Cisco Technology, Inc.

Maxim S Martynov

Qiwen Zhang

Gideon Glass

Etai Lev Ran

Daniel Kaminsky

Etai Lev-Ran

Dan Li

Hamid R Amirazizi

A method and apparatus is disclosed for balancing throughput and compression in a network communication system. A message is received. A first compression mechanism is applied to the message to generate a first compressed message, where applying the first compression mechanism comprises gathering compression information. Based on the compression information, a compression metric associated with the first compressed message is computed. The compression metric is compared to a threshold value. A second compression mechanism is applied to the first compressed message to generate a second compressed message only when the compression metric does not exceed the threshold value. The second compressed message is then transmitted.

Techniques for balancing throughput and compression in a network communication system

A data redundancy elimination system. In particular implementations, a method includes accessing a data block; dividing the data block into a plurality of data chunks; computing chunk signatures for the plurality of data chunks; identifying a plurality of anchor chunks in the plurality of data chunks; accessing a second memory to identify one or more matching chunk signatures; reading one or more signature blocks corresponding to the matching chunk signatures from a first memory into the signature block search space of the second memory; and matching one or more computed chunk signatures to chunk signatures of the one or more signature blocks in the signature block search space.

Data redundancy elimination mechanism including fast lookup of data patterns exhibiting spatial locality

A data redundancy elimination system. In particular implementations, a method includes storing in a memory one or more aggregation trees, each aggregation tree comprising one or more base chunk nodes and one or more super chunk nodes, wherein each base chunk node comprises a chunk signature and corresponding raw data, and wherein super chunk nodes correspond to child base chunk nodes and include a chunk signature; receiving a data block; dividing the data block into a plurality of base chunks, each base chunk having a degree value characterizing the occurrence probability of the base chunk; computing chunk signatures for the plurality of base chunks; applying a super chunk rule to contiguous sequences of base chunks of the plurality of base chunks to create one or more aggregation trees, wherein the super chunk rule aggregates base chunks based on the respective occurrence probabilities of the base chunks; identifying one or more nodes in the one or more created aggregation trees that match corresponding nodes of the aggregation trees in the memory; compressing the received data block based on the identified nodes; and conditionally adding the one or more created aggregation trees to the memory.

Redundancy elimination by aggregation of multiple chunks

Growing community of inventors

Los Gatos, CA, United States of America

Maxim S Martynov