Patent Application
20080270959
Simulating system execution of applications made up of instructions (viz., processor instructions) is commonly performed. Typically, such simulation is performed to ensure that the system and the applications operate as they are expected or intended to operate. Furthermore, such simulation may be performed to verify that systems, applications, and computer programs that were designed for older, outdated hardware are properly executed using more modern hardware. This type of simulation is known as functional simulation. However, other types of simulation of system execution of instructions are also important.
For instance, another type of simulation simulates how long it will take a system to execute the instructions of a given application. This type of simulation is known as timing simulation. As another example, a type of simulation may simulate how much electrical power the system uses to execute the instructions of a given application. This type of simulation is known as power simulation.
The functional simulation of system execution of instructions is commonly decoupled from the timing simulation of system execution of instructions. This typically permits the functional simulation to be performed much more quickly. However, by divorcing the functional simulation from the timing simulation, the functional simulation of system execution of instructions may not be performed as accurately, since some aspects of system execution of instructions can be dependent on the timing of this execution.
The functional simulation component 102 simulates functional execution of instructions 108 on a target system 110, whereas the timing simulation component 104 simulates timing execution of the instructions 108 on the target system 110. The instructions 108 may be part of one or more applications to ultimately be executed on the target system 110. The target system 110 itself can be or include one or more computing devices, such as single-processor computer systems, as well as multiple-processor computer systems.
The functional simulation component 102 simulates the functional execution of the instructions 108 on the target system 110, and thus simulates the functional execution of the target system 110, in that the simulation component 102 determines whether the system 110 and the instructions 108 operate as they are expected or intended to operate. That is, the functional simulation component 102 simulates how the target system 110 operates with respect to the instructions 108. The functional simulation component 102 by itself may not provide any timing information, or may not be able to provide sophisticated or substantially accurate timing information; rather, the functional simulation component 102 may just provide relatively simple timing information. Timing information includes the time it takes for the target system 110 to execute the instructions 108.
By comparison, the timing simulation component 104 simulates how long, or how much time, the target system 110 takes to execute the instructions 108. That is, the timing simulation component 104 simulates how long, or how much time, the target system 110 takes to perform an intended task. The timing simulation component 104 does not test whether the target system 110 properly or correctly executes the instructions 108, which instead is the domain of the functional simulation component 104.
The system 100 can include a communication interface 106, which may be a defined communications protocol, for instance, that defines the exchange of information between the simulation components 102 and 104. The functional simulation component 102 provides functional simulation information 116 to the timing simulation component 104. Similarly, the timing simulation component 104 provides timing simulation information 114 to the functional simulation component 102.
Other than the exchange of information between the simulation components 102 and 104 via the communication interface 106, the functional simulation component 102 and the timing simulation component 104 are at least substantially decoupled from one another. For instance, the timing simulation component 104 can be switched out for, or replaced with, a different timing simulation component 104 that is also capable of operating with the functional simulation component 102. The communication interface 106—i.e., the exchange of information between the simulation components 102 and 104—can be considered as recoupling these decoupled simulation components 102 and 104.
In general, the timing simulation component 104 performs its timing simulation more slowly than the functional simulation component 102 performs its functional simulation. As can be appreciated by those of ordinary skill within the art, the functional simulation component 102 can perform various simulation techniques that speed the functional simulation component 102 performing functional simulation of the target system 110 executing the instructions 108. By comparison, lesser such simulation techniques are available for the timing simulation component 104 to speed performance of the timing simulation of the target system 110 executing the instructions 108.
Furthermore, however, the functional simulation component 102 in at least some situations may not be able to accurately simulate the execution of the instructions 108 by the target system 110 without any timing information. In other words, in many cases, the behavior of the target system 110 may directly depend on the simulated time of various events. For example, many multithreaded libraries use active wait loops, as opposed to yielding the processor on a lock, because of the performance advantage of the resulting short waits. As another example, within an application accessing a network, packets may be resent or network parameters adjusted depending on network congestion. In these and other examples, functionally simulating the target system 110 executing the instructions 108 without taking into account timing information results in a less-than-desirable functional simulation.
Therefore, the functional simulation component 102 takes into account the timing simulation of the target system 110 executing the instructions 108 while performing functional simulation of the target system 110. Advantageously, such influence of the functional simulation by the timing simulation occurs without greatly affecting the performance of the functional simulation. In general, the timing simulation component 104 provides periodic timing-related feedback to the functional simulation component 102, as the timing information 114. The functional simulation component 102 then performs its functional simulation based on this timing-related feedback. The timing simulation component 104 can determine this timing information 114 (i.e., the timing-related feedback) based on the functional simulation information 116 generated continuously by the functional simulation component 102.
