| Apparatus and method for efficient communication of producer/consumer buffer status -> Monitor Keywords |
|
|
 
Apparatus and method for efficient communication of producer/consumer buffer statusRelated Patent Categories: Electrical Computers And Digital Processing Systems: Multicomputer Data Transferring, Multicomputer Data Transferring Via Shared MemoryApparatus and method for efficient communication of producer/consumer buffer status description/claimsThe Patent Description & Claims data below is from USPTO Patent Application 20070174411, Apparatus and method for efficient communication of producer/consumer buffer status. Brief Patent Description - Full Patent Description - Patent Application Claims
BACKGROUND [0001] 1. Technical Field [0002] The present application relates generally to an improved data processing system. More specifically, the present application is directed to an apparatus and method for efficient communication of producer/consumer buffer status. [0003] 2. Description of Related Art [0004] In a data processing system, one processor or device typically acts as a producer by placing data into a shared buffer while another processor or device acts as a consumer by reading and processing the data placed there by the producer. For example, consider a data processing system where a producer and a consumer, independently operating on processors or devices within the data processing system, have access to a shared data buffer region. The shared data buffer region may be located, for example, in shared system or I/O memory and is commonly accessible by both parties. In addition to the shared buffer region, two state variables are shared between the producer and consumer: a head pointer (or index) and a tail pointer (or index), as shown in FIG. 1A. [0005] FIG. 1A is an exemplary diagram of a shared buffer region in accordance with a prior art system. As shown in FIG. 1A, the shared buffer region 100 has an associated tail pointer 120 and head pointer 110. The head pointer 110 represents the next unused location in the shared buffer region 100. The tail pointer 120 represents the last unprocessed data location in the shared buffer region 100. [0006] As a producer places data into the shared buffer region 100, the producer increments the head pointer 110. In this way, the producer indicates to consumers that new data is available for the consumers. Likewise, as the consumer retrieves data from the shared buffer region 100, the consumer increments the tail pointer 120, thereby indicating that data has been read from the shared buffer region 100. The amount of data available in the shared buffer region 100 is indicated by the difference between the head pointer 110 and tail pointer 120. This is referred to as the "active" data region 140 in the shared buffer region 100. The portion of the shared buffer region 100 that is behind the tail pointer 120 is referred to as the "inactive" (or old) data region 130. An empty shared buffer region is identified by setting the head pointer 110 and tail pointer 120 to the same location in the shared buffer region 100. [0007] When there is no constraint on the size of the shared buffer region, i.e. when there is an infinite amount of storage space available for the shared buffer region 100, the head pointer 110 and tail pointer 120 may be incremented indefinitely. However, in most systems, it is more realistic to have only a fixed amount of storage space available for the shared buffer region 100. In this case, it is possible to reclaim the inactive region 130 by implementing what is known as a circular first-in-first-out (FIFO) type shared buffer region 100. This type of shared buffer region is illustrated in FIG. 1B. [0008] As shown in FIG. 1B, the head pointer 110 and tail pointer 120 are incremented as described above with regard to FIG. 1A, however when the end of the shared buffer region 100 is encountered, the head and tail pointers 110 and 120 are permitted to wrap back to the beginning of the shared buffer region 100. As with the shared buffer region 100 shown in FIG. 1A, the active data region 140 is defined as the portion of the shared buffer region 100 between the head pointer 110 and tail pointer 120. However, with the FIFO shared buffer region, both the producer and consumer must now account for the situation where the head pointer 110 has wrapped beyond the end of the shared buffer region 100, i.e. back to the beginning of the shared buffer region 100, but the tail pointer 120 has not wrapped. [0009] This requires efficient communication of the current values for the head and tail pointers 110 and 120 to both the producer and the consumer. The consumer is primarily concerned with monitoring updates to the head pointer 110 and the producer is primarily concerned with monitoring updates to the tail pointer 120. Before a consumer can begin processing, it must know that the head pointer 110 and tail pointer 120 point to different entries in the shared buffer region 100. Likewise, before the producer can add data to the shared buffer region 100, the producer must ensure that the amount of data to be added will not overflow the tail pointer 120. [0010] A simple implementation may involve storing the state variables, i.e. the head pointer 110 and tail pointer 120, in shared system memory and monitoring these state variables remotely. FIG. 2A illustrates this exemplary implementation. [0011] As shown in FIG. 2A a producer 250 performs updates to the head pointer 210 in shared memory 230 and monitors the current state of tail pointer 220 in shared memory 230 when writing data to the shared buffer region 240. Similarly, the consumer 260 updates the tail pointer 220 in shared memory 230 and monitors the current state of the head pointer 210 in shared memory 230 when reading data from the shared buffer region 240. [0012] The producer 250 and consumer 260 each perform updates and monitoring of the head pointer 210 and tail pointer 220 in shared memory 230 remotely. That is, the head pointer 210 and tail pointer 220 are not located in a memory local to the producer 250 or the consumer 260. Thus, the producer 250 and consumer 260 must access these pointers 210 and 220 via a bus or other communication interface that is not local to the producer 250 or consumer 260. Typically, the producer 250 and consumer 260 poll the remote shared memory 230 in order to obtain information regarding the head and tail pointers 210 and 220. As a result, this updating and monitoring requires additional resources, e.g., processor cycles, bus bandwidth, etc., within the system to support these operations. Because of the need to utilized additional system resources, this solution may be considered too detrimental to overall system performance. [0013] In an alternative approach, the head and tail pointers 210 and 220 are maintained in a memory or registers that are local to either the producer 250 or the consumer 260 in order to improve performance of that one party, i.e. the producer 250 or the consumer 260, to which the pointers 210 and 220 are made local. FIG. 2B illustrates such an alternative implementation in which the head pointer 210 and tail pointer 220 are stored in a local memory region 270 to one