CN104734907A - Method for actively measuring end-to-end path performance of OpenFlow network and system adopted by the same - Google Patents

Abstract

A method for actively measuring the end-to-end path performance of an OpenFlow network and the system used in it. Aiming at the problem that the end-to-end path performance of an OpenFlow network cannot be measured at present, an OpenFlow active measurement protocol OFMP and a controller-centered Efficient measurement methods and systems. The measurement method and system can provide technical means and practical tools for debugging SDN control programs, quantitatively analyzing SDN architecture and mechanisms, and evaluating network behavior. In the case of device clock synchronization, the system can efficiently measure the routing, one-way delay, round-trip delay, segment-by-segment one-way delay, delay between the control plane and the data plane, and packet loss rate of a specific flow by sending a packet. End-to-end performance parameters. In addition, the measurement system can also measure some important performance indicators when the device clock is not synchronized or has a non-OFMP-enabled device, and has the characteristics of low overhead for processing the OFMP protocol.

Description

A method for actively measuring the end-to-end path performance of an OpenFlow network and the system used therein

technical field

The invention belongs to the field of network communication, and specifically proposes a method and system for actively measuring the end-to-end path performance of an OpenFlow network.

Background technique

At present, the network architecture of Software Defined Networking (SDN) provides a new technical approach to strengthen network functions, shorten network innovation cycles and solve Internet problems. Although OpenFlow-based SDN has been applied in data centers, network management, network security and other fields, and has verified the feasibility and usability of the SDN architecture, the SDN architecture is more complex and lacks the ability to debug SDN control programs and quantitatively analyze SDN. The architecture and mechanism and technical means of evaluating network behavior seriously hinder the scientific development of SDN.

Network measurement is a basic means to understand network behavior and an important method to quantitatively evaluate network performance. With the development of Internet technology, network measurement technology including two modes of active measurement and passive measurement has been developed. However, the current OpenFlow specification in SDN only provides an interface for real-time acquisition of flow information from the controller, which provides a centralized passive measurement method to a certain extent. Passive measurement has its limitations, it is difficult to measure various parameters such as connectivity between two points and connectivity performance, and the implementation technology is complex. The experience of IP network tells us that the end-to-end path performance of the network is one of the most important indicators in the network, and the main means of obtaining performance parameters is active measurement. Active measurement sends measurement packets (sequences) to the path source, and then observes and analyzes changes in the measurement packets (probe) during network transmission, thereby inferring network status and related performance parameters. This measurement method of actively injecting and tracking the real route of the packet can better reflect the real situation of the problem. However, SDN currently lacks practical active measurement mechanisms and methods.

Comparing the active measurement mechanism in the IP network, it is not difficult to find that there are great difficulties in developing an efficient and easy-to-use active measurement mechanism in the SDN. The first is feasibility. There is an end-to-end path between any two points in an IP network by default, which makes it possible to measure end-to-end performance. In SDN (the OpenFlow network is used as an example below), the path between two points It may not exist. Even if it exists, the packet must follow the flow forwarding rules issued by the controller to the switch. The second is ease of use. IP networks have protocols that support network measurements such as Internet Control Messages Protocol (Internet Control Messages Protocol, ICMP), which makes active measurements easy to perform, but there is no such protocol in SDN. The third is efficiency. The IP node protocol stack has built-in support for active measurement functions, which enables the measurement tasks to be completed efficiently, but SDN nodes cannot provide this support. To realize an efficient and easy-to-use SDN active measurement mechanism must face and solve these problems.

Contents of the invention

Aiming at the problem that the end-to-end path performance cannot be effectively measured in the OpenFlow network at present, the present invention proposes a method for actively measuring the end-to-end path performance of the OpenFlow network and the system adopted therein.

Technical scheme of the present invention is:

A method for actively measuring the end-to-end path performance of an OpenFlow network, based on the OpenFlow specification, it includes the following steps:

A. The user sends the measurement instruction of the network end-to-end path performance to the OFMP-enabled controller. After receiving the measurement instruction of the network end-to-end path performance sent by the user, the OFMP-enabled controller analyzes the measurement instruction to obtain the measurement index and measurement path , construct an OFMP measurement message and write the local measurement information of the OFMP-enabled controller in the measurement message, and send the measurement message to the first-hop OFMP-enabled switch according to the measurement path (the first-hop and last-hop switches are both Enable switch for OFMP), said OFMP measurement message includes the local measurement information of each OFMP enabled switch in the measurement path and the measurement path;

B. The first-hop OFMP-enabled switch receives the OFMP measurement message from the OFMP-enabled controller, writes local measurement information in the measurement message, and sends it to the next-hop switch according to the measurement path. If the switch is OFMP-enabled switch, write the local measurement information of the OFMP-enabled switch in the measurement message, and then send the measurement message to the next-hop switch; if the switch is not an OFMP-enabled switch, then send it to the next-hop switch The measurement message; traverse all switches in the measurement path in turn to complete the forwarding of the OFMP measurement message, and the last hop OFMP enables the switch to return the OFMP enable controller after the OFMP measurement message is written into the local measurement information;

C. The OFMP enabled controller processes the local measurement information sequence of each enabled switch in the OFMP measurement message according to the requirements of the measurement index, obtains end-to-end path performance parameters, and displays these performance parameters to the user.

