OTL: On-Demand Thread Stack Allocation Scheme for Real-Time Sensor Operating Systems

Abstract

Abstract. In wireless sensor networks, each sensor node has severe resource constraints in terms of energy, computing device, and memory space. Especially, the memory space of the platform hardware is much smaller than that of the other computing systems. In this paper, we propose a OTL, which is an on-demand thread stack allocation scheme for MMU-less real-time sensor operating systems. The OTL enables to adaptively adjust the stack size by allocating stack frame based on the amount of each function's stack usage. The amount of the function's stack usage is checked at compile-time, and the adaptive adjustment of the stack occurs at run-time. Our experimental results show that the OTL significantly minimizes the spatial overhead of the threads' stacks with tolerable time overhead compared with fixed stack allocation mechanism of the existing sensor operating systems.

A Robust and Space-Efficient Stack Management Method for Wireless Sensor Network OS with Scarce Hardware Resources

“…To allocate and release memory blocks, they use "malloc()" and "free()" APIs in the library. The works of [23,24] allocate a stack memory block as much as a function call requires. The basic concept for SESAME and OTL is similar to RESM.…”

“…In the resource constraint sensor nodes, a saturated condition can easily occur as the number of tasks rapidly increases due to bursts of targeted events. Although there is no available memory space, tasks in [23,24] continuously request additional stack memory at the entries of function calls in the saturated condition, and then system is halted. Since [23,24] do not propose any method to detect and handle the saturated conditions, the only way to resolve deadlock problem would be system reset.…”

“…Although there is no available memory space, tasks in [23,24] continuously request additional stack memory at the entries of function calls in the saturated condition, and then system is halted. Since [23,24] do not propose any method to detect and handle the saturated conditions, the only way to resolve deadlock problem would be system reset. Second, they propose only a concept for stack management per a function call.…”

“…Typically, the information of some MCU registers is pushed into the current stack to save the current context at a task switching or at a jump to ISR. Accordingly, an estimated stack space in [23,24] can easily overflow when an interrupt or a task switching occur during the execution of a subroutine. This is a critical problem in the multithreaded OS.…”

“…It seems to use such "malloc()" and "free()" as the library APIs in the compiler or the OS. The use of the library makes the performance, for example, the time to search empty memory blocks, and the space overhead to maintain the blocks, of [23,24] nondeterministic since it depends on the used library.…”