In order to design parallel applications that exploit efficiently modern supercomputers with their complex architecture, it is required to apply a variety of programming models and techniques. The development of such applications as well as their tuning to achieve the optimal performance is practically impossible without performance analysis tools. In this paper, we present a scalable binary trace library -- Lightweight Trace Library (LiTL). LiTL aims at providing performance analysis tools with a low-overhead event recording service. In order to enhance the scalability and the performance of such tools, we implement various optimization strategies and techniques in LiTL. The experiments on a suite of synthetic and standard computation kernels show that the overhead caused by LiTL in conjunction with EZTrace remains low. Furthermore, LiTL often performs better than the existing event recording libraries, saving up to 88% of the CPU time and up to 68% of the disk space. Finally, LiTLis a generic library that can be used in conjunction with a wide range of performance analysis tools.
