{"title":"Ultra low delay switching for networked music performance","authors":"G. Baltas, G. Xylomenos","doi":"10.1109/IISA.2014.6878798","DOIUrl":null,"url":null,"abstract":"Low latency is essential for videoconferencing applications such as presence and collaboration between remote participants. In modern videoconferencing systems, the Selective Forwarding Unit (SFU) has the role of transparently duplicating and forwarding media streams between participants, hence it must be able to process large volumes of incoming packets at very high rates. SFU performance is heavily affected by the network I/O mechanisms employed to move packets from the Network Interface Card (NIC) to user space, and then move the copies back to the NIC. Traditional mechanisms, such as POSIX sockets, are not designed for high performance networking and prove to be a major bottleneck in such scenarios, by increasing packet latency and undermining the SFU's scalability. In this paper, we present a novel SFU platform which was designed to handle the ultra-low latency requirements of Networked Music Performance (NMP) applications, that is, the collaboration of musicians in real time. We implement a prototype SFU based on POSIX sockets and outline its performance bottlenecks. To overcome them, we turn to the netmap framework for fast packet I/O, which provides direct but safe access to the NIC buffers. We argue that ultralow latency videoconferencing is a natural application for netmap and thus design and implement a netmap-based SFU.","PeriodicalId":298835,"journal":{"name":"IISA 2014, The 5th International Conference on Information, Intelligence, Systems and Applications","volume":"15 1","pages":"0"},"PeriodicalIF":0.0000,"publicationDate":"2014-07-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"11","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"IISA 2014, The 5th International Conference on Information, Intelligence, Systems and Applications","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1109/IISA.2014.6878798","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 11
Abstract
Low latency is essential for videoconferencing applications such as presence and collaboration between remote participants. In modern videoconferencing systems, the Selective Forwarding Unit (SFU) has the role of transparently duplicating and forwarding media streams between participants, hence it must be able to process large volumes of incoming packets at very high rates. SFU performance is heavily affected by the network I/O mechanisms employed to move packets from the Network Interface Card (NIC) to user space, and then move the copies back to the NIC. Traditional mechanisms, such as POSIX sockets, are not designed for high performance networking and prove to be a major bottleneck in such scenarios, by increasing packet latency and undermining the SFU's scalability. In this paper, we present a novel SFU platform which was designed to handle the ultra-low latency requirements of Networked Music Performance (NMP) applications, that is, the collaboration of musicians in real time. We implement a prototype SFU based on POSIX sockets and outline its performance bottlenecks. To overcome them, we turn to the netmap framework for fast packet I/O, which provides direct but safe access to the NIC buffers. We argue that ultralow latency videoconferencing is a natural application for netmap and thus design and implement a netmap-based SFU.