Multi-tasking adapter for parallel network applications

• US 6,072,781 A
• Filed: 10/22/1996
• Issued: 06/06/2000
• Est. Priority Date: 10/22/1996
• Status: Expired due to Fees

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Wilhelm, George William Jr., Feeney, James William, Olnowich, Howard Thomas
• Primary Examiner(s): Maung, Zarni
• Assistant Examiner(s): DINH, KHANH Q
• Application Number:
  US08/734,946
• Time in Patent Office: 1,323 Days
• Field of Search: 370/282, 370/276, 370/229, 370/417, 370/400, 710/22, 710/28, 709/215, 709/212, 709/234
• US Class Current: 370/282
• CPC Class Codes: G06F 9/544 Buffers; Shared memory; Pipes

Abstract

A communications apparatus is provided comprising a plurality of FIFO buffers, each with independent control and priority logic under software control for supporting different types of message traffic, both send and receive, such as comprise a multimedia server system. Processor software directs messages to specific, optimized FIFO buffers. Further, a system is provided including a plurality of nodes wherein a sending node specifies the communications path through the system, selecting specific FIFO buffers in each node for buffering its messages.

Claims

1. A store-and-forward adapter for interconnecting a nodal processor to a network comprising:

communication means for communicating messages with respect to said network;

a plurality of addressable send FIFO buffers for storing and forwarding messages from said nodal processor to said communication means;

a plurality of addressable receive FIFO buffers for storing and forwarding messages from said communication means to said nodal processor;

priority means for assigning a priority level to each said FIFO buffer;

selection means responsive to said priority means for determining which send FIFO buffer is to forward a first next message to said communication means;

routing means responsive to said priority means for

determining which said receive FIFO buffer is to store a second next message received at said communication means;

said send FIFO buffers and said receive FIFO buffers being implemented within an adapter memory separate from a nodal processor memory;

a plurality of sets of control registers programmable and readable by said nodal processor, one said set of control registers for controlling each said send and receive FIFO buffer;

each said send and receive FIFO buffer is programmable by said nodal processor to a unique size and priority, and to a specific location in said adapter memory;

said unique set of control registers including;

an options register for defining for a given FIFO buffer its size, priority, and location in said adapter memory;

a FIFO write pointer for defining the address in adapter memory where the next message is to be written;

a FIFO write counter for defining the address in adapter memory where each subsequent word of said message is to be written;

a FIFO read pointer for defining the address in adapter memory from where the next message is to be read;

a FIFO read counter for defining the address in adapter memory from where each subsequent word of said message is to be read; and

a FIFO status register for defining for said given FIFO buffer its interrupt, error, empty, full, and enabled conditions.

2. A store-and-forward adapter for interconnecting a nodal processor to a network comprising:

communication means for communicating messages with respect to said network;

a plurality of addressable send FIFO buffers for storing and forwarding messages from said nodal processor to said communication means;

a plurality of addressable receive FIFO buffers for storing and forwarding messages from said communication means to said nodal processor;

priority means for assigning a priority level to each said FIFO buffer;

selection means responsive to said priority means for determining which send FIFO buffer is to forward a first next message to said communication means;

routing means responsive to said priority means for determining which said receive FIFO buffer is to store a second next message received at said communication means;

said send FIFO buffers and said receive FIFO buffers being implemented within an adapter memory separate from a nodal processor memory;

a unique send and receive FIFO buffers pair being allocated to each of a plurality of software applications being executed on said nodal processor;

a different priority level being assigned to each of a plurality of said FIFO buffers pairs;

three send FIFO buffers and three receive FIFO buffers forming three FIFO buffers pairs for executing three software applications simultaneously on said nodal processor, said three FIFO buffers pairs including a first FIFO buffers pair assigned the highest priority for processing control messages for co-ordination of activities amongst a plurality of nodal processors connected to said network, a second FIFO buffers pair assigned the middle priority for processing medium-sized messages associated with high-level protocol communication amongst said plurality of nodal processors, and a third FIFO buffers pair assigned the lowest priority for processing multimedia data messages.

