Distributed-structure-based parallel module structure and parallel processing method

• US 7,373,481 B2
• Filed: 07/18/2001
• Issued: 05/13/2008
• Est. Priority Date: 07/18/2000
• Status: Expired due to Fees

• Current Assignee: Zhaochang Xu
• Original Assignee: Zhaochang Xu
• Inventors: Xu, Zhaochang
• Primary Examiner(s): Patel; Niketa
• Assistant Examiner(s): Moll; Jesse
• Application Number:
  US10/333,618
• Publication Number:
  US 20040044879A1
• Time in Patent Office: 2,491 Days
• Field of Search: 712/25, 712/32, 712/37, 712/235
• US Class Current: 712/25
• CPC Class Codes: G06F 9/46 Multiprogramming arrangements

Abstract

A Distributed-Structure-based parallel module structure and parallel processing method. One object is to provide a novel sequence-net computer architecture. A parallel operating structure with N+1 independent flow-sequences is created, and the N+1 flow-sequences control independently the distributed token via the sequence-net instructions to realize the parallel operating of module. Wherein N flow-sequences is regular type, a new consistency flow-sequence Sc running independently is composed by consistency tokens. The distributed token connecting among multi-machines support the co-operation running among N+1 flow-sequences.

Claims

1. A parallel processing module stored in a computer readable medium, the module having N (N>

1) flow-sequences, each of said flow-sequences having a plurality of source tokens and destination tokens, said module comprising;

a consistency flow-sequence having at least one consistency token for building a distributed token structure together with at least one of said source tokens and at least one of said destination tokens corresponding to said at least said one source token in said N flow-sequences; and

said distributed token structure being controlled by said at least one consistency token resided in said distributed token to perform a consistency operation, and when the source token value in said distributed token structure is invalid, said consistency flow-sequence being controlled by said consistency token at waiting status, and when the source token value in said distributed token structure is valid, said destination token value in said distributed token structure being valid;

wherein the source token and destination token in said flow-sequence are idle tokens that selectively are transformed into the source token or the destination token when said flow-sequence runs into a position where any one of the source token and the destination token is set and said distributed token is tested by said consistency token, and said consistency token is held at a waiting status when the tokens are idle tokens and/or no source token exists in distributed token.

2. A parallel processing module stored in a computer readable medium, the module having N(N>

1) flow-sequences, each of said flow-sequences having a plurality of source tokens and destination tokens, said module comprising;

a consistency flow-sequence having at least one consistency token for building a distributed token structure together with at least one of said source tokens and at least one of said destination tokens corresponding to said at least said one source token in said N flow-sequences; and

said distributed token structure being controlled by said at least one consistency token resided in said distributed token to perform a consistency operation, and when the source token value in said distributed token structure is invalid, said consistency flow-sequence being controlled by said consistency token at waiting status, and when the source token value in said distributed token structure is valid, said destination token value in said distributed token structure being valid;

wherein said source token comprises data to form a source data-token, said destination token comprises data to form a destination data-token, and the distributed token has at least one of said consistency tokens at least one of said source data-tokens, and a plurality of said destination data-tokens, said distributed token is tested by said consistency token, said consistency token waits when an invalid value of said source data-token results from the test, and the data in said source data-token is transferred to the destination data-token and the token value of the destination token is valid when a valid value of said source data-token results from the test.

3. The parallel processing module recited in claim 2, wherein said source data-token and said destination data-token are idle data-tokens for building a distributed idle data-token together with said consistency token, and said idle data-token is transformed into a source data-token or a destination data-token when said flow-sequence runs to a position where any one of the source token and the destination token is set.

4. The parallel processing module recited in claim 2, wherein the consistency token in said consistency flow-sequence comprises data consistency instruction, the source data-token in said flow-sequence comprises write data instruction, and the destination data-token in said flow-sequence comprises read data instruction.