In step A of the present invention: the OFMP-enabled controller refers to an OpenFlow controller capable of understanding the OpenFlow measurement protocol OFMP and performing corresponding functions; the OFMP switch refers to an OpenFlow switch capable of understanding the OpenFlow measurement protocol OFMP and performing corresponding functions.

In step A of the present invention: the measurement path refers to the path in which the measurement message is sent from the OFMP enabled controller in the OpenFlow network, and the switch sequence set by the measurement index requirements until returning to the OFMP enabled controller. The part of the sequence is consistent with the path of the flow being measured.

In step A of the present invention: the measurement index refers to the performance measurement of the end-to-end path, including routing, that is, the path node sequence that the measurement packet passes through, the round-trip delay, the one-way delay, the packet loss rate, and the hop-by-hop delay of the switch One or more of delays between the control plane and the data plane.

In step A of the present invention: OFMP is a protocol for communication between OFMP enabled switches and OFMP enabled controllers or OFMP enabled switches, and the message format of OFMP includes basic fields: the Version field is used to represent the OFMP version number, and the Flag field Including the identification bits of multiple measurement methods (such as single measurement or multiple measurements, one-way measurement or round-trip measurement), the Identification field and the Sequence Number field are used to identify different active measurement processes and different reports in the same active measurement process In the text, the Path Pointer field indicates the pointer to insert the current position of the path measurement record, the Measurement Start field and the Measurement End field are used to identify the starting point and the Dpid value of the end switch of the measurement path, and the Path Record field is used to store the node identification Dpid or the host The time when the MAC and measurement packets arrive at the device.

The local measurement information in the present invention includes an identifier of a local device and a local clock time, and the identifier of the local device is a Dpid or a MAC address.

In step B of the present invention, the first hop OFMP enabled switch receives the OFMP measurement message from the OFMP enabled controller in the PACKET OUT mode, and the last hop switch returns the measurement message to the OFMP enabled controller in the PACKET IN mode message.

A system adopted in a method for actively measuring OpenFlow network end-to-end path performance is characterized in that, on the basis of the OpenFlow specification, the OpenFlow active measurement system includes an OFMP enabled controller and at least two OFMP enabled switches, When the system measures an end-to-end path, the first-hop and last-hop switches of the path are both OFMP-enabled switches.

In the system, clocks of the OFMP-enabled controller and all OFMP-enabled detection devices can be synchronized by methods such as Network Time Protocol (NTP) or satellite timing. If the system clock is synchronized, all performance indicators defined in 4. can be measured; if the system clock is not synchronized, some performance indicators such as round-trip delay and packet loss rate can also be measured;

The function of traversing non-OFMP-enabled switches. For systems with non-OFMP-enabled switches (such as OpenFlow switches), this measurement system can still measure performance indicators such as one-way delay/packet loss rate and two-way delay/packet loss rate of the end-to-end path.

Beneficial effects of the present invention:

The present invention proposes for the first time a method and system for measuring the end-to-end path performance of the OpenFlow network with a centralized architecture, which solves the problem that the end-to-end path performance cannot be measured at present.

The present invention also has the following advantages: (1) under the condition that the device clock is synchronized and there is no need to change the forwarding strategy of the flow, the controller can efficiently obtain the route, one-way delay, round-trip delay, End-to-end performance parameters such as segment-by-segment one-way delay and packet loss rate; (2) When the device clock is not synchronized, it can measure some important end-to-end performance parameters of the specified flow; (3) It can span non-OFMP enabled (4) The overhead of processing OFMP is small; (5) The time delay between the control plane and the data plane can be measured.

Description of drawings

Figure 1 is a schematic diagram of the structure of the active measurement system.

FIG. 2 is a flow chart of OFMP message processing.

FIG. 3 is a schematic diagram of the experimental results of the average value of the one-way cumulative time delay when all the path nodes are enabled with OFMP.

Detailed ways

The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments.

