Communications of the ACM
The performance of message-passing using restricted virtual memory remapping
Software—Practice & Experience
A family of routing and communication chips based on the Mosaic
Proceedings of the 1993 symposium on Research on integrated systems
Fbufs: a high-bandwidth cross-domain transfer facility
SOSP '93 Proceedings of the fourteenth ACM symposium on Operating systems principles
The use of message-based multicomputer components to construct gigabit networks
ACM SIGCOMM Computer Communication Review
High performance messaging on workstations: Illinois fast messages (FM) for Myrinet
Supercomputing '95 Proceedings of the 1995 ACM/IEEE conference on Supercomputing
Computer architecture (2nd ed.): a quantitative approach
Computer architecture (2nd ed.): a quantitative approach
ACM Computing Surveys (CSUR)
FLIPC: a low latency messaging system for distributed real time environments
ATEC '96 Proceedings of the 1996 annual conference on USENIX Annual Technical Conference
VISA: Netstation's virtual Internet SCSI adapter
Proceedings of the eighth international conference on Architectural support for programming languages and operating systems
Application provided checksums
ICCC '02 Proceedings of the 15th international conference on Computer communication
Hi-index | 0.01 |
This paper describes a "zero-scan" mechanism that reduces Internet checksumming overhead from a per-byte scan (or copy) cost, to a small and constant per-message cost. Unlike previous techniques, this mechanism requires no message buffering within the source. This will allow Internet transport protocols to achieve transfer latencies comparable to specialized protocols implemented directly on high-speed LAN (link-layer) services. In addition, this mechanism is transparent to systems outside of the source LAN. Hence, this mechanism affords applications the portability of Internet protocols without sacrificing the high performance of specialized LAN transport protocols.The proposed zero-scan checksumming scheme eliminates the last requirement for an additional data copy/scan, beyond the scan required to transmit or receive from the network channel. If this checksumming mechanism is combined with zero-copy operating system mechanisms that provide low-overhead transfer across application and kernel boundaries, a network interface architecture that provides separate message buffering is no longer required. A consequence is that the network interface may be reduced, essentially, to DMA engines plus link- and physical-layer logic. Taken one step further, the network interface could be integrated with the CPU to create an "internet microprocessor". These alternative interface designs are discussed, along with their requirements and effects upon operating system and computer system architectures.