of the producer 250 and consumer 260 (in the depicted example, the local memory region 270 is local to the consumer 260). [0014] As shown in FIG. 2B, the head pointer 210 and tail pointer 220 are stored in local memory region 270 that is local to consumer 260. This local memory region 270 is separate from shared memory 230 and shared buffer region 240. The producer 250 must remotely monitor, i.e. poll, the value of the tail pointer 220 in order to determine the amount of free space remaining in the shared buffer region 240. This may lead to serious degradation of performance for the producer, as the latency for reading remote locations can be several orders of magnitude worse than for main shared memory storage. Of course the situation could be reversed, requiring that the consumer 260 remotely monitor the head pointer 210 in a local memory region local to the producer 250, for example. SUMMARY OF THE INVENTION [0015] In view of the above, it would be beneficial to have an improved mechanism for efficient communication of shared buffer status between producers and consumers in a data processing system. The illustrative embodiments provide such an improved mechanism by utilizing already existing communication channels established with processing elements of the data processing system to communicate this shared buffer status information. [0016] With the illustrative embodiments, synergistic processor elements (SPEs) are provided in a data processing system. The SPEs are a simple, yet highly optimized, processor with a single instruction, multiple data (SIMD) instruction set. The SPEs have a large register file, a high speed local memory, and a channel interface that serves as the primary communication pathway between an SPE and external processors or devices. [0017] The channel interface is a message passing interface that overlaps input/output (I/O) with data processing and minimizes power consumed by synchronization. The channel facilities of the channel interface are accessed with three types of instructions: a read channel instruction, write channel instruction, and read channel count instruction which measures channel capacity. These channels may be configured as blocking or non-blocking channels. Channels that are defined as blocking channels cause the SPE to stall when accessing a channel whose channel count is zero. In this sense, the channel count indicates the number of outstanding operations that can be issued for that channel without causing the SPE to stall. Aside from blocking on channel access, an SPE program can monitor a variety of channel events using SPE interrupt facilities. [0018] The SPE channel interface supports two signal notification channels, which may be used by the SPE to receive signals originating from other processors or devices in the system. The signal notification channels can be programmed to operate in over-write mode. Over-write mode is a mode in which the latest notification messages written to the channel "overwrite" previous notification messages written to the channel such that the processors may access the most recent information written to the signal notification channel. The over-write mode of the signal communication channels is useful in a one-to-one communications environment, or in a logical-OR communications environment, such as in a many-to-one signaling environment. [0019] With the illustrative embodiments, the producer, running on an SPE, may read its signal notification channel(s) and operate in a blocking mode until the consumer updates the tail pointer information written to the signal notification channel(s). This provides a low-power, low-bandwidth alternative to monitoring, i.e. polling, a remote memory or register location that is local to the consumer. Further, this solution provides low latency responses for real-time systems since the SPE is automatically awoken as soon as data becomes available on the signal notification channel(s). [0020] In another illustrative embodiment, the producer running on the SPE may use the processor's event and interrupt facilities to receive asynchronous notification of consumer's completion. This solution may be of use in environments where there is other work to be done by the SPE and blocking on a channel is not an acceptable alternative. [0021] Of course these illustrative embodiments may be extended by reversing the roles and having the SPE program act as a consumer, while an external processor or device acts as a producer. In such a case, the producer would notify the SPE of buffer state updates by writing the head pointer to a signal notification register. The embodiments may be further extended by having the SPE program operate as both a consumer (receiving head pointer updates from an external device or processor on a first signal notification channel, for example) and as a producer (receiving tail pointer updates from an external device or processor on a second signal notification channel, for example). [0022] In one illustrative embodiment, a method is provided in which an operation on a shared buffer region is performed using a first device in a data processing system. The method may further comprise writing a pointer update to a signal notification channel associated with a second device in the data processing system. The pointer update may be an update to one of a head pointer or a tail pointer of the shared buffer region. The method may further comprise updating a pointer stored in a local store of the second device based on the pointer update. The signal notification channel may be a memory mapped address region of a shared memory of the data processing system. Continue reading about Apparatus and method for efficient communication of producer/consumer buffer status... Full patent description for Apparatus and method for efficient communication of producer/consumer buffer status Brief Patent Description - Full Patent Description - Patent Application Claims Click on the above for other options relating to this Apparatus and method for efficient communication of producer/consumer buffer status patent application. ###  How KEYWORD MONITOR works... a FREE service from FreshPatents1. Sign up (takes 30 seconds). 2. Fill in the keywords to be monitored. 3. Each week you receive an email with patent applications related to your keywords. Start now! - Receive info on patent apps like Apparatus and method for efficient communication of producer/consumer buffer status or other areas of interest. ### Previous Patent Application: Dynamic network fusion in wireless ad-hoc networks Next Patent Application: Recording/transferring program, recording/transferring apparatus, and recording/transferring method Industry Class: Electrical computers and digital processing systems: multicomputer data transferring or plural processor synchronization ### FreshPatents.com Support Thank you for viewing the Apparatus and method for efficient communication of producer/consumer buffer status patent info. IP-related news and info Results in 0.28742 seconds Other interesting Feshpatents.com categories: Qualcomm , Schering-Plough , Schlumberger , Seagate , Siemens , Texas Instruments , 174 |
 * Protect your Inventions * US Patent Office filing 
PATENT INFO |
|
