Method and system for network communication

The present invention relates to network systems, and more particularly, to offloading host system tasks for managing network related operations.

Computer networks are commonly used in various applications. Computing systems typically use a network adapter to communicate with other networked devices. Continuous efforts are being made to efficiently improve processing of network data.

In one embodiment, a method for network communication is provided. The method comprises setting a data size for a network connection, wherein the data size represents an amount of network data a network adapter can send to a host system for the network connection before the network adapter waits for an application to accept any data that has been sent to the host system; monitoring the amount of network data that is received by a host system driver; monitoring the amount of network data that is sent by the network adapter; and suspending transfer of network data to the host system, if the amount of network data sent by the network adapter is similar to the set data size.

In another embodiment, a method for network communication is provided. The method comprises setting a data size for a network connection, wherein the data size represents an amount of network data a network adapter can send to a host system for the network connection before the network adapter waits for an application accept any data that has been sent to the host system; monitoring an amount of network data that is received by a host system driver; wherein a value for a first counter for a host system memory indicates the amount of network data received by the host system driver; monitoring an amount of network data that is sent by the network adapter; wherein a value for a second counter for a network adapter memory indicates the amount of data that is sent to the host system; and suspending transfer of network data to the host system, if the amount of network data sent by the network adapter is similar to the set data size.

In yet another embodiment, a system for network communication is provided. The system comprises a host system executing an application for communicating with at least one networked device; and a network adapter interfacing with the host system and receiving network data from the at least one network device; wherein a data size is set for a network connection and the data size represents an amount of network data the network adapter can send to the host system for the network connection before the network adapter waits for the application to accept any data that has been sent to the host system; and the network adapter suspends transfer of network data to the host system, if the amount of network data sent by the network adapter is similar to the set data size after monitoring the amount of network data that is received by a host system driver and monitoring the amount of network data that is sent by the network adapter.

This brief summary has been provided so that the nature of the invention may be understood quickly. A more complete understanding of the invention can be obtained by reference to the following detailed description of the various embodiments thereof concerning the attached drawings.

The foregoing features and other features of the present invention will now be described with reference to the drawings of various embodiments. In the drawings, the same components have the same reference numerals. The illustrated embodiments are intended to illustrate, but not to limit the invention. The drawings include the following figures:

FIG. 1A shows an example of various protocol layers in a networking system;

FIG. 1B shows a networking system, according to one embodiment;

FIG. 1C shows a block diagram of an adapter having a TOE Module, according to an embodiment;

FIG. 1D shows a top-level block diagram of software architecture, according to an embodiment;

FIG. 1E shows a top-level block diagram of buffer pools created in a host memory, according to an embodiment;

FIG. 1F shows a top-level block diagram of counters location, according to an embodiment;

FIG. 2 shows a process flow diagram for initialization of a system for TOE operation;

FIG. 3 shows a process flow diagram of establishing network connection for the TOE operation; and

FIGS. 4A-4F show a process flow diagram of a TOE Driver's receive algorithm, according to an embodiment.

To facilitate the understanding of the various embodiments, a top-level description of common network protocols/standards and the general architecture/operation of a host system will be described. The specific architecture and operation of the various embodiments will then be described with reference to the general architecture.

Computing systems and devices use various protocols/standards for network communication. Computer networks typically use a layered protocol structure to manage network traffic. One common model that is typically used is an ISO model that includes a physical layer, a data link layer that includes a MAC layer, a network layer and other layers also. Upper level protocol layers (ULPs) (for example, iSCSI and RDMA) described below, interface with a network layer to send and receive data from the network.

FIG. 1A shows an example of various protocol layers ( 10 ) that may be used in networking systems. Data link layer (which includes a MAC layer) 20 interfaces with an IP (Internet Protocol) layer 30 . TCP (Transmission Control Protocol) layer 40 typically sits on top of the IP layer 30 . Upper Layer Protocols (ULPs) 50 may include plural layers, for example, iSCSI layer 70 , RDMA layer 60 and others. The following provides a brief introduction to some of these standards/protocols:

Transmission Control Protocol/Internet Protocol (“TCP/IP”): TCP ( 40 ) is a standard network protocol (incorporated herein by reference in its entirety) that provides connection-oriented, reliable, byte stream service. This means that two nodes establish a logical connection before sending data and TCP maintains state information regarding the data transfer. Reliable means that data is delivered in the same order that it was sent. A byte stream service means that TCP views data to be sent, as a continuous data stream that is sent in any way it sees fit and delivers it to a remote node as a byte stream.

