CN1275429C - A method for performance measurement data acquisition of embedded distributed system - Google Patents

Abstract

The invention discloses a method for collecting performance measurement data of an embedded distributed system, which includes the following processing steps: the first step: master-slave processor initialization; second step: the master processor starts to collect performance data and sends data to each slave processor Send the collection command; Step 3: Each slave processor collects performance data and sends it to the master processor; Step 4: The master processor backs up all the sampling data in this cycle, clears the unified buffer, and starts the next step. Periodic measurement and counting; the fifth step: if the timer T3 on the main processor is overtime to the background report, then the main processor performs statistical calculations on all sampling data in the T3 period and sends them to the background, and then turns to the second step, otherwise Go directly to step two. By adopting the method of the invention, compared with the prior art, the stability of the embedded distribution system and the reliability of data transmission are improved, and the influence on other business modules is reduced.

Description

A performance measurement data acquisition method for an embedded distributed system

technical field

The invention relates to a method for collecting performance measurement data of an embedded distributed system, which belongs to the field of communication.

Background technique

In the field of communication, especially embedded distributed systems, performance data is the basis to reflect the system business volume and the success rate of business processes, and is also an important indicator for evaluating system reliability and stability. At the same time, faults in the positioning system also play a very important role Therefore, it is essential to measure and count various performance indicators in the system. A good data collection method is particularly important, because the collection of performance measurement will have a certain impact on the entire business system. Since the performance indicators come from various business modules, and the business modules may be distributed on the main processor and the slave processors, the performance measurement includes the data collection and reporting of the master processor and the slave processor business modules. The current performance measurement data collection method for embedded distributed systems is as follows:

1. The performance measurement module on the main processor sets the periodic acquisition timer T2, and sets up a unified performance measurement buffer, and the business module on the main processor writes a unified performance measurement buffer; each slave processor counts each counter from the beginning.

2. After the main processor receives the performance measurement command from the background, it sets the reporting timer T3 to the background (T3 is an integer multiple of T2), and sends a performance measurement reporting command (including the list to be tested) to each slave processor when T2 times out , reset timer T2.

3. After receiving the performance measurement reporting command, each slave processor will reply to the master processor one by one with the measurement results of all measured objects according to the measurement list, and reset the counter on the slave processor to start the next step. Period counter counts.

4. When the main processor receives the responses from the slave processors, record the results, and send the measurement command to the corresponding slave processors again when T2 times out, and start the next cycle of data collection.

5. When T3 times out, the performance measurement module on the main processor reports the sampling results within T3 to the background.

6. When the performance measurement module on the main processor receives the background measurement stop command, it kills the timer T3 and clears the measurement list.

The above method has the following disadvantages in the case of a large business volume:

1. Greatly reduced system stability. This is mainly reflected in the data transmission between the slave processor and the master processor (the communication between the master and slave processors can be through T network or Ethernet). Equipped with 48 slave processors (maximum business conditions), since the measurement data reporting messages of the slave processors are uniformly sent by the master processor at the same time, each slave processor reports data almost at the same time, so that the master processor will receive data continuously each time Up to 80*48=3840 messages (ie measurement data). For the main processor with high load requirements, processing so much data in a short time is bound to make things worse and cause the system to crash.

2. Reduce the business processing rate. An important feature in the embedded environment is real-time, and the speed is a matter of great concern to users, so the processing speed of each slave processor plays a decisive role in the speed of the entire distributed system. At present, the performance measurement and reporting of slave processors is done by the business processing modules on the slave processors. When there are many measurement objects, the business processing capability of each slave processor is bound to be reduced, thereby affecting the speed of the distributed system.

3. Poor transmission reliability. In the case of a large business, even if the main processor does not crash, frequent packet sending will make packet loss very easy during transmission, making the data (including business data) received by the main processor unreliable.

Contents of the invention

The invention proposes a method for collecting performance measurement data of an embedded distributed system, which overcomes the shortcomings affecting system stability in the prior art, and solves the problems in the prior art that affect the business processing rate and unreliable measurement data.

The performance measurement data acquisition method of the embedded distributed system of the present invention comprises the following processing steps:

Step 1: Master-slave processor initialization; the master processor opens a unified buffer area, and starts the measurement and counting on the master processor, and directly writes the unified cache; the slave processor increases the measurement data reporting process, and starts the measurement and counting on the slave processor, The process has low priority.

