摘要:
A load balancing mechanism maps a binary tree representation of a routing table into a set of fixed size memories. The mechanism efficiently utilizes the memory in the routing table without violating the tree precedence constraints and the memory access requirements of a pipelined system. The mechanism stores a subtree associated with a densely populated level of the binary tree in memory associated with lower levels.
摘要:
We present a lookup table which allows sparse subtree descriptors and dense subtree descriptors to be stored in the same memory. A subtree entry in the memory stores a dense subtree descriptor for a dense subtree or a plurality of sparse subtree descriptors for sparse subtrees. The subtree entry is indexed by a leaf in the previous subtree. The sparse subtree descriptor stores at least one node descriptor. The node descriptor describes a set of leaves in the sparse subtree having a common value. The common value is encoded in the node descriptor using run length encoding.
摘要:
We present a method and apparatus for increasing the depth of a lookup table. The lookup table includes at least two lookup units. A search request for a final result value corresponding to the key is issued to each of the lookup units in parallel. Each of the lookup units performs a multi-level search in parallel for the final result value. The final result value stored in only one of the lookup units is provided by one of the lookup units.
摘要:
A lookup table for searching for a longest prefix match for a key is disclosed. The lookup table provides a match for a key in a single search cycle. The number of matches stored in the lookup table is maximized by storing each match in only one location in the lookup table. The binary tree is divided into a plurality of levels and each level has a plurality of subtrees. A subtree descriptor stored for a subtree includes a field for each node in the subtree. The state of the field indicates whether an entry for the node is stored in the table. The bit vector allows indexing of the single match stored for the key.
摘要:
Management of congestion level, in a computer-related context, is disclosed. Also disclosed is a system generating a plurality of computer network-related tables during system operation. A number of the tables are each separately indexed by a different index. The system includes at least one tangible computer-readable medium adapted to store, at each indexed location, a swap count providing an indication of the congestion level of the indexed location. The system also includes insert logic stored as instructions on the at least one medium for execution. When executed, the insert logic is operable to: i) insert, when a predetermined condition has been satisfied, a new entry by overwriting the current entry stored in the indexed location having the lowest swap count; and ii) update the swap counts in each of the indexed locations in a manner that maintains the total swap count at least substantially constant over time.
摘要:
A circuit which accurately controls the word line (pass transistor gate) driving voltage to a voltage which is both controlled and is not significantly greater than is needed to drive the word line. The circuit eliminates the need for a double-boot-strapping circuit, and ensures that no voltages exceed that necessary to fully turn on a memory cell access transistor. Voltages in excess of that which would reduce reliability are avoided, and accurate driving voltages are obtained. A DRAM includes word lines, memory cells having enable inputs connected to the word lines, a gate receiving word line selecting signals at first logic levels Vss and Vdd, and for providing a select signal at levels Vss and Vdd, a high voltage supply source Vpp which is higher in voltage than Vdd, a circuit for translating the select signals at levels Vss and Vdd to levels Vss and Vpp and for applying it directly to the word lines whereby an above Vdd voltage level word line is achieved without the use of double boot-strap circuits.
摘要:
In a ring-of-clusters network topology, groups of slave devices are accessed in parallel, such that the latency around the ring is proportional to the number of clusters and not proportional to the number of integrated circuits. The devices of a cluster share input and output ring segments such that packets arriving on the input segment are received and interpreted by all the devices in a cluster. In other embodiments, none, some or all but one slaves per cluster are asleep or otherwise disabled so that they do not input and interpret incoming packets. Regardless, in all embodiments, the slaves of a cluster cooperate, potentially under the controller's direction, to ensure that at most one of them is actively driving the output segment at any given time. The devices may be addressed through a device ID, a cluster ID, or a combination thereof. Embodiments of the invention are suited to exploit multi-chip module implementations and forms of vertical circuit stacking.
摘要:
A delay locked loop includes initialization circuitry that ensures that a DLL is initialized to an operating point that is not to close to either end of a delay vs. control voltage characteristic. The initialization circuitry forces the DLL to initially search for a lock point starting from an initial delay, the delay is varied in one direction, forcing the DLL to skip the first lock point. The initialization circuitry only allows the DLL to vary the delay of the voltage controlled delay loop in the one direction from the initial delay until the operating point is reached.
摘要:
An initialization circuit in a delay locked loop ensures that after power up or other reset clock edges are received by a phase detector in the appropriate order for proper operation. After reset of the delay locked loop, the initialization circuit assures that at least one edge of a reference clock is received prior to enabling the phase detector to increase (or decrease) the delay in a delay line. After at least one edge of a feedback clock is received, the initialization circuit enables the phase detector to decrease (or increase) the delay in a delay line.
摘要:
A dynamic random access memory (DRAM) having pairs of bitlines, each pair being connected to a first bit line sense amplifier, wordlines crossing the bitlines pairs forming an array, charge storage cells connected to the bitlines, each having an enable input connected to a wordline, the bit line sense amplifiers being connected in a two dimensional array, pairs of primary databuses being connected through first access transistors to plural corresponding bit line sense amplifiers in each row of the array, apparatus for enabling columns of the first access transistors, databus sense amplifiers each connected to a corresponding data bus pair, a secondary databus, the secondary databus being connected through second access transistors to the databus sense amplifiers, and apparatus for enabling the second access transistors, whereby each the primary databus pair may be shared by plural sense amplifiers in a corresponding row of the array and the secondary databus may be shared by plural primary databus pairs.