JP2005347894A - Computer system and reception interruption processing method thereof - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a computer system capable of preventing consumption of a CPU time due to reception of excessive packets without the need for changing a scheduling mechanism of an operating system and preventing excessive consecution of an interruption disabled state. <P>SOLUTION: The computer system measures the number of reception interruption times from a NIC 10 and the utilizing time of a CPU due to the reception interruption. The measurement results are reset by a prescribed time period. On every occurrence of interruption, the computer system compares a threshold value with a result of measurement of the number of interruption times, inhibits the reception processing when the measurement result exceeds the threshold value and aborts received packets. Similarly, the computer system compares the measurement result of the CPU utilizing time with a threshold value, inhibits the reception processing when the measurement result exceeds the threshold value and aborts the received packets. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to a computer system connectable to a network and a method for processing a reception interrupt from a network connection apparatus.

  In a computer equipped with a modern operating system, when an interrupt from hardware occurs, a program being executed at that time is interrupted where possible, and interrupt processing is executed with priority. About the interrupted program, execution is restarted from the place where it interrupted after completion | finish of interruption processing. Some interrupt processing from hardware has a higher priority that must be executed immediately, while others have a lower priority than other programs interrupted by an interrupt. For this reason, in general, when an interrupt from hardware occurs, only the interrupt processing with a high priority is executed, and the execution timing of other interrupts is left to the scheduler.

  However, in this method, when interrupt processing frequently occurs, CPU time is consumed by the interrupt processing, and CPU time allocated to other processing, for example, an application program is reduced. In the worst case, the application program cannot be operated at all. In this case, it seems to the user that the operation of the application program being executed is stopped and the computer itself is stopped.

  As a specific example, when a small packet is frequently received by a network interface card (NIC) that is a network connection device, CPU time is consumed only by processing the received packet, and the operation of other application programs There are situations where it becomes very slow or completely inoperable. The act of intentionally generating such a situation to stop a service provided by a computer is called a DoS (Denial of Services) attack.

  In addition, even if not intentional, this situation may occur due to a misconfiguration of the network device or the like, and there may occur a situation in which an application that the computer originally wants to execute cannot be executed.

  In order to avoid such a situation, a method of changing the scheduling mechanism of the operating system and reducing the CPU time allocated to the execution of the interrupt processing from the hardware according to the importance of the application can be considered. However, there is a case where the internal change of the operating system is impossible, and even if it is possible, there is a problem that a great cost is required.

In addition, in order to avoid overloading the CPU while receiving bursts of data from the network, receive a sufficient number of packets after the last interrupt and wait for a certain time after the last interrupt There is a technique for enabling an interruptable state by continuing the state (see, for example, Patent Document 1).
JP-A-9-223091 (paragraph 0017)

  However, with this known technique, depending on the number of packets and the time threshold, there is a possibility that the interrupt disabled state may continue excessively, especially when burst reception occurs regularly due to misconfiguration of network equipment. Becomes higher.

  The present invention provides a computer system and an interrupt control method thereof capable of preventing consumption of CPU time due to excessive packet reception and preventing excessive continuation of an interrupt disabled state without changing the scheduling mechanism of the operating system. It is intended to provide.

  In order to achieve the above object, a computer system according to a first aspect of the present invention includes a network connection device that generates an interrupt with respect to reception of a packet from a network, and an interrupt count measuring unit that measures the number of interrupts from the network connection device. And the reset means that resets the measurement result of the interrupt count measurement means at a predetermined time period, and every time an interrupt occurs, the measurement result of the interrupt count measurement means is compared with the threshold value, and the measurement result exceeds the threshold value. If there is a reception process, the reception process restriction means for restricting the reception process is provided.

  According to a second aspect of the present invention, there is provided a network connection device that generates an interrupt for reception of a packet from a network, and a CPU usage time measuring unit that measures a CPU usage time due to a reception interrupt from the network connection device. , The reset means for resetting the measurement result of the CPU usage time measurement means at a predetermined time period, and the measurement result of the CPU usage time measurement means is compared with the threshold value every time a reception interrupt occurs, and the measurement result sets the threshold value. If it exceeds the maximum number, reception processing restriction means for restricting reception processing is provided.

