The computing models for high-energy physics experiments are becoming ever more globally distributed and grid-based, both for technical reasons (e.g., to place computational and data resources near each other and the demand) and for strategic reasons (e.g., to leverage equipment investments). To support such computing models, the network and end systems, computing and storage, face unprecedented challenges. One of the biggest challenges is to transfer scientific data sets -now in the multi-petabyte (10 15 bytes) range and expected to grow to exabytes within a decade -reliably and efficiently among facilities and computation centers scattered around the world. Both the network and end systems should be able to provide the capabilities to support high bandwidth, sustained, end-toend data transmission. Recent trends in technology are showing that although the raw transmission speeds used in networks are increasing rapidly, the rate of advancement of microprocessor technology has slowed down. Therefore, network protocol-processing overheads have risen sharply in comparison with the time spent in packet transmission, resulting in degraded throughput for networked applications. More and more, it is the network end system, instead of the network, that is responsible for degraded performance of network applications. In this paper, the Linux system's packet receive process is studied from NIC to application. We develop a mathematical model to characterize the Linux packet receiving process. Key factors that affect Linux systems' network performance are analyzed.

No abstract

Data transfer is now an essential function for science discoveries, particularly within big data environments. To support data transfer for big data science, there is a need for high performance, scalable, end-to-end, and programmable networks that enable science applications to use the network most efficiently. The existing network paradigm that support big data science consists of three major components: terabit networks that provide high network bandwidths, Data Transfer Nodes (DTNs) and Science DMZ architecture that bypasses the performance hotspots in typical campus networks, and on-demand secure circuits/paths reservation systems, such as ESNet OSCARS and Internet2 AL2S, which provides automated, guaranteed bandwidth service in WAN. This network paradigm has proven to be very successful. However, to reach its full potentials, we claim that existing network paradigm for big data science must address three major problems: the last mile problem, the scalability problem, and the programmability problem. To address these problems, we proposed a solution called AmoebaNet. AmoebaNet applies Software Defined Networking (SDN) technology to provide "QoS-guaranteed" network services in campus or local area networks. AmoebaNet complements existing network paradigm for big data science: it allows application to program networks at run-time for optimum performance; and, in conjunction with WAN circuits/paths reservation system such as ESNet OSCARS and Internet2 AL2S; it solves the last mile problem and the scalability problem. • Programmability. This feature enables science applications to program networks at run-time to suit their needs. A powerful and rich

No abstract

TCP performs poorly in networks with serious packet reordering. Processing reordered packets in the TCP layer is costly and inefficient, involving interaction of the sender and receiver. Motivated by the interrupt coalescing mechanism that delivers packets upward for protocol processing in blocks, we propose a new strategy, Sorting Reordered Packets with Interrupt Coalescing (SRPIC), to reduce packet reordering in the receiver. SRPIC works in the network device driver; it makes use of the interrupt coalescing mechanism to sort the reordered packets belonging to the same TCP stream in a block of packets before delivering them upward; each sorted block is internally ordered. Experiments have proven the effectiveness of SRPIC against forward-path reordering