Latency analysis of UDP and BPI on Myrinet
Hyun-Wook Jin, Chuck Yoo
1999 IEEE International Performance, Computing and Communications Conference (Cat. No.99CH36305)
High-speed networks such as ATM, Myrinet, and Gigabit Ethernet are available today, and many researchers make efforts to enhance the performance of end-to-end communication on these high-speed networks. One of the efforts is to develop new light-weight communication primitives for high-speed networks. However; the latency of the new primitives has not been characterized thoroughly, partly because existing measurement methodologies do not take into account the features of high-speed networks.
... refore, there are only incomplete comparisons of the new primitives and traditional protocols, and they cannot really prove the usefulness of new primitives. In order to address this issue, this paper suggests a new measurement methodology and uses the methodology to perform a detailed latency analysis of UDP and a light-weight primitive, called BPI, on Myrinet. Our results clearly show the difference of perbyte overhead between BPI and UDR A surprising result is that BPI is found to be slower than UDP for 4KB or larger data size. data in modern systems (e.g., multimedia systems) is getting larger than that of traditional systems. The second is the development of new light-weight communication primitives. Most of the new primitives, such as FM (Fast Messages) , U-Net , and Myrinet Software [ 5 ] , bypass the kernel in order to reduce communication overhead. The new light-weight communication primitives are less general and portable than traditional protocols but these have the potential to achieve the best performance on a specific network hardware. Although light-weight communication primitives are attractive, these primitives need to prove their usefulness and be compared thoroughly with traditional protocols. The goal of this paper is to analyze the latency of a light-weight communication primitive in details and to compare it with a traditional protocol. The latency measurement is performed on Myrinet because it is the fastest LAN available today. The measured light-weight communication primitive is BPI [SI that is designed to support high-speed communication on Myrinet. Because BPI does not support reliable delivery, it is compared with UDP rather than TCP. In addition, this paper suggests a new measurement methodology for the latency analysis of high-speed networks. The measurement is performed at the firmware level of NIC (Network Interface Card), which is more advanced than the existing measurements that are performed at the device driver level. As a result, this paper shows what factors influence the performance of a light-weight communication primitive and what should be considered in designing a new light-weight communication primitive. The rest of this paper is organized as follows. Section 2 describes related works. Section 3 introduces a new measurement methodology to measure the latency on high-speed networks. Section 4 explains Myrinet Software. Section 5 presents experiment environment in details. Sections 6 and 7 show the measurement results and the comparison of BPI and UDP. Finally, the paper concludes in Section 8.