IP ( 30 ) is a standard protocol (incorporated herein by reference in its entirety) that provides a datagram service whose function is to enable routing of data through various network subnets. Each of these subnets could be a different physical link such as Ethernet, ATM, or others. IP layer 30 is responsible for fragmentation of transmit data to match a local link's maximum transmission unit (MTU). IP layer 30 fragments data at a source node or at any intervening router between a source and a destination node.

iSCSI Protocol: Internet SCSI (iSCSI) protocol as defined by the Internet Engineering Task Force (IETF) maps the standard SCSI protocol on top of the TCP/IP protocol. iSCSI (incorporated herein by reference in its entirety) is based on the Small Computer Systems Interface (“SCSI”) standard, which enables host computer systems to perform block level input/output (“I/O”) operations with a variety of peripheral devices including disk and tape devices, optical storage devices, as well as printers and scanners. The iSCSI and TCP/IP protocol suite consist of four protocol layers: the application layer (of which iSCSI is one layer); the transport layer (TCP), the network layer (IP) and the link layer (i.e. Ethernet).

A traditional SCSI connection between a host system and peripheral device is through parallel cabling and is limited by distance and device support constraints. For storage applications, iSCSI was developed to take advantage of network architectures based on Ethernet standards.

The iSCSI architecture is based on a client/server model. Typically, the client is a host system such as a file server that issues a read or write command. The server may be a disk array that responds to the client's request. Typically, the client is an initiator that initiates a read or write command and a disk array is a target that accepts a read or write command and performs the requested operation.

In a typical iSCSI exchange, an initiator sends a “read” or “write” command to a target. For a read operation, the target sends the requested data to the initiator. For a write command, the target sends a “Ready to Transfer Protocol Data Unit (“PDU”)” informing the initiator that the target is ready to accept the write data. The initiator then sends the write data to the target.

Once the data is transferred, the exchange enters the response phase. The target then sends a response PDU to the initiator with the status of the operation. Once the initiator receives this response, the exchange is complete. The use of TCP guarantees the delivery of the PDUs.

Typically, logical units in a target process commands. Commands are sent by a host system in Command Descriptor Blocks (“CDB”). A CDB is sent to a specific logical unit and may include a command to read a specific number of data blocks. The target's logical unit transfers the requested data block to the initiator, terminating with a status message, indicating completion of the request. iSCSI encapsulates CDB transactions between initiators and targets over TCP/IP networks.

As iSCSI becomes popular various software solutions to execute the iSCSI layer in software are emerging. Host system software typically executes the iSCSI layer in software. However, this process is slow and may consume host system processor time and resources, especially for generating digests and checking cyclic redundancy code (CRC).

Operating systems, for example, Microsoft Chimney support offloading of TCP/IP protocol stack from a host system to a network adapter, but do not solve digest and data copy problems for the iSCSI layer. For example, Microsoft defines a data receive algorithm for interoperation between a TCP/IP Offload Engine (TOE) Driver (described below) and the Microsoft Chimney Operating System to deliver a limited amount of data (up to a Maximum Segment Size) when data becomes available. The TOE Driver is blocked from delivering any more data to an Upper layer Protocol (ULP) until the ULP accepts or rejects the delivered data. The ULP may post additional data that may be received after the delivered data.

This process has significant latency between arrival of data and availability of a buffer (memory storage) to store data

A TOE Module (described below) holds off delivery of data to the TOE Driver while the TOE Driver is waiting for the ULP to post a buffer. The ULP only posts a buffer large enough for the delivered data and the TOE Driver notifies the TOE Module to deliver the next data. In this case, the TOE Driver/TOE module interaction adds to the latency of data delivery to the ULP.

In another case when the TOE Module does not hold off delivery of data while the TOE Driver is waiting for a response from the ULP, the TOE Driver holds the received data in an anonymous buffer and then copies the data to the buffers posted by the ULP. This unnecessarily uses host system processor time and resources since data that is buffered by the TOE Driver has to be copied to ULP buffers consuming CPU cycles. The present adaptive aspects, described below tackle the latency issue.

