Abstract: A serial buffer includes a first port configured to implement an serial rapid I/O (sRIO) protocol and a second port configured to implement a Lite-weight serial (Lite) protocol. SRIO packets received on the first port are translated into Lite request packets compatible with the Lite protocol. The Lite request packets are transmitted to the second port. Lite response packets compatible with the Lite protocol are returned to the second port in response to the Lite request packets. The Lite response packets are translated into sRIO response packets compatible with the sRIO protocol. These sRIO response packets are returned to the first port, thereby providing a mechanism to acknowledge successful transmissions from the first port to the second port. Unsuccessful transmissions are identified by a timeout mechanism. The serial buffer also enables transfers from the second port to the first port in a similar manner.

Abstract: Within a serial buffer, request packets are written to available memory blocks of a memory buffer, which are identified by a free buffer pointer list. When a request packet is written to a memory block, the memory block is removed from the free buffer pointer list, and added to a used buffer pointer list. Memory blocks in the used buffer pointer list are read, thereby transmitting the associated request packets from the serial buffer. When a request packet is read from a memory block, the memory block is removed from the used buffer pointer list and added to a request buffer pointer list. If a corresponding response packet is received within a timeout period, the memory block is transferred from the request buffer pointer list to the free buffer pointer list. Otherwise, the memory block is transferred from the request buffer pointer list to the used buffer pointer list.

Abstract: A serial buffer monitors an incoming stream of packets to identify single missing packets and multiple consecutive missing packets. Upon detecting multiple consecutive missing packets, an interrupt is generated, thereby stopping the data transfer. Upon detecting a single missing packet, a single missing packet identifier is inserted into the packet header of the packet that resulted in identification of the single missing packet. The incoming packets, including any inserted single missing packet identifiers, are written to a queue. When the water level reaches the water mark of the queue, the stored packets are read to create an outgoing packet stream. When a packet read from the queue includes an inserted single missing packet identifier, a dummy packet (e.g., a packet having a data payload of all zeros) is inserted into the outgoing packet stream. As a result, real-time applications are capable of processing the outgoing packet stream in a constant fashion.

Abstract: A computer system comprises host processor and a network interface, wherein the host processor includes resources supporting a full power mode, a lower power mode and a power down mode, as seen in standard system bus specifications such as PCI and InfiniBand. The network interface includes a medium interface unit coupled to network media supporting a least high speed protocol, such as a Gigabit Ethernet or high-speed InfiniBand, and a lower speed protocol, such as one of 10 Mb and 100 Mb Ethernet or a lower speed InfiniBand. Power management circuitry forces the medium interface unit to the lower speed protocol in response to an event signaling entry of the lower power mode. In the lower power mode, the network interface consumes less than the specified power when executing the lower speed protocol, and consumes greater than the specified power when executing the high speed protocol.

Abstract: A serial buffer having a plurality of virtual queues, which can be allocated to include various combinations of on-chip dual-port memory blocks, on-chip internal memory blocks and/or off-chip external memory blocks. The virtual queues are allocated and accessed in response to configuration bits and size bits stored on the serial buffer. Relatively large external memory blocks can be allocated to virtual queues used for data intensive operations, while relatively small and fast dual-port memory blocks can advantageously be allocated to virtual queues used for passing command and status information. The serial buffer provides an efficient and flexible manner for utilizing available memory, which not only minimizes the access latency but also provides a large amount of buffer space to meet different application needs.

Abstract: A system and method for using a doorbell command to allow sRIO devices to operate as bus masters to retrieve data packets stored in a serial buffer, without requiring the SRIO devices to specify the sizes of the data packets. The serial buffer includes a plurality of queues that store data packets. A doorbell frame request packet identifies the queue to be accessed within the serial buffer, but does not specify the size of the data packet(s) to be retrieved. Upon detecting a doorbell frame request packet, the serial buffer operates as a bus master to transfer the requested data packets out of the selected queue. The selected queue can be configured to operate in a flush mode or a non-flush mode. The serial buffer may also indicate that a received doorbell frame request has attempted to access an empty queue.

Abstract: A serial buffer is configured to transmit a plurality of received data packets through a data packet transfer path to a host processor. A doorbell controller of the serial buffer monitors the number of data packets transmitted to the host processor through the data packet transfer path, and estimates the number of data packets actually received by the host processor. The doorbell controller generates a doorbell command each time that the estimated number of data packets corresponds with a fixed number of data packets in a frame. The doorbell commands are transmitted to the host processor on a doorbell command path, which is faster than the data packet transfer path. The doorbell controller may estimate the number of data packets actually received by the host processor in response to a first delay value, which represents how much faster the doorbell command path is than the data packet transfer path.

Abstract: A multi-port serial buffer having a plurality of queues is configured to include a first set of queues assigned to store write data associated with a first port, and a second set of queues assigned to store write data associated with a second port. The available queues are user-assignable to either the first set or the second set. Write operations to the first set of queues can be performed in parallel with write operations to the second programmable set of queues. In addition, a first predetermined set of queues is assigned to the first port for read operations, and a second predetermined set of queues is assigned to the second port for read operations. Data can be read from the first predetermined set of queues to the first port at the same time that data is read from the second predetermined set of queues to the second port.

Abstract: Status/error reporting is implemented using a doorbell system. A plurality of flag registers are included on a system device, such as a serial buffer. Each flag register has a corresponding address, and stores a plurality of flags. A flag scan controller accesses the flag registers in a predetermined priority order, using the flag register addresses. Upon detecting that one or more of the flags of a flag register are activated, the flag scan controller causes a doorbell command to be generated. The doorbell command includes the flag register address and the corresponding flags. A system processor receives the doorbell command and services the activated flags. Once the activated flags are serviced, the system processor performs one or more software write operations to clear the flags within the system device. The system processor can simultaneously service multiple flags. The system processor can also simultaneously clear multiple flags.