3. A store-and-forward adapter for interconnecting a nodal processor, responsive to two software applications executing simultaneously on said nodal processor, to a network, comprising:

communication means for communicating messages with respect to said network;

a plurality of addressable send FIFO buffers for storing and forwarding messages from said nodal processor to said communication means;

a plurality of addressable receive FIFO buffers for storing and forwarding messages from said communication means to said nodal processor;

priority means for assigning a priority level to each said FIFO buffer;

selection means responsive to said priority means for determining which send FIFO buffer is to forward a first next message to said communication means;

routing means responsive to said priority means for determining which said receive FIFO buffer is to store a second next message received at said communication means;

said send FIFO buffers and said receive FIFO buffers being implemented within an adapter memory separate from a nodal processor memory;

a unique send and receive FIFO buffers pair being allocated to each of a plurality of software applications being executed on said nodal processor;

a different priority level being assigned to each of a plurality of said FIFO buffers pairs;

said FIFO buffers pairs include a first FIFO buffers pair assigned the highest priority for processing small control messages associated with a first application and for processing medium-sized messages associated with a second application for high-level protocol communication amongst said plurality of nodal processors, and second and third FIFO buffer pairs assigned the lowest priority for processing multimedia data messages with respect to said first application.

4. A store-and-forward adapter for interconnecting a nodal processor to a network comprising:

communication means for communicating messages with respect to said network;

a plurality of addressable send FIFO buffers for storing and forwarding messages from said nodal processor to said communication means;

a plurality of addressable receive FIFO buffers for storing and forwarding messages from said communication means to said nodal processor;

priority means for assigning a priority level to each said FIFO buffer;

selection means responsive to said priority means for determining which send FIFO buffer is to forward a first next message to said communication means;

routing means responsive to said priority means for determining which said receive FIFO buffer is to store a second next message received at said communication means;

said send FIFO buffers and said receive FIFO buffers being implemented within an adapter memory separate from a nodal processor memory;

a unique send and receive FIFO buffers pair being allocated to each of a plurality of software applications being executed on said nodal processor;

each said receive FIFO buffer including a plurality of direct memory access (DMA) channels, each said DMA channel comprising;

a DMA control program stored to said adapter memory by said nodal processor for defining the address location in said nodal processor to receive said second next message, and further defining transfer control and program validity;

selection means associated with each received message for selecting and activating a DMA control program in one of said DMA channels; and

alerting means to notify said nodal processor that said DMA channel has stored said second next message to nodal processor memory.

5. The adapter of claim 4, wherein said selection means for selecting and activating said DMA control program is provided by a control field in said received message.

6. The adapter of claim 5, wherein said control field includes indicia for (1) defining whether said message is to be processed by a DMA control program, (2) defining which one of the plurality of receive FIFO buffers is to store-and forward said message, and (3) selecting the DMA control program.

7. The adapter of claim 5, wherein said selection means comprises;

a plurality of pointers stored in said adapter memory, each said pointer pointing to the first word of a DMA control program;

address formation means for combining said control field with a preset base address to form a pointer address in said adapter memory;

access means responsive to said pointer address for accessing the pointer to the first word of a selected DMA control program; and

incrementing means for incrementing said pointer to access successive words of said DMA control program.

8. The adapter of claim 5, said alerting means further comprising:

detection means for detecting the successful completion of a store-and-forward operation controlled by said DMA control program and subsequently posting a completion status;

a status register in said adapter containing a status bit for each said DMA channel, wherein each status bit is uniquely settable upon detection of a successful completion of a store-and-forward operation;

reading means operable under control of said nodal processor for reading a group of status bits from said status register; and

resetting means operable under control of said nodal processor for resetting status bits individually in said group of status bits by performing a write operation to said status register.