  Further, in the computer systems of the first and second inventions, the reception process restricting means may prohibit the reception process for the current reception interrupt and discard the received packet.

  According to the computer system and the reception interrupt processing method of the present invention, it is possible to prevent consumption of CPU time due to excessive packet reception without changing the scheduling mechanism of the operating system and to prevent excessive continuation of the interrupt disabled state. Can do.

  Embodiments of the present invention will be described below with reference to the drawings.

  FIG. 1 is a block diagram showing an electrical configuration of a computer system according to an embodiment of the present invention.

  As shown in FIG. 1, a computer system 100 includes a central processing unit (CPU) 1 that controls the entire system, a main memory 2 that is used as a work area for the CPU 1, and an HDD that is a recording device that can read and write programs and data. Recording device 3 such as (Hard Disk Drive), tuner 4 that receives the digital broadcast signal, decoder 5 that decodes the digital broadcast signal received by the tuner 4, and data that is decoded by the decoder 5 is efficiently recorded in the recording device 3. And an MPEG encoder 6 that encodes the compressed encoded data of MPEG (Moving Picture Experts Group) 2, which is a possible file format. Data encoded by the MPEG encoder 6 is stored in the main memory 2 via the system bus 7 under the control of the CPU 1, and is grouped into blocks and recorded in the recording device 3.

  The computer system 100 also includes an MPEG decoder 9 that receives data read from the recording device 3 through the system bus 7 and decodes the data into a format that can be displayed on the monitor 8 under the control of the CPU 1. Further, a network connection device (hereinafter referred to as “NIC”) 10 for processing connection with the network is connected to the system bus 7.

  FIG. 2 is a diagram showing a software configuration in the computer system 100.

  An OS (operating system) 11 manages the hardware management of the entire computer system 100 and the execution of the application programs 12 and 13. The device drivers 14 and 15 provide the OS 11 with a procedure for accessing each device. Here, the device driver 14 is a device driver for the NIC 10 of FIG. Hereinafter, this is referred to as a NIC driver. The NIC driver 14 has an interrupt handler 21 that is called by the OS 11 when an interrupt is generated from the NIC 10 and processes the interrupt.

  FIG. 3 is a flowchart showing a basic processing flow by the interrupt handler 21.

  The interrupt handler 21 first determines whether the generated interrupt factor is an interrupt factor to be processed (step 301). Examples of interrupt factors to be processed include transmission completion and reception. When the interrupt factor to be processed is determined (YES in step 301), the interrupt handler 21 locks the operation on the NIC 10 (step 302). Subsequently, the interrupt handler 21 determines what is the interrupt factor to be processed (step 303), and executes processing corresponding to the determined interrupt factor (step 304). After executing the process corresponding to the interrupt factor, the operation lock on the NIC 10 is released (step 305).

  FIG. 4 is a flowchart showing an example of a reception process executed as one of processes corresponding to the interrupt factor in FIG.

  In this reception processing, first, after securing a buffer corresponding to the size of the received packet (step 401), the received packet is copied to the buffer (step 402), and the attribute of the received packet is set in the buffer (step 403). ). Subsequently, the buffer is passed to the upper network driver of the reception process (step 404), and the process in the upper network driver is executed.

  In general, these processes are executed in a privileged mode called a kernel mode in UNIX (registered trademark), and are executed with a higher priority than a user application executed in a non-privileged mode. For this reason, when reception interruptions frequently occur, there is a possibility that the CPU time that can be used by the user application is completely lost.

  In the computer system 100, such a problem is solved by the following processing procedure implemented in the NIC driver 14.

  FIG. 5 is a flowchart showing a part of the initialization process of the NIC driver 14, and FIG. 6 is a diagram showing a configuration of an area secured on the main memory 2 in the initialization process of the NIC driver 14.

  In the initialization process, the NIC driver 14 secures an interrupt count storage area 31 on the main memory 2, and sets 0 as the “interrupt count” (step 501). The interrupt count storage area 31 is an area for storing the number of reception interrupts received from the NIC 10 within a predetermined time.

  Next, the NIC driver 14 secures the CPU usage time storage area 32 on the main memory 2, and sets 0 as the “CPU usage time” (step 502). The CPU usage time storage area 32 is an area for storing the total CPU usage time used in the reception process of the interrupt process from the NIC 10 within a predetermined time.

