Preemptive retransmission of buffered data in a network

Abstract

An apparatus for data transmission over a network includes a buffer and a timer mechanism for timing an optimum interval (less than the round-trip error response delay) for retransmission of data. A first accessor accesses data in the buffer for transmission and starts a first timeout clock. Second or further accessors of the buffer are responsive to a timeout to access the data, start a timeout clock and attempt to transmit the data on a path avoiding path elements used by prior accessors. A counter increments and decrements a count of the references to the buffer by accessors and signals when the count reaches zero. A memory manager returns the buffer to a free buffer pool responsive to the reference counter signaling that the count has reached zero. An analysis mechanism can be used to determine the optimum interval and tune the timer mechanism.

Description

technical field [0001] This invention relates to the field of distributed networked fault-tolerant systems, and more particularly, to those systems in which speed and reliability of data transmission are important. Background technique [0002] Distributed systems are becoming an increasingly common means of providing a platform for a variety of functions, including storage controller type functions. Their ubiquity arises from the flexibility and scalability such systems provide. Fault tolerance is achieved in a number of mutually supportive ways, such as providing redundant network infrastructure or redundant storage attachments. Distributed applications rely on network connectivity and communication capabilities to perform their functions. These fault-tolerant features improve the availability of the system. High availability is also increasingly important for many applications. [0003] Many systems implement retry algorithms for their network interfaces that operate ...

AI Technical Summary

Problems solved by technology

[0008] 5. The sender may use alternate hardware to try the request a second time

The private copy can be accessed without being blocked by the virtual memory manager, but this scheme will increase the cost of each send, which includes the majority of the ones that do not encounter the problem, because of the need for each send before the first send attempt copy data

Such additional processing costs may prove unacceptable in most modern networks

Images