5. A parallel processing method used with a parallel processing module stored in a computer readable medium, the module having N (N>

1) flow-sequences, wherein each flow-sequence comprises a plurality of source tokens and destination tokens, said module comprising a consistency flow-sequence having consistency tokens for building a distributed token together with a source token and at least one of said destination tokens corresponding to said at least one source token in said N flow-sequences, said parallel processing method comprising the following steps;

1) executing each said flow-sequence independently and the source token value being valid when said flow-sequence runs into the source token;

2) executing said consistency flow-sequence at the same time of executing said flow-sequence;

3) controlling said distributed token with said consistency token resided to perform consistency operation when said consistency flow-sequence runs into said consistency token;

4) controlling said consistency flow-sequence with said consistency token for waiting when the source token value in said distributed token is invalid and said destination token value in said distributed token being valid when said source token value in said distributed token is valid; and

5) determining whether said destination token value is valid when said flow-sequence runs into said destination token;

5.1) if the destination token value is invalid, said flow-sequence does not run until said destination token value is valid;

5.2) if the destination token value is valid, said flow-sequence continuously runs;

wherein the source token and the destination token in said flow-sequence are idle tokens, and said idle tokens are transformed into the source token or the destination token when said flow-sequence runs into a position where any one of the source token and the destination token is set, and said distributed token is tested by said consistency token, and said consistency token is held at a waiting status when the tokens in said flow-sequence are idle tokens.

6. A parallel processing method used with a parallel processing module stored in a computer readable medium, the module having N (N>

1) flow-sequences, wherein each flow-sequence comprises a plurality of source tokens and destination tokens, said module comprising a consistency flow-sequence having consistency tokens for building a distributed token together with a source token and at least one of said destination tokens corresponding to said at least one source token in said N flow-sequences, said parallel processing method comprising the following steps;

1) executing each said flow-sequence independently and the source token value being valid when said flow-sequence runs into the source token;

2) executing said consistency flow-sequence at the same time of executing said flow-sequence;

3) controlling said distributed token with said consistency token resided to perform consistency operation when said consistency flow-sequence runs into said consistency token;

4) controlling said consistency flow-sequence with said consistency token for waiting when the source token value in said distributed token is invalid, and said destination token value in said distributed token value being valid when said source token value in said distributed token is valid; and

5) determining whether said destination token value is valid when said flow-sequence runs into said destination token;

5.1) if the destination token value is invalid, said flow-sequence does not run until said destination token value is valid;

5.2) if the destination token value is valid, said flow-sequence continuously runs;

wherein said source token comprises data to form a source data-token, said destination token comprises data to form a destination data-token, and the distributed token comprises at least one of said consistency tokens, at least one of said source data-tokens, and a plurality of said destination data-tokens, said distributed token is tested by said consistency token, and said consistency token waits when an invalid value of said source data-token results from the test while the data in said source data-token are transferred to the destination data-token and the token value of the destination token is valid when a valid value of said source data-token results from the test.

7. A parallel processing method used with a parallel processing module stored in a computer readable medium, the module having N (N>

1) flow-sequences, wherein each sequence comprises a plurality of source tokens and destination tokens, said module comprising a consistency flow-sequence having consistency tokens for building a distributed token together with a source token and at least one of said destination tokens corresponding to said at least one source token in said N flow-sequences, said parallel processing method comprising the following steps;

1) executing each said flow-sequence independently and the source token value being valid when said flow-sequence runs into the source token;

2) executing said consistency flow-sequence at the same time of executing said flow-sequence;

3) controlling said distributed token with said consistence token resided to perform consistency operation when said consistency flow-sequence runs into said consistency token;

4) controlling said consistency flow-sequence with said consistency token for waiting when the source token value in said distributed token is invalid, and said destination token value in said distributed token being valid when said source token value in said distributed token is valid; and

5) determining whether said destination token value is valid when said flow-sequence runs into said destination token;

5.1) if the destination token value is invalid, said flow-sequence does not run until said destination token value is valid;

5.2) if the destination token value is valid, said flow-sequence continuously runs;

wherein said source token comprises data to form a source data-token, said destination token comprises data to form a destination data-token, and the distributed token comprises at least one of said consistency tokens, at least one of said source data-tokens, and a plurality of said destination data-tokens, said source data-token and said destination data-token are idle data-tokens for building a distributed idle data-token together with said at least one consistency token, and said idle data-token is transformed into source data-token or destination data-token when said flow-sequence runs to a position where any one of the source token and the destination token is set.