Firstly, the environment required by the present invention is given, and Fig. 1 shows the experimental environment of the system. These include 5 OpenFlow-enabled switches, 1 PICA P3290 common switch, 1 POX-enabled controller, 2 Linux-side hosts, and 1 Spirent traffic generator. The five OpenFlow-enabled switches S1, S2, S3, S5 and S6 are all Linux PCs running openflow-1.0.0 software. These PCs use i5-3470CPU with a main frequency of 3.2GHz, 4GB of memory, and 4-port Gigabit Ethernet. Their Dpid values are set to 1, 2, 3, 5, and 6 respectively, and the Dpid of the switch S4 is set to 4. The switches S1, S2, S3, S5 and S6 and the controller all run the OFM daemon process and are OFMP-enabled nodes, and the switch S4 is not OFMP-enabled. The controller adopts the pox-carp version, and is interconnected with the switch in an in-band manner. The control subnet is 10.0.0.0/24, and the OpenFlow subnet IP address of the data plane is 192.168.1.0/24. In addition, all switches can run the NTP protocol, which allows millisecond-level clock synchronization with the controller.

Figure 3 shows the flow chart of the embodiment of the present invention, the process starts at step S101, and then in step S102, the user sends a measurement command, OFMP enables the controller to analyze the measurement path and index, constructs a measurement message and writes it into the measurement information, send measurement message; in step S103, the first-hop OFMP enable switch parses the measurement message, and writes the measurement information; in step S104, it is judged to measure the time delay between the control plane and the data plane, if so, In the step S111, send the measurement message to the OFMP controller, otherwise turn to the step S105, and send the measurement message to the next hop switch; then go to the step S106, judge whether it is an OFMP switch, if it is a step S107, judge whether the OFMP switch It is the last hop of the path; if yes, turn to step S111, otherwise turn to step S108, judge whether it is a round-trip measurement and the OFMP switch is the midpoint of the path; if yes, turn to step S109, send a measurement message in the opposite direction of the original path, and then Go to step S105, otherwise directly go to step S105. In step S111, turn to step S112, the OFMP controller processes the measurement message and displays the measurement information, and finally goes to S113, the processing flow ends.

Example

The present embodiment provides that the working process of carrying out end-to-end path active measurement to network environment as shown in Figure 1 is as follows: at first generate TCP background flow between S1 to S6 with Spirent Avalanche traffic generator, and be by controller for The background flow establishes a flow table so that its route is S1-S2-S3-S5-S6 (that is, path 1 in Figure 1). The NTP protocol is used for clock synchronization between the switch and the controller. Then, the controller calls the measurement application program to measure the end-to-end performance of the path, and uses PACKET-OUT to initiate the measurement. The starting point of the measurement is S1, and the end point is S6. When the two-way measurement mode is used, the measurement message arrives at S6 and returns to S1 according to the original path, and S1 uses PACKET-IN to return the measurement message with the time stamp of the controller and each switch to the controller, and the controller processes these node identifications and timestamp to get the measurement result. Figure 3 shows the average value of hop-by-hop one-way delay along the route (the average value of 100 measurements).

The test results show that by using the Dpid and time stamp sequences obtained from the active measurement messages sent by the system, we obtain performance parameters such as hop-by-hop delay and end-to-end one-way delay. Among them, the time delay of each switch node is greater than 0.1ms. Figure 3 shows that as the background traffic increases, the flow packet data processed by the switch increases, and the hop-by-hop delay and end-to-end one-way delay also increase, which is in line with the actual situation of the network. Hop-by-hop average delay and variance calculations are shown in Table 1.

Table 1

From the measurement data in Figure 3 and Table 1, it can be seen that when the background traffic gradually increases from 10 Mbps to 600 Mbps, the delay of each hop of the network is gradually increasing, which is caused by the increase of network packet queuing delay. Under 100 measurements, the variance of the hop-by-hop delay is small, which indicates that the measurement results of the system are relatively stable. For network switches, since the flow identifiers of measurement packets and background flows are consistent, the forwarding behavior is also consistent. Therefore, the Dpid and timestamp sequence carried by measurement packets can truly reflect the performance parameters of the current end-to-end path in the network under test. .

Both ends are OFMP-enabled nodes and there are some non-OFMP-enabled switches (that is, ordinary OpenFlow switches) in the middle of the route. Other conditions are the same as the previous experiments. Specifically, the route established by the controller is: S1-S2-S4-S5-S6 (path 2 in FIG. 1 ). Among them, the commercial OpenFlow switch S4 runs OpenvSwitch1.9.2, the OpenFlow protocol version is 1.0, it does not support OFMP, and other nodes are OFMP-enabled. At this time, the controller initiates two-way measurement, the starting point is S1, and the end point is S6. After receiving the measurement message, S6 transmits it in the opposite direction along the path, and finally S1 returns the measurement message to the controller. The analysis shows that the measurement message has the node identification and time stamp sequence written by the controller, S1, S2, S5 and S6, but does not have any information of S4.