RDMA: Remote Direct Memory Access (RDMA) is a standard upper layer protocol (incorporated herein by reference in its entirety) that assists one computer to directly place information in another computer's memory with minimal demands on memory bus bandwidth and CPU processing overhead. RDMA over TCP/IP defines interoperable protocols to support RDMA operations over standard TCP/IP networks.

FIG. 1B shows an example of networking system 100 , used according to one embodiment. System 100 includes host system 102 , which typically includes several functional components. These components may include central processing unit (CPU) 104 , host memory (or main/system memory) 106 , system bus 108 , an input/output (“I/O”) device (not shown), read only memory (not shown), and other components.

Host memory 106 is coupled to CPU 104 via system bus 108 . Host memory 106 provides CPU 104 access to data and program information that is stored in host memory 106 at execution time. Typically, host memory 106 is composed of random access memory (RAM) circuits.

Host System 102 includes adapter interface 110 , which couples host system 102 to network adapter 114 via bus/connection 112 and host interface 116 . The structure of host interface 116 depends on bus/connection 112 . For example, if bus 112 is a PCI bus, then host interface 116 includes logic and structure to support PCI bus based communication.

Adapter 114 connects host system 102 to network 122 via network interface 118 and network connection 120 . The structure of network interface 118 depends on the type of network, for example, Ethernet, Fibre Channel and others.

FIG. 1C shows a block diagram for adapter 114 interfacing with host system 102 via link (for example, a PCI bus) 112 and host interface 116 . Adapter 114 may be on a PCI development board with a Field Programmable gate Array (FPGA). Adapter 114 may also be integrated into an Application Specific Integrated Circuit (ASIC).

Adapter 114 includes a TCP/IP accelerator module (also referred to as TCP Offload Engine (TOE) 132 ) that executes the TCP/IP protocol stack in the hardware, instead of a software stack at host system 102 . Details of a TOE Module are provided in co-pending patent application Ser. No. 10/620,040, filed on Jul. 15, 2003, incorporated herein by reference in its entirety.

Adapter 114 includes processor 124 that has access to adapter memory 128 . Processor 124 controls overall adapter 114 functionality by executing firmware instructions from memory 128 .

Adapter 114 also includes a direct memory access (“DMA”) engine 126 , which performs direct memory access functions in sending data to and receiving data from host system 102 .

Adapter 114 also includes iSCSI module 130 which includes a dedicated processor or state machine to accomplish this purpose. Instead of the software layer in host system 102 , iSCSI module 130 performs various iSCSI layer operations in adapter 114 , including processing digests, performing data copy offload and large PDU transmission offload operations.

RDMA offload module 134 executes the RDMA protocol functionality in adapter 114 .

FIG. 1D shows a block diagram of a top-level software architecture for implementing the various embodiments disclosed herein. CPU 104 executes operating system (“OS”) 136 in host system 102 . In one example, OS 136 may be based on Microsoft Chimney. Application layer (may also be referred to as ULP) 138 is executed in host system 102 . Application 138 can send read and write requests via driver 142 (may also be referred to as TOE Driver 142 ). OS interface 140 interfaces between TOE Driver 142 and application 138 . Adapter firmware 144 is executed by processor 124 out of memory 128 to control overall adapter 114 functionality.

FIG. 1E shows a top-level block diagram of buffer pools that are created by TOE Driver 142 in host memory 106 to facilitate efficient TOE operation. TOE Driver 142 creates Anonymous Buffer Pool 103 with anonymous buffers (or kernel buffers) 1 -N 107 to temporally store incoming data from the network until application 138 is ready to accept data and move it to a named memory buffer 109 from amongst named memory buffer pool 105 . Named buffers 109 are allocated to store information for a particular connection/application (for example, 138 ). Push Timers are used to make buffers available for a predetermined time only. Push Timers apply to Named buffers and are used to return Named Buffers 109 to an issuing application. Once a push timer has expired the associated buffer is released to anonymous buffer pool 103 . The usage of buffer pools 103 and 105 are described below.

Before describing the details of buffer pools 103 and 105 operations the following defines certain terms that are used to explain the functionality of various embodiments described herein: “indicateWindowSize” is a parameter used by TOE Driver 142 and adapter 114 specify a window size in bytes, representing the amount of data, which can be delivered by adapter 114 to TOE Driver 142 , before adapter 114 suspends data transfer to wait for an application to pass a named buffer to adapter 114 or to acknowledge that data has been stored in anonymous buffer 107 .