Abstract: A serial buffer includes queues configured to store data packets received from a host. A direct memory access (DMA) engine receives data packets from the highest priority queue having a water level that reaches a corresponding watermark. The DMA engine is configured in response to a DMA register set, which is selected from a plurality of DMA register sets. The DMA register set used to configure the DMA engine can be selected in response to information in the header of the read data packet, or in response to the queue from which the data packet is read. Each DMA register set defines a corresponding buffer in system memory, to which the data packet is transferred. Each DMA register set also defines whether the corresponding buffer is accessed in a wrap mode or a stop mode, and whether doorbell signals are generated in response to transfers to the last address in the corresponding buffer.

Abstract: On-chip resources of a serial buffer are accessed using priority packets of a Lite-weight protocol. A priority packet path is provided on the serial buffer to support priority packets. Normal data packets are processed on a normal data packet path, which operates in parallel with the priority packet path. The system resources of the serial buffer can be accessed in response to the priority packets, without blocking the flow of normal data packets. Thus, normal data packets may flow through the serial buffer with the maximum bandwidth supported by the serial interface. The Lite-weight protocol also supports read accesses to queues of the serial buffer (which reside on the normal data packet path). The Lite-weight protocol also supports doorbell commands for status/error reporting.

Abstract: A serial buffer having a parser and multiple parallel processing paths is provided. The parser receives incoming packets, determines the type of each packet, and then routes each packet to a processing path that corresponds with the determined packet type. Packet types may include blocking priority packets (which implement bus slave operations), non-blocking priority packets (which access on-chip resources of the serial buffer) and data packets (which implement bus master operations). Because the different packet types are processed on parallel processing paths, the processing of one packet type does not interfere with the processing of other packet types. As a result, blocking conditions within the serial buffer are minimized.

Abstract: A network interface comprises the first port on which incoming data is transmitted and received at the data transfer rate of the network, a buffer memory coupled to the first port, and a second port coupled with the buffer memory, and through which transfer of packets between the host system, and the buffer memory is executed. A driver in the host system allocates a plurality of sets of receive buffers, where each set of receive buffers is composed of receive buffers having different sizes. A receive buffer descriptor cache located at the interface level stores receive buffer descriptors corresponding to receive buffers in the plurality of sets. As incoming packets arrive at the interface, logic determines the size of the incoming packet and assigns the packet to a receive buffer descriptor in the receive buffer descriptor cache according to the determined size.

Abstract: A method and apparatus for performing computationally intensive calculations on a database server that provides a recovery mechanism is provided. These calculations involve a series of updates performed on data stored as data items in a database server. An existing recovery mechanism of the database server is used to store previous versions of the data items. The previous versions of the data items are subsequently retrieved from the recovery mechanism and used to determine whether the formula involved in the calculation should be modified, whether to continue the calculations with the current formula, or whether to terminate the calculations. By extending the functionality of the SELECT statement, the program can access the previous versions of the data items.

Abstract: A system which provides for generation of alert packets using network interfaces. The alert packets are downloaded into a network interface, for example during a transition from an OS-present state to an OS-absent state. The alert packets are provided with control fields which, in various combinations, indicate alert conditions upon which such packets are to be transmitted, and which indicate a repetition mode for transmission of the packet after a match of the alert condition. Logic in the network interface, causes scanning in the plurality of packets downloaded into the network interface at scan intervals to identify packets having control codes matching alert signals received by the network interface. The process allows for more than one alert packet to be matched with a single alert signal during a given scan interval, and to be transmitted.

Abstract: An architecture for a high performance IPSEC accelerator. The architecture includes components for scanning fields of packets, programming an IPSEC services device according to the scanned fields, and modifying the scanned packet with an output from the IPSEC security services device. Preferably, the architecture is implemented in hardware, and attached to a host machine. Hardware devices, fast in comparison to software processing and network speeds, allows the computationally intensive IPSEC processes to be completed in real-time and reduce or eliminate bottlenecks in the path of a packet being sent or received to/from a network.

Abstract: A plurality of virtual paths in a network interface between a host port and a network port are managed according to respective priorities. Thus, multiple levels of quality of service are supported through a single physical network port. Variant processes are applied for handling packets which have been downloaded to a network interface, prior to transmission onto the network. The network interface also includes memory used as a transmit buffer, that stores data packets received from the host computer on the first port, and provides data to the second port for transmission on the network. A control circuit in the network interface manages the memory as a plurality of first-in-first-out FIFO queues having respective priorities. Logic places a packet received from the host processor into one of the plurality of FIFO queues according to a quality of service parameter associated with the packets. Logic transmits the packets in the plurality of FIFO queues according to respective priorities.

Abstract: A hardware packet accelerator parses incoming packets to retrieve header data for building a frame status and for verifying the incoming packets are part of an established connection with a host. The accelerator includes a connection database that allows retrieval of connection information based on an index constructed from a hashed TCP connection address. The frame status comprises information needed to perform packet re-assembly and is stored in a memory that is local (directly accessible) by a processing device that performs the packet re-assembly. Among other advantages, the processing device does not need to read packet header data from a packet buffer, saving large amounts of header data retrieval time.

Abstract: Autonomous retransmission of data packets onto a network from a Network Interface Card level upon command from a host processor is support. Efficient FIFO buffering in an ASIC is retained. Uses for autonomous retransmission include hardware and software testing and in network management.

Abstract: In one embodiment of a networking module, a first block receives a serial digital media signal, and provides a parallel digital media signal based on the serial digital media signal. A second block, operative with the first block, stores the parallel digital media signal in a corresponding slot in an outgoing frame, and sends the outgoing frame in response to receiving an incoming frame.