9. The adapter of claim 8, wherein said detection means is further operable for posting an interrupt to said nodal processor concurrently with posting said completion status.

10. The adapter of claim 9, further comprising interrupt resetting means responsive to said nodal processor for resetting said posted interrupt concurrently with the resetting of said status bits.

11. The adapter of claim 8, wherein said detection means further modifies a first pointer in adapter memory to contain a new value as specified by said selected DMA control program.

12. The adapter of claim 11, wherein said first pointer is modified to a new value as specified by a control field of said DMA control program, and wherein a first new value disables further processing by said DMA channel until said nodal processor writes a second new value to adapter memory to enable said DMA channel.

13. The adapter of claim 11, wherein said first pointer is modified to a new value as specified by a control field of said DMA control program, and said new value enables further processing by said DMA channel by pointing to the same or a different DMA control program.

14. The adapter of claim 8, wherein said DMA control program comprises a plurality of linked DMA programs, each containing a different nodal processor address for defining where an associated received message is to be forwarded, for scattering said received message to a plurality of nodal processor addresses as defined by said plurality of linked DMA programs.

15. The adapter of claim 8, wherein a flag in said control field in said received message selects for processing said received message either (a) a DMA control program, or (b) a processing apparatus comprising

posting means for posting an interrupt from said receive FIFO buffer to said nodal processor to alert said nodal processor that said received message is waiting in said adapter memory;

reading means at said nodal processor for reading said received message directly from said adapter memory; and

interrupt resetting means at said nodal processor for deleting said received message from said receive FIFO buffer, resetting said interrupt, and enabling the receive FIFO buffer to process subsequent received messages.

16. The adapter of claim 8, including apparatus, responsive to said DMA control program indicating said DMA control program is not valid, for processing said received message, said apparatus comprising:

posting means for posting an interrupt from said receiving FIFO to said nodal processor to alert said processor that said received message is waiting in said adapter memory;

reading means at said processor for reading control fields from said received message directly from said adapter memory;

storing means at nodal processor for storing a valid DMA control program to said adapter memory; and

resetting means at said nodal processor for resetting said interrupt, starting said DMA channel, and enabling said receive FIFO buffer to process said received message by DMA.

17. The adapter of claim 7, wherein said plurality of pointers are stored in a pointer area of adapter memory specified as separate from the FIFO buffer areas by a hardware control register.

18. A store-and-forward adapter for interconnecting a nodal processor to a network, comprising:

communication means for communicating messages with respect to said network;

a plurality of addressable send FIFO buffers for storing and forwarding messages from said nodal processor to said communication means;

a plurality of addressable receive FIFO buffers for storing and forwarding messages from said communication means to said nodal processor;

priority means for assigning a priority level to each said FIFO buffer;

selection means responsive to said priority means for determining which send FIFO buffer is to forward a first next message to said communication means;

routing means responsive to said priority means for determining which said receive FIFO buffer is to store a second next message received at said communication means;

said send FIFO buffers and said receive FIFO buffers being implemented within an adapter memory separate from a nodal processor memory;

a unique send and receive FIFO buffers pair being allocated to each of a plurality of software applications being executed on said nodal processor;

a plurality of Direct Memory Access (DMA) sending channels;

each said send FIFO further comprising a sending list of pointers to said DMA sending channels;

each said sending channel including;

a DMA Control Program stored to said adapter memory by said nodal processor defining the nodal processor or system address from where the associated message is to be forwarded, and further defining transfer control and program validity;

selection means associated with each said sending list for selecting and activating one of said plurality of DMA Control Programs; and

alerting means for notifying said nodal processor that said DMA channel has completed forwarding said send message to said network.

19. The adapter of claim 18, wherein said sending list contains a plurality of pointers to DMA Control Programs stored individually to said sending list by said nodal processor writing to a preset listing area of adapter memory, such that the first pointer written becomes the top of said sending list, and the last pointer written becomes the bottom of said sending list.