By providing timing-related feedback periodically, instead of continuously (e.g., such as at the simulation of each of the instructions 108), the timing simulation component 104 helps the functional simulation component 102 provide more accurate functional simulation, without greatly slowing performance of the functional simulation. For example, the functional simulation component 102 may simulate a real-time clock (RTC) of the target system 110, where the system 110 may be able to execute a certain number of the instructions 108 per each clock cycle, which is referred to as instructions per cycle (IPC). Rather than providing timing-related feedback at the simulation of each of the instructions 108 by the functional simulation component 102, the timing simulation component 104 may instead provide such timing-related feedback once per a number of such clock cycles, such as once per one million such simulated clock cycles, where one million (or other number) of such simulated clock cycles is referred to herein as a simulation cycle. As such, a substantially optimal compromise between functional simulation performance and functional simulation accuracy is achieved.
Once the functional and the timing simulations have been completed, the functional simulation component 102 outputs functional simulation results 112, and the timing simulation component 104 may output timing simulation results 112. These simulation results 112 may then be used by a developer or a designer to ensure that the target system 110 is properly executing the instructions 108, in the proper length of time. Described next in the detailed description is a particular approach by which the functional simulation of the target system 110 takes into account the timing simulation of the target system 110 that is periodically performed.
During each such period of time (202), the functional simulation component 102 performs functional simulation based on a specified number of instructions per cycle (204). That is, rather than functionally simulating the target system 110 without regards to how many instructions the system 110 can execute within a given simulation cycle, the functional simulation component 102 instead functionally simulates the target system 110 based on the knowledge that a specified number of instructions per cycle can currently be executed by the target system 110 being simulated. In this sense, the functional simulation component 102 takes into account timing simulation-related information when performing functional simulation. For the first simulation cycle, the specified number of instructions per cycle may be set to a default value. During functional simulation, and as a part of functional simulation, the functional simulation component 102 outputs functional simulation information 116 (206).
Next, after each period of time (208), the timing simulation component 104 performs timing simulation. Particularly, the timing simulation component 104 adjusts the specified number of instructions per cycle that is to be used by the functional simulation component 102 during the next period of time (210). Adjusting the specified number of instructions per cycle may be achieved based on the functional simulation information 116 output by the functional simulation component 102. For instance, the timing simulation component 104 may analyze this information 116 to determine whether there is significant congestion occurring within the target system 110 that may result in a decreased number of instructions that can be executed within system 110 during the next simulation cycle. The timing simulation information 114 (i.e., the specified number of instructions per cycle as adjusted) is output by the timing simulation component 104 (212), for utilization by the functional simulation component 102 when functionally simulating the target system 110 during the next period of time.
It is noted that the actual number of instructions for a given simulation cycle is not able to be determined by the timing simulation component 104, insofar as the specified number of instructions per cycle is determined for a simulation cycle before the functional simulation component 102 simulates the target system 110 in relation to that simulation cycle. In other words, the timing simulation component 104 predicts how many instructions are likely to be able to be executed by the target system 110 during the functional simulation of the system 110 by the functional simulation component 102 in the next simulation cycle. As such, only previous history (i.e., historical data) regarding the simulation of the target system 110 can be employed to predict the instructions per cycle for the next simulation cycle. This previous history can include previous numbers of instructions per cycle actually simulated by the functional simulation component 102, as well as functional simulation information 116 that has been output by the functional simulation component 102 regarding the functional simulation of the target system 110.
In one embodiment, it has been found that an auto-regressive, moving-average (ARMA) model, adequately predicts the number of instructions per cycle for the next simulation cycle. This model can be mathematically described as the inverse of instructions per cycle—i.e., as cycles per instruction—as follows:
In equation one CPIx is the average cycles per instruction (CPI) for time period x. The constants a0, a1 . . . aN represent the autoregressive parameters (or numbers of poles in the z-transform domain), and the constants m0, m1 . . . mM represent the moving average parameters (or numbers of zeros in the z-transform domain). Furthermore, C{circumflex over (P)}Ix is the previously estimated value of the CPI for time period x, so that the term CPIx−C{circumflex over (P)}Ix represent the previous prediction error at step x. Some examples of specific ARMA models may include an ARMA22 model (two poles, two zeros), corresponding to equation (1) with N=2 and M=2, or a “next-step predictor” model, corresponding to equation (1) with N=1, M=0 and a0=1.0 In one embodiment, captured by the moving average components of equation (1), any error of the instructions per cycle predicted for a current simulation cycle as compared to the actual number of instructions per cycle for that simulation cycle can be diffused over one or more future simulation cycles.