Step 2: The master processor starts statistics and sends collection commands to the slave processor;

Step 3: Collect performance data from the processor and send it in a package;

Step 4: The main processor backs up the data collected in this period. When the periodic recording timer T2 on the main processor expires, the main processor records the data collection results of this cycle (including the collection results of the main processor business module and the results reported by the slave processor), and clears the unified buffer to start the next measurement cycle count of

Step 5: If the timer T3 reporting to the background on the main processor is overtime, the main processor will perform statistical calculations on the T3/T2 sampling results within the T3 period and send them to the background, and then go to the second step, otherwise go directly to Go to the second step (that is, repeat the second, third, and fourth steps).

Step 6: When the main processor receives the background stop measurement command, it clears the corresponding measurement list, clears the periodic collection timer T1, the periodic recording timer T2 and the reporting period T3 to the background, and no longer collects measurement data.

Wherein the second step includes the following steps:

(1) Start the measurement and counting of the service modules on the main processor. Since the unified buffer is directly written when counting the values of each counter, no special message trigger is required.

(2) The main processor sets a periodic acquisition timer T1 and a periodic recording timer T2 (T1<T2).

(3) After the performance measurement module on the main processor receives the measurement command from the background (the performance measurement list contains the objects to be measured to be collected), it starts to set the reporting timer T3 to the background (T3 is an integer multiple of T2).

(4) When the periodic acquisition timer T1 expires, the performance measurement module on the master processor sends a measurement data reporting command to the corresponding slave processor according to the measurement list, and there is a delay t1 between multiple commands.

Wherein the third step includes the following steps:

(1) Count the respective counters from the business process on the processor.

(2) Collect the measurement data of the measured object from the processor measurement reporting process.

(3) Backup the data of this period from the measurement and reporting process of the processor, and start the statistics of the next period.

(4) The measurement data reporting process of the slave processor classifies the backup measurement data according to the measurement list, the size of each measured object data and the size of the master processor and network communication data packet, and forms multiple measured objects into one Bag.

(5) The measurement data reporting process of the slave processor sends each packet of data to the main processor, and there is a certain delay t2 between packets.

(6) The performance measurement module of the master processor returns a confirmation message after receiving the data packet, and the slave processor sets the reception success bitmap according to the confirmation message.

(7) Set a retransmission timer T4 after sending all the assembled data packets from the measurement data reporting process of the processor.

(8) When the retransmission timer T4 expires, it is decided whether to retransmit the corresponding packet according to the received success bitmap (that is, the confirmation message of each packet).

T1, T2, T3 are timers on the master processor, and T4 is a timer on the slave processor. The third step above must be completed within (T2-T1-t1×n) seconds, where n is the number of slave processors, thus requiring (t2×maximum possible number of packets)×2+T4+t1×n<(T2 -T1). In addition, for some basic measurement items that do not require background measurement commands, the performance measurement method is similar to the above method, except that the performance measurement module of the master processor sends measurement commands to all slave processors that have business data, instead of according to the measurement task list.

Using the method of the present invention has the following advantages:

1. Reduce the impact of performance measurement on the processor on other business modules. A process is added to the slave processor, which has a lower priority and is specially used for reporting measurement data, thus reducing the burden on the slave processor business processing module, improving the slave processor business processing capability, and helping to improve the business of the distributed system rate.

2. Improved the stability of the system. Using the method of the present invention, the packet sending greatly reduces the number of messages from the processor and the master processor, and the delayed transmission between each message packet reduces the burden on the communication between the processor and the master processor, (T2-T1-t1 ×n) The buffer time not only reduces the transmission frequency, but also greatly reduces the burden on the main processor, which is very beneficial to improve the stability of the system.

3. Improve the reliability of data transmission. Through the retransmission mechanism of message confirmation, the probability of incorrect measurement data due to packet loss in transmission is greatly reduced, and the transmission reliability and authenticity of measurement data are improved.

4. By adopting the method of the present invention, compared with the prior art, the stability of the embedded distribution system and the reliability of data transmission are improved, and the influence on other business modules is reduced.

Description of drawings

Figure 1 is a schematic diagram of performance measurement.

Fig. 2 is a flowchart of performance measurement data realization.

Fig. 3 is the decomposition of "the master processor starts counting and sends a collection command to the slave processor" in Fig. 2 .

Figure 4 is the decomposition of "collecting data from the processor and sending it in packets" in Figure 2.

