Reducing memory fragmentation in network applications with dynamic memory allocators optimized for performance

Abstract

The needs for run-time data storage in modern wired and wireless network applications are increasing. Additionally, the nature of these applications is very dynamic, resulting in heavy reliance on dynamic memory allocation. The most significant problem in dynamic memory allocation is fragmentation, which can cause the system to run out of memory and crash, if it is left unchecked. The available dynamic memory allocation solutions are provided by the real-time Operating Systems used in embedded or general-purpose systems. These state-of-the-art dynamic memory allocators are designed to satisfy the run-time memory requests of a wide range of applications. Contrary to most applications, network applications need to allocate too many different memory sizes (e.g., hundreds different sizes for packets) and have an extremely dynamic allocation and de-allocation behavior (e.g., unpredictable web-browsing activity). Therefore, the performance and the de-fragmentation efficiency of these allocators is limited. In this paper, we analyze all the important issues of fragmentation and the ways to reduce it in network applications, while keeping the performance of the dynamic memory allocator unaffected or even improving it. We propose highly customized dynamic memory allocators, which can be configured for specific network needs. We assess the effectiveness of the proposed approach in three representative real-life case studies of wired and wireless network applications. Finally, we show very significant reduction in memory fragmentation and increase in performance compared to state-of-the-art dynamic memory allocators utilized by real-time Operating Systems.

Statistical analysis of blocking probability and fragmentation based on Markov modeling of elastic spectrum allocation on fiber link

“…Frag int (in [32] referred to as internal fragmentation) considers the number of remaining slots in the idle spectrum interval of size m ͠ slots, which a request of n req slots would be allocated in, and cannot be used for any future request no matter the amount of resources demanded. The remaining slots in such empty block are even smaller than the smallest request size and therefore are considered wasted.…”

“…The remaining slots in such empty block are even smaller than the smallest request size and therefore are considered wasted. Frag ext (in [32] referred to as external fragmentation) accounts for the idle spectrum intervals that cannot be used for the actual request of n req slots and are considered holes in between the otherwise allocated blocks. To calculate Frag ext in (5), the sum is over the number of slots belonging to all the unallocated blocks of slots, whose sizes are smaller than a required number of slots n req for the incoming request.…”

Dynamic Memory Management for Embedded Systems

“…A lot of research has been performed in the optimization of dynamic memory allocation [14,107]. And in that research, the actual behavior of the application in terms of allocation calls is relevant since the aim of these optimizations is to map data that is frequently allocated in memories closer to the processor.…”