3com Corporation

Eoin O'Brien

Daniel M O'Keeffe

Neil J Clifford

Eugene O'Neill

Edele O'Malley

Peter Furlong

Eoghan Stack

Annette E Sohor

Derry M O'Donoghue

A content addressable memory stores entries each comprising a rule and as part of the entry a mask identifying all the entities to which the rule is applicable. A search pattern of data and a bit mask identifying the actual entity (or entities) associated with the data is applied as a search word along with a comparison mask that excludes all the other entities from the comparison of the search word with the entry. The CAM can thereby store efficiently in a single entry a rule that may be applicable to some but not all of a multiplicity of entities such as possible ingress ports of a network unit.

Content addressable memory organized to share entries between different entities such as ports of a network unit

In a stack of multi-port network communication units each unit has a forwarding database, the units are connected by way of a cascade, and at least some of the units are connected to links constituting a trunk. When a unicast data packet is received at a first of said units and the unicast data packet has a destination address which is not the subject of an entry in the forwarding database of the first unit, the unicast data packet is sent by way of the cascade to the other units in the stack, accompanied by a flag. When a second unit has in its forwarding database an entry, associating the destination address with forwarding data, it sends a management packet indicating said destination address and the identity of said second unit, so that the database of the first unit can be immediately updated.

Method for synchronizing databases in stacked network units

A network unit includes a forwarding database which includes entries relating packet addresses to forwarding data and is operative to form linked chains of entries wherein the entries in a chain represent different packet addresses which are reduced by hashing to the same pointer for the database. A seed for a hashing algorithm is provided by a combination, such as an exclusive-OR combination, of part of an input address word and a selected word. Hashing is preferably performed in at least two cycles, wherein in one cycle a hash function is applied to a portion of the input address not including the part which is combined with the selected word to constitute a hash seed and in another (preceding) cycle a further part of the address and the aforementioned hash seed are subject the hash function. The selected word may be a software programmable mask and it is possible thereby to ensure a high degree of randomness in the hash seed and to reduce the incidence of convergence of different addresses to a common pointer.

Network unit including address hashing

Growing community of inventors

Dublin, Ireland

Eoin O'Brien