Intel Corporation

Anguo Tony Huang

Juan-Carlos Calderon

Jing Ling

Jean-Michel Caia

Vivek Joshi

Simon S Chong

Man Dieu Trinh

Ryszard Bleszynski

David A Stelliga

Steve John Clohset

Systems and methods for queuing and de-queuing packets in a two-dimensional link list data structure. A network processor processes data for transmission for a plurality of Virtual Connections (VCs). The processor creates a two-dimensional link list data structure for each VC. The data field of each data packet is stored in one or more buffer memories. Each buffer memory has an associated buffer descriptor that includes a pointer to the location of the buffer memory, and a pointer pointing to the memory of the next buffer descriptor associated with a buffer memory storing data for the same packet. Each data packet also has an associated packet descriptor including a pointer pointing to the memory location of the first buffer descriptor associated with that packet, and a pointer pointing to the memory location of the packet descriptor associated with the next data packet queued for transmission. A VC descriptor for each VC keeps track of the memory locations of the next packet descriptor and the next buffer descriptor to be de-queued, and the memory locations for storing the next packet descriptors and the next buffer descriptors to be queued.

Two-dimensional queuing/de-queuing methods and systems for implementing the same

A method and apparatus for providing a virtual output queue (VoQ) from a received set of data packets in a multi-service system. Each packet is divided into at least one partition, including a last partition that includes packet information, such as error status and packet length. The system receives the packet from a flow, parses the packet into partitions, including a first partition and the last partition, places each last partition into a linked list based on a time when the last partition was received, links the last partition to the first partition, and employs the linked list as the output queue. This system allows for rapid compilation and transmission of different sized packets, and obviates the need for the receiving processor to wait for the last partition to discard a bad packet.

Virtual output queue (VoQ) management method and apparatus

A rate policing algorithm for packet flows is based on counters and threshold checking. The rate policing algorithm utilizes a state machine having four links: (1) compliant state to compliant state; (2) transition from compliant state to non-compliant state; (3) non-compliant state to non-compliant state; and (4) transition from non-compliant state to compliant state. Depending on the values obtained from the counters and utilizing the threshold values, it is determined whether a flow rate for packets is compliant or non-compliant.

Rate policing algorithm for packet flows

Memory access efficiency for packet applications may be improved by transferring full partitions of data. The number of full partitions written to external memory may be increased by temporarily storing packets using on-chip memory that is on a chip with the processor. Before writing packets to external memory, packets of length smaller than the external memory partition size may be temporarily stored in the on-chip memory until an amount corresponding to a full or nearly full partition has been collected, at which point the data can be efficiently written to an external memory partition.

Increasing memory access efficiency for packet applications

The rate-based scheduling for a network application is used to control the bandwidth available to a flow while scheduling the transmission of the flow. The rate-based scheduling uses rate credits to represent the amount of data a flow is permitted to transmit and only permits a flow to transmit if the flow has rate credit available. A flow is permitted to transmit only if the peak packet rate for the scheduler has not been exceeded.

Rate-based scheduling for packet applications

Interleaving memory access includes enabling data included in a receive flow of data to be stored in a first memory bank, enabling data included in a transmit flow of data to be stored in a second memory bank, and alternating access of data in the first memory bank with access of data in the second memory bank.

Interleaving memory access

A memory is divided into a number of partitions. The partitions are grouped into a first group of partitions and a second group of partitions. When required by a port, a partition is assigned to the port from a pool of un-assigned partitions. The pool of un-assigned partitions comprises of un-assigned partitions from the first group of partitions and un-assigned partitions from the second group of partitions. The un-assigned partitions from the first group of partitions are assigned to the port until a first threshold is reached. The un-assigned partitions from the second group of partitions are assigned to the port after the first threshold is reached. A second threshold is used to limit a total number of partitions assigned to the port.

Dynamic memory allocation for assigning partitions to a logical port from two groups of un-assigned partitions based on two threshold values

Systems and methods for assisting multiple protocol conversion in a network accelerator. A network device includes a transmit processing engine, a receive processing engine and one or more memories, each memory including one or more buffers for storing packets. When a packet is received, the receive engine adds a 4, 8, 12 or 16-byte tag to the front of the packet on a per-VC basis and stores the packet to a buffer. Additionally, the receive engine is able to add an offset to the starting address of the packet in the buffer relative to the beginning of the buffer. When a packet is to be transmitted, the transmit engine is able to transmit the packet from an address that is offset from the starting address of the buffer by one or more bytes. Additionally, the transmit engine is able to add one of several predefined packet headers on a per-packet basis.

Multi-protocol conversion assistance method and system for a network accelerator

A CBR/VBR traffic scheduler includes multiple CBR/VBR shapers to shape traffic over a wide range of peak cell rates for multiple CBR and VBR connection. Each shaper points to one or more VCs in a link list and includes a PCR counter initialized to a first value, an SCR counter initialized to a second and an arbitration counter. Each shaper is also connected to one of several clock sources, each having an associated clock cycle. A priority encoder, coupled to each arbitration counter, provides for determining priority between shapers having one or more associated VCs ready for transmission. Both the PCR counter and the SCR counter for each shaper is decremented during each associated clock cycle. For each shaper, when the PCR counter is decremented to a value of zero, the arbitration counter is initialized to a preset value and enabled for selection by the priority encoder. For each shaper, when the SCR counter is decremented to zero, a credit count parameter associated with each of the VCs in the associated link list is incremented by a predetermined value. The shaper having the lowest arbitration count value is selected by the priority encoder for cell transmission. When a shaper is selected for cell transmission, the associated link list of VCs is walked through in order. Only VCs that have a credit count parameter above a threshold value are transmitted.

CBR/VBR traffic scheduler

Systems and methods for storing, or caching, VC descriptors on a single-chip network processor to enhance system performance. The single-chip network processor includes an on-chip cache memory that stores VC descriptors for fast retrieval. When a VC descriptor is to be retrieved, a processing engine sends a VC descriptor identifier to a content-addressable memory (CAM), which stores VC descriptor identifiers in association with addresses in the cache where associated VC descriptors are stored. If the desired VC descriptor is stored in the cache, the CAM returns the associated address to the processing engine and the processing engine retrieves the VC descriptor from the cache memory. If the VC descriptor is not stored in the cache, the CAM returns a miss signal to the processing engine, and the processing engine retrieves the VC descriptor from an off-chip memory. In this manner, VC descriptors associated with high bandwidth VCs are stored to the cache and retrieved much quicker from the cache than from the off-chip memory.

Growing community of inventors

Mountain View, CA, United States of America

Anguo Tony Huang