Method and apparatus for efficient memory bank utilization in multi-threaded packet processors
A method and apparatus for efficient memory bank utilization in multi-threaded packet processors is presented. A plurality of memory access requests, are received and are buffered by a plurality of memory First In First Out (FIFO) buffers, each of the memory FIFO buffers in communication with a memory controller. The memory access requests are distributed evenly across said memory banks by way of the memory controller. This reduces and/or eliminates memory latency which can occur when sequential memory operations are performed on the same memory bank.
1. An apparatus comprising:
a memory including a plurality of memory banks;
a memory controller in bidirectional communication with said memory; and
a plurality of memory First In First Out (FIFO) buffers, each of said memory FIFO buffers in bidirectional communication with said memory controller, each memory FIFO buffer operational to buffer a request to said memory and wherein said memory controller distributes memory accesses evenly across said memory banks, and wherein a depth of said memory FIFO is determined empirically based on processor speed, memory speed, and a number of threads, and wherein said memory FIFO depth is large enough to prevent said memory FIFO from filling during steady state operation.
2. The apparatus of claim 1 wherein said memory, said memory controller and said memory FIFOs are used as part of a multithread environment.
3. The apparatus of claim 1 wherein each of said memory FIFOs includes an address FIFO.
4. The apparatus of claim 1 wherein each of said memory FIFOs includes a read and write data FIFO.
5. The apparatus of claim 1 wherein said memory is organized as at least one of the group consisting of an inter-bank memory wherein said memory is organized such that each sequential address is the same memory bank and an intra-bank memory wherein said memory is organized such that each sequential address is in a different memory bank.
6. The apparatus of claim 1 wherein each memory bank comprises one of the group consisting of at least one Reduced Latency Dynamic Random Access Memory (RLDRAM) and at least one Synchronous Dynamic Random Access Memory (SDRAM).
7. The apparatus of claim 1 wherein a memory FIFO is provided for each memory bank.
8. The apparatus of claim 1 wherein said memory, said memory controller and said memory FIFOs are used as part of a packet processing environment.
9. A computer-implemented method in which a computer system performs operations comprising:
receiving a plurality of memory access requests in a system having a memory including a plurality of memory banks, a plurality of memory First In First Out (FIFO) buffers and a memory controller in bidirectional communication with a memory;
buffering said memory access requests with said memory FIFO buffers, each of said memory FIFO buffers in bidirectional communication with said memory controller; and
distributing said memory access requests evenly across said memory banks by way of said memory controller, and wherein a depth of said memory FIFO is determined empirically based on processor speed, memory speed, and a number of threads, and wherein said memory FIFO depth is large enough to prevent said memory FIFO from filling during steady state operation.
10. The method of claim 9 wherein said memory, said memory controller and said memory FIFOs are used as part of a multithread environment.
11. The method of claim 9 wherein each of said memory FIFOs includes an address FIFO.
12. The method of claim 9 wherein each of said memory FIFOs includes a read and write data FIFO.
13. The method of claim 9 wherein said memory is organized as at least one of the group consisting of an inter-bank memory wherein said memory is organized such that each sequential address is the same memory bank and an intra-bank memory wherein said memory is organized such that each sequential address is in a different memory bank.
14. The method of claim 9 wherein each memory bank comprises one of the group consisting of at least one Reduced Latency Dynamic Random Access Memory (RLDRAM) and at least one Synchronous Dynamic Random Access Memory (SDRAM).
15. The method of claim 9 wherein said memory, said memory controller and said memory FIFOs are used as part of a packet processing environment.
16. The method of claim 10 wherein a memory FIFO is provided for each memory bank.