Accelerated SMPP decoder implementation based on GPU

Abstract

Graphic processing unit (GPU) provides a low-cost but powerful hardware platform for implementing massively parallel high performance systems. The capabilities of GPUs have been used to provide fast and low cost solutions in areas such as machine learning, complex simulations such as global warming and genetic engineering and network traffic processing. Our research is focused on using GPUs to accelerate the decoding process of the Short Message Peer to Peer (SMPP) protocol. SMPP protocol is used to exchange Short Messages between Short Message Service Centers (SMSCs) and TCP/IP based applications. From the point of view of a user, a SMS taking few seconds is acceptable and therefore a SMSC is mostly focused on achieving a higher throughput than a low per packet latency. We have developed a SMPP decoder library in C with GPU support. It supports both CPU based and GPU based decoding. The library also includes two primary APIs. The first API is for general usage by any C based application and the second API could be used by Java Native Interface (JNI) based application. We have evaluated the performance of the library in both CPU and GPU modes and compared it with a SMPP server based on Cloudhopper, a Java implementation of SMPP protocol. The evaluation shows around five times throughput gain in GPU mode over both Java and C based CPU modes.