8. A parallel processing method used with a parallel processing module stored in a computer readable medium, the module having N (N>

1) flow-sequences, wherein each sequence comprises a plurality of source tokens and destination tokens, said module comprising a consistency flow-sequence having consistency tokens for building a distributed token together with a source token and at least one of said destination tokens corresponding to said at least one source token in said N flow-sequences, said parallel processing method comprising the following steps;

1) executing each said flow-sequence independently and the source token value being valid when said flow-sequence runs into the source token;

2) executing said consistency flow-sequence at the same time of executing said flow-sequence;

3) controlling said distributed token with said consistency token resided to perform consistency operation when said consistency flow-sequence runs into said consistency token;

4) controlling said consistency flow-sequence with said consistency token for waiting when the source token value in said distributed token is invalid, and said destination token value in said distributed token being valid when said source token value in said distributed token is valid; and

5) determining whether said destination token value is valid when said flow-sequence runs into said destination token;

5.1) if the destination token value is invalid, said flow-sequence does not run until said destination token value is valid;

5.2) if the destination token value is valid, said flow-sequence continuously runs;

wherein said source token comprises data to form a source data-token, said destination token comprises data to form a destination data-token, and the distributed token comprises at least one of said consistency tokens, at least one of said source data-tokens, and a plurality of said destination data-tokens, and the consistency token in said consistency flow-sequence comprises data consistency instruction, the source data-token in said flow-sequence comprises write data instruction, and the destination data-token in said flow-sequence comprises read data instruction.

9. A parallel processing module having N (N>

1) unit modules, a consistency flow-sequence, and a data-token bus, wherein each of said unit modules is connected with said data-token bus and the consistency flow-sequence through consistency flow-sequence interface in the module, each of said unit modules comprising a computer, data-token control modules and data-token arrays, said computer connecting with data-token arrays via a computer connecting device in said data-token control module and performing reading and writing operation of said data-token arrays via said data-token control module, said consistency flow-sequence being connected with a multi-computer connecting device in said data-token control module to transmit data and tokens via said data-token bus to realize data-token consistency operation among N unit modules, said data-token array comprising data bits and token bits and said consistency token sent by consistency flow-sequence comprising a trigger time sequence and a data-token address.

10. The parallel processing module recited in claim 9, wherein the data is transmitted into an address specified in a local data-token array while a source token that owns the same address is created to present that said data is valid when said computer runs to write data instruction.

11. The parallel processing module recited in claim 10, wherein said consistency flow-sequence sends a consistency token that comprises trigger time sequence and address of data-token to a multi-computer connecting device when said consistency flow-sequence runs to a data consistency instruction, said data-token arrays in said unit module being addressed by the same data-token address, and the data and token in this data-token array broadcast to all the data-token arrays of said unit module via data-token bus when said source token is valid.

12. The parallel processing module recited in claim 11, wherein the data in the local data-token array is read when said computer runs to read data instruction, the destination token which owns the same address is created to present that said destination token is tested, and if said destination token is invalid, said computer waiting or re-reading this destination token.

13. The parallel processing module recited in claim 9, wherein said consistency flow-sequence has N cell-computers, each said cell-computer being connected with at least one said data-token module and the cell-computers being connected by a cell-computer bus.

14. The parallel processing module recited in claim 13, wherein said cell-computer, data-token control module, and data-token array are integrated into a single card.

15. The parallel processing module recited in claim 13, wherein said computer, cell-computer, data-token control module, and data-token array are integrated into a single chip.