Comparison of measurement results in Table 2 Test 1 and Test 3

Table 2 compares the node sequence and path round-trip delay carried by the measurement message in Test 1 and Test 3. It can be seen that although paths 1 and 2 are different, the average round-trip delays of the two paths are basically the same.

The parts not involved in the present invention are the same as the prior art or can be realized by adopting the prior art.

Claims (8)

1. a method for Active measuring OpenFlow network end-to-end path performance, is characterized in that, on the basis of OpenFlow specification, it comprises the following steps:

A. user sends the measurement instruction of network end-to-end path performance to the enable controller of OFMP, after the measurement instruction of the network end-to-end path performance that user sends received by the enable controller of OFMP, resolve measurement instruction and obtain measurement index and measuring route, structure OFMP measured message also writes the local measurement information of the enable controller of OFMP in measured message, jump OFMP enabled switch according to measuring route head wherein and send this measured message, described OFMP measured message comprises the local measurement information of each OFMP enabled switch in measuring route and measuring route; Wherein, first jumping and final jump switch are OFMP enabled switch;

B. first jumping OFMP enabled switch receives the OFMP measured message from the enable controller of OFMP, local measurement information is write in measured message, send according to measuring route down hop switch, if this switch is OFMP enabled switch, in measured message, then write the local measurement information of this OFMP enabled switch, then down hop switch sends this measured message; If this switch is not OFMP enabled switch, then send this measured message to its down hop switch again; The all switches traveled through in measuring route complete the forwarding of OFMP measured message successively, and final jump OFMP enabled switch is returned the enable controller of OFMP after OFMP measured message write local measurement information;

C. the enable controller of OFMP is according to the requirement of measurement index, processes, obtains end-to-end path performance parameter, and show these performance parameters to user to the local measurement information sequence of each enabled switch in OFMP measured message.

2. the method for Active measuring OpenFlow network end-to-end path performance according to claim 1, it is characterized in that, in steps A: the enable controller of OFMP refers to that can understand OpenFlow measures agreement OFMP, and performs the OpenFlow controller of corresponding function; OFMP switch refers to that can understand OpenFlow measures agreement OFMP, and performs the OpenFlow switch of corresponding function.

3. the method for Active measuring OpenFlow network end-to-end path performance according to claim 1, it is characterized in that, in steps A: measuring route refers to that measured message sends from the enable controller of OFMP in OpenFlow network, required by measurement index and the switch sequence set until return to the path of the enable controller of OFMP, in this path the part of switch sequence and the path of measured stream consistent.

4. the method for Active measuring OpenFlow network end-to-end path performance according to claim 1, it is characterized in that, in steps A: measurement index refers to the performance measure of end-to-end path, comprise route namely measure grouping the path node sequence of process, round-trip delay, One Way Delay, packet loss, switch hop-by-hop time delay and between control plane and datum plane time Yanzhong one or more.

5. the method for Active measuring OpenFlow network end-to-end path performance according to claim 1, it is characterized in that, in steps A: OFMP is the agreement that OFMP enabled switch communicates with between the enable controller of OFMP or OFMP enabled switch, the message format of OFMP comprises elementary field: Version field is for representing OFMP version number, Flag field comprises the flag of multiple metering system, Identification field and Sequence Number field are for identifying different Active measuring processes and the different messages in identical Active measuring process, Path Pointer field represents the pointer of the current location inserting path measurements record, Measurement Start field and Measurement End field are for the Dpid value of the starting point and terminal switch that identify measuring route, Path Record field arrives the moment of this equipment for the MAC and measured message depositing this node identification Dpid or main frame.

6. the method for Active measuring OpenFlow network end-to-end path performance according to claim 1, it is characterized in that, local measurement information comprises identifier and the local clock time of local device, and the identifier of described local device is Dpid or MAC Address.

7. the method for Active measuring OpenFlow network end-to-end path performance according to claim 1, it is characterized in that, in step B, first jumping OFMP enabled switch receives the OFMP measured message from the enable controller of OFMP in PACKET OUT mode, final jump switch, then return this measured message in PACKET IN mode to the enable controller of OFMP.

8. the system that adopts of the method for an Active measuring OpenFlow network end-to-end path performance, it is characterized in that, on the basis of OpenFlow specification, OpenFlow Active Measurement System comprises an enable controller of OFMP and at least two OFMP enabled switch, during systematic survey end-to-end path, the head in path jumps and final jump switch is OFMP enabled switch.