe.

E: The idle rate continues to drop below the threshold q3, requiring a limit;

Action: The system determines an initial default receiving ratio according to the interface type and bandwidth, and takes the time difference from the critical state to the current time as the initial statistical period, and corrects the initial default receiving ratio according to the received and received data collected in this cycle. Obtain the constraint condition for the next cycle to refer to, and obtain the state of the next cycle of each interface according to the current data, then the system transitions to the restricted state, and starts the next adjustment cycle.

Restricted status:

In this state, the application for enrollment is reviewed. According to the current system status, whether or not to enter the team is determined. Each application will trigger one of four actions: F, 0, 11, and I:

F: The adjustment period has not ended, and the idle rate has no obvious downward trend;

Action: Check the current status of the receiving interface. If it is in the free state, it will directly enter the queue. If it is in the discard state, it will be directly discarded. In the restricted state, the number of enqueues allowed, the type of the packet, and the enqueue of the corresponding data stream will be allowed according to the current interface. Comprehensive judgment, decide whether you can join the team.

G: The adjustment period ends, or the idle rate drops too fast during the period;

Action: End this cycle; If the idle rate drops too fast, it indicates that the cycle time set by the critical state is too long, the granularity is too large, which is not conducive to control. It is necessary to end this cycle ahead of time and re-correct the cycle time. According to the statistical value of this cycle, the restriction condition is re-adjusted. If the idle rate continues to decrease, the restriction condition is increased, otherwise the appropriate adjustment period is relaxed, and the next adjustment period is started. H: The adjustment period ends, the idle rate rises above the threshold q2, and the number of discarded N consecutive packets is 0;

Action: The system transitions to a free state.

I: (emergency control event) Whether the current period has ended or not, if the idle rate is lower than the threshold q4, the event is triggered directly;

Action: Do not force the end of this cycle, directly set the interface state and flow state to the most severe control. You can directly set the interface state to the discard state and start actively discarding the message. The interface in the discard state needs to wait until the resource is slowed down and the state is automatically corrected to the restricted state when the period ends.

The main advantage of the present invention lies in its intelligence, which is reflected in several aspects: 1. Judging the busyness of the main processor according to the current queue state, and triggering the intelligent running state machine by the message; 2. Introduction of the critical state, intelligence Calculate the time of the adjustment period, and dynamically adjust it in each subsequent period, so that it can adapt to the message congestion caused by various reasons; 3. The enqueue ratio of the interface is weighted and intelligently adjusted according to the latest statistics; Intelligently adjust the strength of message restrictions based on current status.

Through the integrated application of these intelligent strategies, channel congestion is avoided at the source, instead of thinking about ways to discard later.

Claims

Claim

The master-slave multiprocessor system message channel control method is characterized in that: according to the channel queue idle rate q, the system state is switched; in different system states, the system policy is adjusted, and the channel is actively discarded before the channel is congested. Message.

2. The message channel control method of a master-slave multiprocessor system according to claim 1, comprising the steps of:

a. Set the threshold of the idle rate q in the system state machine: ql>q2>q3>q4; b. In the initial stage, the message forwarding is not restricted, the system is defined as free state; c when the system is in the free state, check the queue Current idle rate, when the idle rate q < q2, the system is defined as a critical state;

If the queue idle rate is gradually higher and q q2, the system returns to the free state, or; the idle rate continues to drop rapidly, then enter the following steps - f. According to the current active interface type, the system determines a default receive ratio, according to the critical state After the captured network status is adjusted to the default ratio, it is used as the initial receiving ratio; at the end of each subsequent period, the ratio is adjusted according to the network traffic change;

g. Each interface adjusts its state according to the proportion of the received receiver and the status of the received message;

Each interface processes the packets according to their different states: The packets are directly forwarded in the free state, and are controlled according to the priority in the restricted state. When the idle rate is lower than q4, the interface is discarded and the packets are discarded. .

The message channel control method of the master-slave multiprocessor system according to claim 1 or 2, wherein the idle rate q is defined as: a ratio of the idle length of the current queue to the total length of the queue.