Detailed ways

Below in conjunction with accompanying drawing, the implementation of technical scheme is described in further detail:

As shown in Figure 1, the main processor includes a performance measurement module and other business modules. On the main processor, the performance measurement module uniformly opens a buffer, and other business modules are responsible for counter counting and directly write to the unified buffer. After the performance measurement module of the main processor receives the measurement command from the background, if the periodic acquisition timer T1 on the main processor is overtime, the measurement command is sent to the performance measurement report process of the corresponding slave processor according to the measurement list, and the slave processor is in (T2- T1-t1×n) time to complete the packaged transmission of measurement data (there is a delay t2 between packets) and clear the measurement counter on the slave processor, and then decide whether to send it for the second time according to the response of the master processor. The performance counters on the slave processor are accumulated and counted by other business modules, and the performance measurement reporting process of the slave processor is only responsible for the packaging and sending of the measurement results and the clearing of the counter; after the periodic recording timer T2 on the master processor expires, the master The computer records the measurement results returned from the slave processor and the counting results accumulated by other modules of the main processor, clears the unified buffer, resets timers T1 and T2, and starts counting in the next cycle. When the reporting timer T3 times out to the background, the performance measurement module of the main processor performs simple calculations on the measurement results in T3/T2 sampling periods and sends them to the background, and the background generates a report for a certain period of time according to the reported results.

Refer to Figures 2, 3, and 4. First, the system is powered on and initialized. The main processor sets up a unified performance measurement buffer, and starts to accumulate the counters that need to be counted by the main processor. After the slave processor is powered on, the process of reporting measurement data is increased. The business module starts counter accumulation; after the measurement task is built in the background, the performance measurement list is sent to the main processor, and includes the value of reporting period T3 to the background (T3 is an integer multiple of T1); the performance measurement module of the main processor receives the measurement of the background After the command, set the periodic acquisition timer T1, the periodic recording timer T2, and the reporting timer T3 to the background; when T1 times out, the master processor sends the acquisition command to the measurement data reporting process of the corresponding slave processor, and there is a gap between multiple commands. Delay t1 to stagger the time of reporting data from different slave processors as much as possible; after receiving the acquisition command, the measurement data reporting process of each slave processor, according to the measurement list, the size of each measured object data, the main processor it belongs to, and the network communication Due to the limitation of data packet size, the measurement data of multiple measured objects will be combined into one packet, and all counters will be cleared at the same time, and then each packet will be sent to the main processor. The performance measurement module of the machine returns a confirmation message after receiving a packet of data every time, and informs the slave processor whether the packet data is received successfully; a retransmission timer T4 is set after the slave processor measurement data reporting process has sent all the data packets of the time; When the sending timer T4 times out, the process of reporting measurement data from the processor decides whether to resend the corresponding packet according to the bitmap of successful reception (that is, the confirmation message of each packet). (T2-T1-t1×n) time to complete. When the periodic recording timer T2 is overtime, the main processor performance measurement module records the data collection results of this cycle, clears all counters, resets timers T2 and T1, and starts the counting of the next measurement cycle; when the periodic collection timer T1 When timeout occurs again, the measurement command is sent to the slave processor again, and the above process is repeated; when the reporting cycle T3 times out to the background, the master processor performs a simple calculation on the T3/T2 sampling results in the T3 cycle and sends it to the background; if the master processor When the machine performance measurement module receives the performance measurement end command from the background, it clears the corresponding measurement list, and clears the reporting period T3 to the background, and the performance measurement reporting process ends.

Example:

Since the whole method involves the value of some parameters, it will be explained by giving an example of the parameter value in the process of realizing the performance measurement data acquisition method in the GPRS system.

Performance measurement data acquisition method in GPRS system: GPRS system is an embedded distributed system, the main processor is the front desk, and the slave processor is all A-type single boards and B-type single boards. It is assumed that a maximum of 48 A-type single boards and 16 One class B board and one class A board can be configured with up to 80 cells. The communication data packet size is limited to less than 1400Byte, and the measurement data size of each cell is 80×4Byte=320Byte, so according to our method, 4 cells can be sent in one packet, and the size of each packet is 4×320Byte=1280Byte<1400Byte. If a class A board is configured with 80 cells, 80/4=20 packets need to be reported. Can set:

Foreground cycle sampling timer T1 = 270 seconds

Foreground cycle recording timer T2 = 300 seconds

Foreground and background synchronization timer T3 = 900 seconds

The measurement command time interval t1=8 milliseconds sent by the master processor to each slave processor

Slave processor (type A single board) retransmission timer T4 = 4 seconds

Interval time t2 for sending packets from the processor (type A single board) = 400 milliseconds