Based on this timing-related feedback from the timing simulation component 104 to the functional simulation component 102 (i.e., as the timing simulation information 114), the functional simulation of the target system 110 executing the instructions 108 is more accurate. For instance, an application being simulated that encompasses the instructions 108 acts as if execution by the target system 110 were faster or slower. Such timing-related influence on the functional simulation ensures that the functional simulation takes into account real-world issues that the target system 110 may encounter, such as network congestion, and so on. Without such timing-related influence on the functional simulation, the functional simulation otherwise acts as if all the instructions 108 can be executed without regards to time.
Therefore, the simulations of parts 202 and 208 of the method 200 can be seen to operate as follows. During each simulation cycle, the functional simulation component 102 functionally simulates the target system 110 executing the instructions 108, based on the specified number of instructions per cycle. Once per simulation cycle, the timing simulation component 104 performs timing simulation of the target system 110 executing the instructions 108 to adjust the number of instructions per cycle for the next simulation cycle within which functional simulation is to be performed. Such periodic punctuation of the functional simulation with timing simulation ensures that the functional simulation is performed in relation to timing information, without affecting performance of the functional simulation too much.
It is noted that most of the simulation of the target system 110 involves simulating the target system 110 as the system 110 executes the instructions 108. However, relatively infrequent other events also occur. Such events include peripheral device events, such as accessing a network adapter, accessing a hard disk drive, and so on. These peripheral device events are relatively infrequent in that their number is much smaller than the number of instructions executed by the target system 110. However, such peripheral device events may nevertheless affect the functional simulation of the target system 110, in that processing such events can take considerably longer than executing processor instructions. For example, waiting for a hard disk drive access can take milliseconds, whereas a processor instruction can be executed in microseconds.
Therefore, upon the functional simulation being performed by the functional simulation component 102 encounter such a peripheral device event, the functional simulation component 102 calls the timing simulation component 104 to provide a corresponding time delay (214). This time delay is used by the functional simulation component 102 in performing functional simulation of the target system 110, in addition to the other timing simulation information 114 regularly and periodically provided by the timing simulation component 104. For example, the time delay for a hard disk drive write access can be propagated to the model simulating such a hard disk drive, to make the simulated disk drive unavailable for an appropriate duration of time.
Ultimately, the functional simulation that takes into account the timing simulation is completed, at which time the results of the simulation are output (216). The simulation results may be stored in an electronic file, for instance, for later review and analysis by a developer or designer of the target system 110 or of the application encompassing the instructions 108. The simulation results may further be displayed on a display device, or printed as hardcopy by a printing device.
The timing simulation predicts at a given sample point the instructions per cycle to be used during the functional simulation within the next time period. Thus, at sample point 302A, the timing simulation predicts the instructions per cycle to be used during the functional simulation within the time period 304A. At sample point 302B, the timing simulation predicts the instructions per cycle to be used during the functional simulation within the time period 304B. At sample point 302C, the timing simulation predicts the instructions per cycle to be used during the functional simulation within the time period 304C.
In the example of
In conclusion,
Functional simulation of a target system executing instructions is performed (402). As has been described, such functional simulation includes simulating how the target system in question operates, as opposed to how long it takes for the target system to execution the instructions in question. The functional simulation of the target system is performed at least substantially continuously.
Periodically, timing simulation of the target system executing instructions is performed (404). As has also been described, such timing simulation includes simulating how much time, or how long, it takes for the target system to execute a given portion of the instructions (i.e., perform a given task). The functional simulation of the target system takes into account the periodic timing simulation that is performed. For instance, the timing simulation outputs timing simulation information at the periodic intervals at which it is performed, where this timing simulation information is used by the functional simulation to guide such functional simulation of the target system.
As one example, as has been described in relation to
Thus, the timing simulation simulates how many instructions the target system is to execute during the next period of time, where this information is used by the functional simulation when functionally simulating the system in this next period of time. The instructions-per-cycle value, or other timing simulation information, is predicted, such as based on historical data regarding the numbers of instructions per cycle previously predicted for prior periods of time, and/or on functional simulation information output during functional simulation. Once the simulation is finished, the results of the simulation are output (406), including the results of the timing and/or functional simulation.
