Intel Corporation

Lsi Corporation

Joseph R Hasting

William G Burroughs

David Sonnier

Jerry Pirog

Te Khac Ma

Tsung-Yuan Charles Tai

Ren Wang

Andrew J Herdrich

Debra Bernstein

Jr-Shian Tsai

Stephen R Van Doren

Niall D McDonnell

Hugh Wilkinson

Deepak Mital

Yipeng Wang

Andrew Cunningham

Bruce Richardson

James T Clee

Narender R Vangati

Gage Eads

Jonathan Kenny

Stephen Miller

An Yan

Jamison D Whitesell

Edwin Verplanke

Bradley A Burres

Namakkal N Venkatesan

David Allen Brown

Steven J Pollock

Michael R Betker

A system and method are described for work conserving, load balancing, and scheduling by a network processor. For example, one embodiment of a system includes a plurality of processing cores, including a scheduling circuit, at least one source processing core that generates at least one task and at least one destination processing core that receives and processes the at least one task, and generates a response. The scheduling circuit of the exemplary system receives the at least one task and conducts a load balancing to select the at least one destination processing core. In an embodiment, the scheduling circuit further detects a critical sequences of tasks, schedules those tasks to be processed by a single destination processing core, and, upon completion of the critical sequence, conducts another load balancing to potentially select a different processing core to process more tasks.

Work conserving, load balancing, and scheduling

Apparatus and methods implementing a hardware queue management device for reducing inter-core data transfer overhead by offloading request management and data coherency tasks from the CPU cores. The apparatus include multi-core processors, a shared L3 or last-level cache ('LLC'), and a hardware queue management device to receive, store, and process inter-core data transfer requests. The hardware queue management device further comprises a resource management system to control the rate in which the cores may submit requests to reduce core stalls and dropped requests. Additionally, software instructions are introduced to optimize communication between the cores and the queue management device.

Multi-core communication acceleration using hardware queue device

Apparatus and method for multi-core dynamically-balanced task processing while maintaining task order in chip multiprocessor platforms. One embodiment of an apparatus includes: a distribution circuitry to distribute, among a plurality of processing units, tasks from one or more workflows; a history list to track all tasks distributed by the distribution circuitry; an ordering queue to store one or more sub-tasks received from a first processing unit as a result of the first processing unit processing a first task; and wherein, responsive to a detection that all sub-tasks of the first task have been received and that the first task is the oldest task for a given parent workflow tracked by the history list, all sub-tasks associated with the first task are to be placed in a replay queue to be replayed in the order in which each sub-task was received.

Method and apparatus for dynamically balancing task processing while maintaining task order

Technologies for a distributed hardware queue manager include a compute device having a processor. The processor includes two or more hardware queue managers as well as two or more processor cores. Each processor core can enqueue or dequeue data from the hardware queue manager. Each hardware queue manager can be configured to contain several queue data structures. In some embodiments, the queues are addressed by the processor cores using virtual queue addresses, which are translated into physical queue addresses for accessing the corresponding hardware queue manager. The virtual queues can be moved from one physical queue in one hardware queue manager to a different physical queue in a different physical queue manager without changing the virtual address of the virtual queue.

Technologies for a distributed hardware queue manager

Described embodiments generate tasks corresponding to each packet received by a network processor. A destination processing module receives a task and determines, based on the task size, a queue in which to store the task, and whether the task is larger than space available within a current memory block of the queue. If the task is larger, an address of a next memory block in a memory is determined, and the address is provided to a source processing module of the task. The source processing module writes the task to the memory based on a provided offset address and the address of the next memory block, if provided. If a task is written to more than one memory block, the destination processing module preloads the address of the next memory block to a local memory to process queued tasks without stalling to retrieve the address of the next memory block.

Task queuing in a multi-flow network processor architecture

Described embodiments provide a network processor having a plurality of processing modules coupled to a system cache and a shared memory. A memory manager allocates blocks of the shared memory to a requesting one of the processing modules. The allocated blocks store data corresponding to packets received by the network processor. The memory manager maintains a reference count for each allocated memory block indicating a number of processing modules accessing the block. One of the processing modules reads the data stored in the allocated memory blocks, stores the read data to corresponding entries of the system cache and operates on the data stored in the system cache. Upon completion of operation on the data, the processing module requests to decrement the reference count of each memory block. Based on the reference count, the memory manager invalidates the entries of the system cache and deallocates the memory blocks.

Memory manager for a network communications processor architecture

Described embodiments provide a memory manager for a network processor having a plurality of processing modules and a shared memory. The memory manager allocates blocks of the shared memory to requesting ones of the plurality of processing modules. A free block list tracks availability of memory blocks of the shared memory. A reference counter maintains, for each allocated memory block, a reference count indicating a number of access requests to the memory block by ones of the plurality of processing modules. The reference count is located with data at the allocated memory block. For subsequent access requests to a given memory block concurrent with processing of a prior access request to the memory block, a memory access accumulator (i) accumulates an incremental value corresponding to the subsequent access requests, (ii) updates the reference count associated with the memory block, and (iii) updates the memory block with the accumulated result.

Growing community of inventors

Orefield, PA, United States of America

Joseph R Hasting