BIP: A new protocol designed for high performance networking on Myrinet

Abstract

Abstract. High speed networks are now providing incredible performances. Software evolution is slow and the old protocol stacks are no longer adequate for these kind of communication speed. When bandwidth increases, the latency should decrease as muc:h in order to keep the system balance. With the current network technology, the main bottleneck is most of the time the software that makes the interface between the hardware and the user. We designed and implemented new protocols of transmission targeted to parallel computing that squeeze the most out of the high speed Myrinet network, without wasting time in system calls or memory copies, giving all the speed to the applications. This design is presented here as well as experimental results that lead to achieve real Gigabit/s throughput and less than 5/~s latency on a cluster of PC workstations, with this affordable network hardware. Moreover, our networking results compare favorably with the expensive parallel computers or ATM LANs.

Kernel Mode Linux: Toward an Operating System Protected by a Type Theory

“…To improve the performance further, we should modify the kernel and its interface and exploit the TAL type system more aggressively. For example, we think that a kernel can be modified to export network communication hardware to user programs as user-level communication technologies (e.g., [14,15,4]). The problem of the user-level communication is a tradeoff between performance and safety.…”

HARNESS fault tolerant MPI design, usage and performance issues

“…The handling of these possibly non-contiguous data types is very important in real applications, and is often a neglected area of communication library design [17]. Most communications libraries are designed for low latency and/or high bandwidth with contiguous blocks of data [14]. Although this means that they must avoid unnecessary memory copies, the efficient handling of recursive data structures is often left to simple iterations of a loop that packs a send/receive buffer.…”

“…This problem becomes even more acute when the target systems latency is so low that any buffering, while waiting for slower nodes, drastically changes performance characteristics as is the case with BIP-MPI [14] and SCI MPI [8].…”

Design and implementation of a user‐level Sockets layer over Virtual Interface Architecture

“…More specifically, the traditional communication architecture based on the TCP/IP protocol suite is reported to fail in delivering raw-hardware performance to end users, due to protocol overhead, context switching overhead, and data copying overhead between the user and the kernel space. As a result, a number of user-level communication architectures have been proposed that remove the operating system from the critical communication path [3][4][5][6].…”