20. The adapter of claim 19, further comprising a send locator register for specifying a location in adapter memory unique from the FIFO areas containing said preset listing area.

21. The adapter of claim 20, wherein said send locator register is programmable by said nodal processor.

22. The adapter of claim 21, wherein said send locator register is further programmable from an interface to said network.

23. The adapter of claim 22, wherein said sending list contains a stack of pending pointers to sending DMA Channel Programs issued by said nodal processor.

24. The adapter of claim 23, wherein said nodal processor stacks a plurality of pointers to said sending lists, and continues to stack further pointers to said sending lists prior to the completion of the previous stack of pending pointers.

25. The adapter of claim 24, wherein said adapter processes the top pointer of said sending list next, and upon successful completion of the DMA Channel Program pointed to by said top pointer, said adapter removes said top pointer from said sending list and moves the next sequential pointer in said sending list to the top.

26. The adapter of claim 25, wherein said nodal processor reads a hardware control register in said adapter to determine the status of said sending list, status indications including empty, full, or the number of pending pointers comprising said list sending list.

27. The adapter of claim 25, wherein said selection means selects the pointer at the top of said sending list from one of a plurality of sending lists based on the priority assigned to said sending lists.

28. The adapter of claim 25, wherein the priority assigned to each said sending list is different for each sending list, and the highest priority list is executed continually, one DMA Channel Program after another, until said list is empty, then the next highest priority list is executed continually until it is empty or until the higher priority list becomes not empty.

29. The adapter of claim 25, wherein the priority assigned to said sending lists are equal for some or all of said sending lists, and said sending lists having equal priority are executed using a round robin mechanism, where one DMA Channel Program is executed from each of said sending lists having the same priority before a second DMA Channel Program is executed from any of the lists having the same priority.

30. The adapter of claim 27, wherein said selection means comprises:

means for reading a first pointer from the top of a selected list, said first pointer pointing to the first word of a selected DMA Control Program;

access means responsive to said first pointer for accessing said first word of said selected DMA Control Program;

address incrementing means responsive to said first pointer for incrementally addressing successive words of said selected DMA Control Program.

31. The adapter of claim 19, said alerting means further comprising:

detection means for detecting the successful completion of a store-and-forward operation controlled by said DMA control program and subsequently posting a completion status;

a status register in said adapter containing a status bit for each said DMA channel, wherein each status bit is uniquely settable upon detection of a successful completion of a store-and-forward operation;

reading means operable under control of said nodal processor for reading a group of status bits from said status register; and

resetting means operable under control of said nodal processor for resetting status bits individually in said group of status bits by performing a write operation to said status register.

32. The adapter of claim 31, wherein said detection means is further operable for posting an interrupt to said nodal processor concurrently with posting said completion status.

33. The adapter of claim 32, wherein an interrupt resetting means is provided whereby said nodal processor resets said posted interrupt simultaneously with said resetting means for resetting status bits.

34. The adapter of claim 20, said DMA Control Program comprising:

a plurality of linked DMA Programs, each said linked DMA Program including address means for addressing a nodal processor or system from where an associated send message is to be read, and

means responsive to said linked DMA Programs for gathering said associated send messages from a plurality of nodal processor or system addresses.

35. The adapter of claim 18, further comprising:

posting means responsive to said DMA Control Program indicating said DMA Control Program is not valid for posting an interrupt from said sending FIFO to said nodal processor alerting said processor that said send message is waiting in said adapter memory;

said nodal processor further including

storing means responsive to a posted interrupt for storing a valid DMA Control Program to said adapter memory; and

interrupt resetting means for resetting said posted interrupt, starting said DMA Channel, and enabling said send FIFO to process said send message by DMA.

36. The adapter of claim 31, wherein the status bits in said status register are settable as determined by control bits in said pointers to said DMA Control Program.