Among them: (400 milliseconds × 20) × 2+4 seconds + 8 milliseconds × (48+16) = 20.32 seconds < (T2-T1) = 30 seconds, which can ensure that the class A board can complete the reporting of measurement data within 30 seconds . After receiving the performance measurement collection command from the background, the foreground sets timers T1 (270 seconds), T2 (300 seconds) and T3 (900 seconds) at the same time, and when the timer T1 (270 seconds) times out, it sends a message to all A The measurement data reporting process of similar veneer sends the measurement collection command; after the measurement data reporting process receives the measurement collection command, according to the size of the data packet of each measured object (if it is 320byte), the main processor (assuming it belongs to the same main processing unit) Machine) and network communication packet size limit (less than 1400byte), group the measurement data (as described above, 20 packets can be grouped), and clear the counter on the A-type single board for the statistics of the next cycle, and then Send all assembled packets to the foreground at an interval of t2 (400 milliseconds), and set a timer T4 (4 seconds); the foreground will return a corresponding response (set the success bitmap) every time a packet is received. If T4 (4 seconds) of the board is overtime, according to the success bitmap, resend the packet that has not received a response. After resending, the timer T4 will not be set; the front desk will directly write Unified buffer, and when T2 (300 seconds) times out, back up the data in the unified buffer, and clear the unified buffer, reset timer T1 and T2; =3) the record results are simply calculated and sent to the background, and the timer T3 is reset.

Claims (5)

1. a performance measurement data acquisition method of an embedded distributed system, characterized in that, the method comprises the following processing steps: Step 1: Master-slave processor initialization; the master processor opens a unified buffer area, and starts the measurement and counting on the master processor, and directly writes the unified cache; the slave processor increases the measurement data reporting process, and starts the measurement and counting on the slave processor, The reporting process has a low priority; Step 2: The main processor starts to collect performance data, counts the measurement counts of the business modules on the main processor and writes them into the unified buffer; after receiving the measurement command from the background, the performance measurement module on the main processor regularly reports to the corresponding Send measurement data reporting commands from the processor separately, and set a sending delay between each command; Step 3: Count the counters of the business process on each slave processor, collect the measurement data of the measured object from the measurement reporting process of the processor, classify the backup measurement data, form multiple measured objects into a package, and send each package of data to the master Processor, there is a certain delay between packets; Step 4: The main processor backs up all the sampling data in this cycle, clears the unified buffer, and starts the measurement and counting of the next cycle; Step 5: If the timer T3 reporting to the background on the main processor is overtime, the main processor will perform statistical calculations on all sampling data in the T3 period and send them to the background, and then go to the second step, otherwise go directly to the second step step. 2. the performance measurement data collection method of embedded distribution system according to claim 1, is characterized in that, described method also comprises following processing steps: When the main processor receives the background stop measurement command, it clears the corresponding performance measurement list, clears all cycle timers, and no longer collects measurement data. 3. according to the performance measurement data collection method of embedded distribution system described in claim 1 or 2, it is characterized in that, described second step specifically comprises the following processing steps: (1) counting the measurement counts of the business modules on the main processor in the first step and writing them into a unified buffer; (2) Periodic acquisition timer T1 and periodic recording timer T2 are set on the main processor, and make T1＜T2; (3) After the performance measurement module on the main processor receives the measurement command from the background, it starts to set and report the timer T3 to the background; (4) When the periodic acquisition timer T1 expires, the performance measurement module on the main processor sends measurement data reporting commands to the corresponding slave processors according to the measurement list, and a sending delay t1 is set between each command. 4. according to the performance measurement data collection method of embedded distribution system described in claim 1 or 2, it is characterized in that, described 3rd step specifically comprises the following processing steps: (1) Statistics of respective counters from the business process on the processor; (2) Collect the measurement data of the measured object from the measurement reporting process of the processor; (3) Back up the current period data from the processor measurement reporting process, and start the statistics of the next period; (4) The measurement data reporting process of the slave processor classifies the backup measurement data according to the measurement list, the size of each measured object data and the size of the master processor and network communication data packet, and forms multiple measured objects into one Bag; (5) Send each packet of data to the main processor from the processor measurement data reporting process, and there is a certain delay t2 between packets; (6) the master processor performance measurement module returns a confirmation message after receiving the data packet, and the success bitmap is received from the processor according to the confirmation message setting; (7) set up a retransmission timer T4 after sending all grouped data packets from the processor measurement data reporting process; (8) When the retransmission timer T4 expires, it is determined whether to retransmit the corresponding packet according to the received success bitmap. 5. the performance measurement data collection method of embedded distributed system according to claim 4, it is characterized in that, in the described 3rd step, each gathers performance data from processing machine and packs and sends to main processing machine in (T2-T1 -t1×n) seconds, where n is the number of slave processors.

Images