| Satellite has been widely applied to the social, economic and cultural fields, attracting more and more attentions. However, the complexity of device and software has increasingly led to software bugs. Once the accident happens, great threat will be introduced into human lives, property safety and national security. Therefore, as an expensive cost and high worth applications, the reliability of satellite software must be guaranteed.Software testing is one of the necessary means of improving the reliability. Satellite software run on embedded systems, with specific processor, and they typically characterize with real-time, limited memory, poor I/O channels, code fixed in ROM. All these features have led testing to be difficult, and make the running software tracking even impossible, hampering the progress and quality improvement of testing. In other words, the traditional testing methods do not adapt to the satellite software testing any more.Technology of satellite software testing can be divided into several categories by different standards. According to whether the specific hard device taking part into testing, testing can be physical test, partial physical test and full digital simulation test; according to where the target data source is, test can be open loop test, close loop test. This paper chooses the full digital close test, taking advantage of the simulator to construct the full digital simulation test platform. Simulator implements the system environment of satellite software, especially the CPU, memory, registers and communication interface, enabling the test on ordinary PC for satellite software. Meanwhile, simulator intergrates the debug ability, including step in/forward, set breakpoints and disassemble code showing. All these improve the speed and quality of testing satellite software, and enhance testers'ability to find bugs.After the full digital simulation test platform by demand is set up, problems still exist. The parallel satellite system, including software having interaction with each other, still can't be tested. Several simulators need to run together. However, due to the simulating time of simulator differs with each other, the time synchronization strategy is needed to avoid the causality errors.There are some researchers studying on this, but none adapt to our application. Chengyuan Zhu implement the real-time simulation using dSPace, Jihe Wang studies the double satellite real time simulation. Both implement the real-time simulation taking advantage of the hardware. However, simulator's speed is slower than real running the software. So the real-time simulation is impossible for our application. Jifu Zhang studies the HLA, which need lookahead value achieve good result, to solve multiple planes simulation. However, the satellite software we run is unpredictable. Junli Yang's implement differs with us, she adapt the partical-physical simulation technology, which can't replay the scene where the software interacts with each other.We innovatively propose to solve the synchronization problem among multiple simulators using parallel simulation technology. We add controlling model of simulation, where design and implement the synchronization strategy. None feature have mentioned this solution until now.Synchronization algorithm for parallel simulation can be two kinds:conservative time synchronization algorithm and optimistic time synchronization algorithm. Conservative solution resolutely forbid the causality error(CE), only allowing the safe event to execute. On the other hand, optimistic synchronization algorithm allow the processor execute the events disregarding the CE, and rollback the events and rerun them when CE occurs. For lacking the lookahead, needed by conservative synchronization strategy to perform good result, the paper drops the conservative to use the optimistic algorithm.The paper first introduces a classical optimistic synchronization algorithm Time Warp, introducing the basic process and concepts of optimistic synchronization strategy. The concepts include rollback, anti-message, and GVT. And show how to re-execute the event to support the rollback of events:remove changes done by events to the variables and recall the events should not be executed. To achieve the former we need to save the sate before the event executing, and send anti-messages to achieve the following. So, we know, rollback needs price:state of preservation leads to storage cost; and recursive sending anti-messages leads to the "avalanche", and re-executing events leads simulation going forward slow.We, aiming at reducing the cost base on TW, first introduce several time synchronization strategies:Time Buckets (TB), Breathing Time Buckets (BTB) and improved TW. We analyze the advantage and disadvantage of these algorithms. We apply the Breathing Time Warp (BTW) to our application, proving BTW can correctly solve the problem of time synchronization. And the experiment results show that the simulation speed proves to be faster than both BTB and TW; the rollbacks number is smaller than both BTB and TW; the anti-messages number is smaller than TW. Though BTB avoid sending the anti-messages, it produces intolerable number of rollbacks. Above all, BTW is the ideal solution for solving the time synchronization problem among multiple simulators. |
PDF Full Text Request