37. A store-and-forward adapter for interconnecting a nodal processor to a network, comprising

communication means for communicating messages with respect to said network;

a plurality of addressable send FIFO buffers for storing and forwarding messages from said nodal processor to said communication means;

a plurality of addressable receive FIFO buffers for storing and forwarding messages from said communication means to said nodal processor;

said adapter being operable to control said plurality of send FIFO buffers and said plurality of receive FIFO buffers;

priority means for assigning a priority level to each said FIFO buffer;

selection means responsive to said priority means for determining which send FIFO buffer is to forward a first next message to said communication means;

routing means responsive to said priority means for determining which said receive FIFO buffer is to store a second next message received at said communication means;

said send FIFO buffers and said receive FIFO buffers being implemented within an adapter memory separate from a nodal processor memory;

a plurality of sets of control registers programmable and readable by said nodal processor, one said set of control registers for controlling each said send and receive FIFO buffer;

nodal processor selectively reading and writing said control registers and adapter memory selectively directly or from an addressed FIFO buffer;

a plurality of address bits encoded in the address used by said nodal processor to address said adapter, said address bits being encoded selectively to define;

first, a read or write operation to said adapter hardware control registers, where the specific register is defined by the remainder of the address bits;

second, a read or write operation to a selected send FIFO buffer;

third, a read or write operation to said adapter memory directly, where the specific memory location is defined by the remainder of the address bits; and

fourth, a read or write operation to a selected receive FIFO buffer.

38. A store-and-forward adapter for interconnecting a nodal processor to a network, comprising:

communication means for communicating messages with respect to said network;

a plurality of addressable send FIFO buffers for storing and forwarding messages from said nodal processor to said communication means;

a plurality of addressable receive FIFO buffers for storing and forwarding messages from said communication means to said nodal processor;

selection means for determining which send FIFO buffer is to forward a first next message to said communication means;

routing means for determining which said receive FIFO buffer is to store a second next message received at said communication means;

a unique send and receive FIFO buffers pair being allocated to each of a plurality of software applications being executed on said nodal processor;

a different priority level being assigned to each of a plurality of said FIFO buffers pairs;

three send FIFO buffers and three receive FIFO buffers forming three FIFO buffers pairs for executing three software applications simultaneously on said nodal processor, wherein said three FIFO buffers pairs include a first FIFO buffers pair assigned the highest priority for processing control messages for co-ordination of activities amongst a plurality of nodal processors connected to said network, a second FIFO buffers pair assigned the middle priority for processing medium-sized messages associated with high-level protocol communication amongst said plurality of nodal processors, and a third FIFO buffers pair assigned the lowest priority for processing multimedia data messages.

39. A store-and-forward adapter responsive to two software applications executing simultaneously on a nodal processor for interconnecting said nodal processor to a network, comprising:

communication means for communicating messages with respect to said network;

a plurality of addressable send FIFO buffers for storing and forwarding messages from said nodal processor to said communication means;

a plurality of addressable receive FIFO buffers for storing and forwarding messages from said communication means to said nodal processor;

selection means for determining which send FIFO buffer is to forward a first next message to said communication means; and

routing means for determining which said receive FIFO buffer is to store a second next message received at said communication means;

a unique send and receive FIFO buffers pair being allocated to each of a plurality of software applications being executed on said nodal processor;

said FIFO buffers pairs include a first FIFO buffers pair assigned the highest priority for processing small control messages associated with a first application and for processing medium-sized messages associated with a second application for high-level protocol communication amongst said plurality of nodal processors, and second and third FIFO buffer pairs assigned the lowest priority for processing multimedia data messages with respect to said first application.

