{"title":"灵活的FPGA-to-FPGA通信系统","authors":"Wu An, Xi Jin, Xueliang Du, Shuaizhi Guo","doi":"10.1109/ICACT.2016.7423482","DOIUrl":null,"url":null,"abstract":"In high-performance computing systems, each computing node communicates via a high-speed serial bus to ensure sufficient data transfer bandwidth. However, each computing node of different bus protocols is very difficult to communicate directly, which is not conducive to the extensibility of HPC (High performance computing) clusters. In this paper, we propose UPI, a inter-node communication interface based on FPGA, which can transmit different bus protocols (PCIe protocol and Ethernet protocol) simultaneously. More importantly, many different bus-supported computing nodes can be connected to the same HPC system. We implemented our UPI system on \"Gemini\" prototype verification board with two Xilinx Virtex-6 FPGAs. The results show that the transmission speed of the UPI can reach 11.04Gpbs (PCIe Gen2 X4) and 4.32Gpbs (Gigabit Ethernet) when DMA payload sizes is greater than 260KB and 80KB, respectively.","PeriodicalId":125854,"journal":{"name":"2016 18th International Conference on Advanced Communication Technology (ICACT)","volume":"17 1","pages":"0"},"PeriodicalIF":0.0000,"publicationDate":"2016-03-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"8","resultStr":"{\"title\":\"A flexible FPGA-to-FPGA communication system\",\"authors\":\"Wu An, Xi Jin, Xueliang Du, Shuaizhi Guo\",\"doi\":\"10.1109/ICACT.2016.7423482\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"In high-performance computing systems, each computing node communicates via a high-speed serial bus to ensure sufficient data transfer bandwidth. However, each computing node of different bus protocols is very difficult to communicate directly, which is not conducive to the extensibility of HPC (High performance computing) clusters. In this paper, we propose UPI, a inter-node communication interface based on FPGA, which can transmit different bus protocols (PCIe protocol and Ethernet protocol) simultaneously. More importantly, many different bus-supported computing nodes can be connected to the same HPC system. We implemented our UPI system on \\\"Gemini\\\" prototype verification board with two Xilinx Virtex-6 FPGAs. The results show that the transmission speed of the UPI can reach 11.04Gpbs (PCIe Gen2 X4) and 4.32Gpbs (Gigabit Ethernet) when DMA payload sizes is greater than 260KB and 80KB, respectively.\",\"PeriodicalId\":125854,\"journal\":{\"name\":\"2016 18th International Conference on Advanced Communication Technology (ICACT)\",\"volume\":\"17 1\",\"pages\":\"0\"},\"PeriodicalIF\":0.0000,\"publicationDate\":\"2016-03-03\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"8\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"2016 18th International Conference on Advanced Communication Technology (ICACT)\",\"FirstCategoryId\":\"1085\",\"ListUrlMain\":\"https://doi.org/10.1109/ICACT.2016.7423482\",\"RegionNum\":0,\"RegionCategory\":null,\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"\",\"JCRName\":\"\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"2016 18th International Conference on Advanced Communication Technology (ICACT)","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1109/ICACT.2016.7423482","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
In high-performance computing systems, each computing node communicates via a high-speed serial bus to ensure sufficient data transfer bandwidth. However, each computing node of different bus protocols is very difficult to communicate directly, which is not conducive to the extensibility of HPC (High performance computing) clusters. In this paper, we propose UPI, a inter-node communication interface based on FPGA, which can transmit different bus protocols (PCIe protocol and Ethernet protocol) simultaneously. More importantly, many different bus-supported computing nodes can be connected to the same HPC system. We implemented our UPI system on "Gemini" prototype verification board with two Xilinx Virtex-6 FPGAs. The results show that the transmission speed of the UPI can reach 11.04Gpbs (PCIe Gen2 X4) and 4.32Gpbs (Gigabit Ethernet) when DMA payload sizes is greater than 260KB and 80KB, respectively.