  Next, the NIC driver 14 secures an interrupt count threshold storage area 33 on the main memory 2, and sets an “interrupt count threshold” therein (step 503). As the interrupt count threshold, the number of interrupts received from the NIC 10 that is allowable per unit time is set. However, an optimal value may be prepared for the interrupt frequency threshold value in advance according to the system and the application to be operated, and the interrupt frequency threshold value may be switched and used.

  Next, the NIC driver 14 secures a CPU usage time threshold storage area 34 on the main memory 2 and sets a “CPU usage time threshold” therein (step 504). As the CPU usage time threshold, an allowable value of CPU usage time per unit time used by the reception interrupt process is set. However, this CPU usage time threshold value may be prepared in advance according to the system and the application to be operated, and used by switching.

  Further, as shown in FIG. 7, the NIC driver 14 performs the following processing at regular intervals. That is, the NIC driver 14 clears (sets 0) the “interrupt count” in the interrupt count storage area 31 (step 701) each time a certain time elapses, and then continues to “CPU” in the CPU usage time storage area 32. “Use time” is cleared (set to 0) (step 702).

  Furthermore, the NIC driver 14 performs the following process when the reception process is activated by an interrupt from the NIC 10.

  FIG. 8 is a flowchart showing the flow of processing by the NIC driver 14 when reception processing is activated by an interrupt.

  The NIC driver 14 starts measuring the “CPU usage time” in the reception process (step 801), and increments the “interrupt count” in the interrupt count storage area 31 (step 802).

  Subsequently, the NIC driver 14 compares the “CPU usage time” stored in the CPU usage time storage area 32 with the “CPU usage time threshold” stored in the CPU usage time threshold storage area 34 (step 803). If the “CPU usage time” is greater than the “CPU usage time threshold” (YES in step 803), the measurement of the “CPU usage time” is terminated, and the measurement result is added to the contents of the CPU usage time storage area 32 for processing. Is finished (step 806). That is, the received packet is discarded without performing the reception process for the occurrence of the current interrupt. At this time, if necessary, the NIC 10 is controlled to discard the received packet.

  When “CPU usage time” is equal to or less than “CPU usage time threshold” (NO in step 803), “interrupt count” in the interrupt count storage area 31 and “interrupt count threshold” in the interrupt count threshold storage area 33 are compared. (Step 804). When the “interrupt count” is equal to or less than the “interrupt count threshold” (NO in step 804), the reception process of FIG. 4 is performed (step 805).

  If “interrupt count” is greater than “interrupt count threshold” (YES in step 804), the measurement of “CPU usage time” is terminated, and the measurement result is added to the contents of the CPU usage time storage area 32 for processing. Is finished (step 806). That is, the reception process shown in FIG. 4 is not performed and the packet is discarded. If necessary, the NIC 10 is controlled to discard the received packet.

  In the above processing, when “CPU usage time” is greater than “CPU usage time threshold” or “interrupt count” is greater than “interrupt count threshold”, the number of interrupts or CPU time required for reception processing is assumed as an amount. It means that the upper limit of is exceeded. If this state is left unattended, the application programs 12 and 13 cannot be executed in the worst case, and the service as the computer system 100 may not be provided. Therefore, if “CPU usage time” is larger than “CPU usage time threshold” or “interrupt count” is larger than “interrupt count threshold”, the received packet is lost by discarding the packet without performing the reception process. Nevertheless, the provision of services as the computer system 100 can be continued. Further, there is no problem that a packet that could not be received is subjected to reception processing when it is retransmitted after a time interval. Furthermore, since the contents of the CPU usage time storage area 32 and the interrupt count storage area 31 are reset at regular intervals, the interrupt disabled state does not continue for an excessively long time.

  As described above, according to the computer system 100 of the present embodiment, it is possible to avoid excessive consumption of the CPU time due to the reception interrupt of the NIC 10, so that the recording apparatus 3 undergoes data compression encoding from decoding of the digital broadcast signal received by the tuner 4. It is possible to prevent the operation of the CPU 1 from becoming a bottleneck in the process of recording data and the process of reading out the data recorded in the recording device 3 and decompressing and decoding the data.