40. Method for interconnecting a nodal processor to a network, comprising the steps of:

storing a plurality of DMA control programs in respective DMA channels in a plurality of receive FIFO buffers in adapter memory;

responsive to a received message from said network, selecting one of said receive FIFO buffers and activating a DMA control program in one of said DMA channels;

storing said received message to nodal processor memory;

notifying said nodal processor that said DMA channel has stored said received message to nodal processor memory by posting a completion status to a status register;

reading to said nodal processor a croup of status bits from said status resister; and

resetting individual status bits in said status register by writing from said nodal processor,

responsive to a flag in a control field in said received message, selecting for processing said received message either (a) a DMA control program, or (b) a process including the steps of;

posting an interrupt from said receive FIFO buffer to said nodal processor to alert said nodal processor that said received message is waiting in said adapter memory;

reading said received message directly from said adapter memory to said nodal processor; and

from said nodal processor, deleting said received message from said receive FIFO buffer, resetting said interrupt, and enabling the receive FIFO buffer to process subsequent received messages.

41. Method for interconnecting a nodal processor to a network, comprising the steps of:

storing a plurality of DMA control programs in respective DMA channels in a plurality of receive FIFO buffers in adapter memory;

responsive to a received message from said network, selecting one of said receive FIFO buffers and activating a DMA control program in one of said DMA channels;

storing said received message to nodal processor memory;

notifying said nodal processor that said DMA channel has stored said received message to nodal processor memory by posting a completion status to a status register;

reading to said nodal processor a group of status bits from said status register;

resetting individual status bits in said status register by writing from said nodal processor;

responsive to said DMA control program indicating said DMA control program is not valid, processing said received message according to the steps of;

posting an interrupt from said receiving FIFO to said nodal processor to alert said processor that said received message is waiting in said adapter memory;

reading control fields from said received message directly from said adapter memory;

for storing a valid DMA control program to said adapter memory; and

resetting said interrupt, starting said DMA channel, and enabling said receive FIFO buffer to process said received message by DMA.

42. Method for interconnecting a nodal processor to a network comprising:

operating a plurality of addressable send FIFO buffers for storing and forwarding messages from said nodal processor to said communication means;

operating a plurality of addressable receive FIFO buffers for storing and forwarding messages from said network to said nodal processor;

assigning a priority level to each said FIFO buffer;

responsive to said priority level, determining which send FIFO buffer is to forward a first next message to network;

responsive to said priority level, determining which said receive FIFO buffer is to store a second next message received from said network;

allocating a unique send and receive FIFO buffers pair to each of a plurality of software applications being executed on said nodal processor;

operating three FIFO buffer pairs simultaneously, including;

coordinating activities amongst a plurality of nodal processors connected to said network by processing control messages in a first FIFO buffers pair assigned the highest priority;

processing medium-sized messages associated with high-level protocol communication amongst said plurality of nodal processors in a second FIFO buffers pair assigned the middle priority; and

processing multimedia data messages in a third

FIFO buffers pair assigned the lowest priority.

43. Method for interconnecting a nodal processor as a node via an adapter to a network wherein a sending node transmits a message from a first adapter across the network to a second adapter of a receiving node, comprising the steps of:

storing a plurality of DMA control programs in respective DMA channels in a plurality of receive FIFO buffers in memory in said second adapter at the receiving node;

commanding priority logic at said second at the receiving node adapter by the nodal processor to prioritize the sending of received messages from said plurality of receive buffers to the nodal processor;

receiving a receive message at said second adapter from the network and routing said message to one of the plurality of receive FIFO buffers at the receiving node, wherein the selection of which receive FIFO buffer is made at said second adapter is based on the header of the received message;

responsive to priority logic and a received message from said network, selecting one of said receive FIFO buffers at the receiving node and activating a DMA control program in one of said DMA channels based on snooping the message header of the receive message;

storing said received message to nodal processor memory at the receiving node;

notifying said nodal processor that said DMA channel has stored said received message to nodal processor memory by posting a completion status to a status register;

reading to said nodal processor a group of status bits from said status register; and

resetting individual status bits in said status register by writing from said nodal processor.