  It should be noted that the present invention is not limited to the illustrated examples described above, and it is needless to say that various modifications can be made without departing from the scope of the present invention.

  For example, in the description of the above embodiment, the “interrupt count” of reception within a certain time and the “CPU usage time” used for the determination are used for the determination, but the “interrupt count”, “ An average value of “CPU usage time” or a moving average value may be used.

  In the description of the above embodiment, when “CPU usage time” is greater than “CPU usage time threshold” or “interrupt count” is greater than “interrupt count threshold”, the reception process is not performed and the packet is discarded. However, the following variations can also be applied.

  In the reception process of FIG. 4, the process is performed until the received packet in step 402 is copied to the buffer, and the process of setting the packet attribute in the subsequent step and the process of passing the buffer to the upper network driver are performed a plurality of times. It is also possible to execute all of them after receiving the reception interrupt. As a result, the number of executions of the upper network driver can be reduced. This leads to an effect that the overhead for starting the scheduler can be reduced in the case of an implementation in which the scheduler is started to execute the upper network driver.

  Also, if the “CPU usage time” is greater than the “CPU usage time threshold” or if it is determined that the “interrupt count” is greater than the “interrupt count threshold”, it is determined that the network is in an abnormal state and received. It is also possible to prohibit the interruption and allow the NIC 10 to receive a reception interrupt after a certain period of time, in anticipation of the possibility of the network state recovering thereafter.

It is a block diagram which shows the electric constitution of the computer system concerning one Embodiment of this invention. It is a figure which shows the structure of the software in the computer system of FIG. It is a flowchart which shows the flow of the basic process by an interrupt handler. FIG. 4 is a flowchart illustrating an example of a reception process executed as one of processes corresponding to an interrupt factor in FIG. 3. 5 is a flowchart showing a part of initialization processing of a NIC driver. It is a figure which shows the structure of the area | region ensured on the main memory in the initialization process of FIG. It is a flowchart which shows the reset process performed for every fixed time of a NIC driver. It is a flowchart which shows the flow of a process by a NIC driver when a reception process is started by interruption.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 1 ... CPU, 2 ... Main memory, 3 ... Recording apparatus, 10 ... Network connection apparatus (NIC), 14 ... NIC driver, 21 ... Interrupt handler, 31 ... Interrupt count storage area, 32 ... CPU use time storage area, 33 ... Interrupt count threshold storage area, 34... CPU usage time threshold storage area.

Claims (5)

A network connection device that generates an interrupt for reception of a packet from the network;
Interrupt number measuring means for measuring the number of interrupts from the network connection device; reset means for resetting the measurement result of the interrupt number measuring means at a predetermined time period;
A reception processing restriction unit that compares a measurement result of the interruption frequency measurement unit with a threshold every time the interruption occurs, and gives a restriction to reception processing when the measurement result exceeds the threshold; A featured computer system. A network connection device that generates an interrupt for reception of a packet from the network;
CPU usage time measuring means for measuring CPU usage time due to a reception interruption from the network connection device;
Resetting means for resetting the measurement result of the CPU utilization time measuring means at a predetermined time period;
Each time a reception interrupt occurs, a measurement result of the CPU usage time measurement means is compared with a threshold, and if the measurement result exceeds the threshold, a reception processing restriction means for restricting reception processing is provided. A computer system characterized by the above.   The computer system according to claim 1, wherein the reception processing restriction unit prohibits reception processing for a current reception interrupt and discards the received packet. Measuring the number of interrupts from a network connection device that generates an interrupt for reception of a packet from the network;
Resetting the interrupt count measurement result at a predetermined time period;
And a step of comparing the measurement result of the number of interruptions with a threshold each time the interruption occurs, and prohibiting reception processing for the interruption when the measurement result exceeds the threshold. System receive interrupt handling method. Measuring the CPU usage time due to a reception interrupt from a network connection device that generates an interrupt for reception of a packet from the network;
Resetting the measurement result of the CPU usage time at a predetermined time period;
Comparing the measurement result of the CPU usage time with a threshold every time the interrupt occurs, and prohibiting reception processing for the interrupt if the measurement result exceeds the threshold. Receiving interrupt processing method for a computer system.

Images