The value for “indicateWindowSize” may be set globally (i.e. for every interaction between TOE Driver 142 and adapter 114 ) or on a per network connection basis. TOE Driver 142 may import the “indicateWindowSize” value from a User Interface, a driver parameter file, flash read only memory, NVRAM, as a default value or from TOE Module 132 . Generally, TOE Driver 142 passes the indicateWindowSize value to TOE Module 132 during initialization of adapter 114 . TOE Driver 142 may also specify a unique value for indicateWindowSize for a connection when the connection is offloaded to TOE Module 132 . The term “indicatewindowSize is open” means that data can be accepted from TOE module 132 .

An “indicatedBytes” parameter value indicates the number of bytes that have been received by TOE Driver 142 from adapter 114 (via TOE module 132 ) at any given time. As shown in FIG. 1F , the “indicatedBytes” parameter is based on the value of counter 148 that monitors the number of bytes received by TOE Driver 142 from adapter 114 and placed in host memory 106 .

TOE Driver 142 resets “indicatedBytes” value to zero when “indicatedBytes” equals “indicateWindowSize” and received data has been copied to a named buffer or accepted by application 138 . When the “indicatedBytes” value (i.e. counter 148 ) is reset to zero and there are no named buffers pending, TOE Driver 142 issues an “indicateAcknowledge” command to TOE Module 132 to open a new “indicateWindow” that allows TOE Module 132 to resume sending data to TOE Driver 142 . TOE Driver 142 may optionally send an “indicateAcknowledge” command while data is being accepted to keep the indicate window open.

An “indicatedBytesPending” parameter is based on the value of counter 150 ( FIG. 1F ). Counter 150 is incremented by TOE Module 132 when a segment is sent to TOE Driver 142 . The value of Counter 150 is incremented by the amount of data in the segment sent to TOE Driver 142 . TOE Module 132 suspends sending data to TOE Driver 142 when counter 150 value is equal to “indicateWindowSize”.

When TOE Module 132 receives an “indicateAcknowledge” command from TOE Driver 142 , TOE Module 132 decrements counter 150 by a value specified in an “AcknowledgedBytes” field of the “indicateAcknowledge” command. When TOE Module 132 receives a Named Buffer from TOE Driver 142 , TOE Module 132 resets counter 150 to the value 0.

Once TOE Module 132 has sent a number of bytes equal to or similar to a value specified by the “indicateWindowSize” to TOE Driver 142 , TOE Module 132 stops sending data received from the network. TOE Module 132 holds pending data until TOE Driver 142 posts a buffer for application 138 for pending data or future received data, or acknowledges that some or all of data has been received. At that time, TOE Module 132 resumes sending pending data or future data to TOE Driver 142 .

TOE Driver 142 receives network data sent by TOE Module 132 (in adapter 14 ). When TOE Driver 142 receives a buffer from application 138 , TOE Driver 142 passes the buffer to TOE Module 132 . In one example, the size of the buffer may be greater than the “indicateWindowSize”. In this example, TOE Module 132 DMAs data outside of “indicateWindow” to the buffer passed by TOE Driver 142 . TOE Driver 142 will first copy data in “indicateWindow” to the buffer. The buffers passed to TOE Module 132 are returned when the buffer is filled or a timeout expires.

If the buffer received from application 138 is smaller than the amount of data in “indicateWindow”, then TOE Driver 142 copies the amount of data based on the “indicateWindow” size and returns the buffer to application 138 . TOE Driver 142 then indicates to application 138 that more data is available.

If TOE Driver 142 does not receive a buffer from application 138 and application 138 signals that it has accepted the sent data, TOE Driver 142 then sends the next data based on the “indicateWindow” size to application 138 .

As TOE Driver 142 moves data based on the “indicateWindow” size, it updates counter 148 ( FIG. 1F ). TOE Driver 142 may also inform TOE Module 132 that data has moved by issuing an “indicateAcknowledgement” command to TOE Module 132 .

FIG. 2 shows a process flow diagram for initializing system 100 for network operations. The process starts in step S 200 , when adapter 114 is initialized.

In step S 202 , TOE Driver 142 loads default operating parameters including a default value for “indicateWindowSize”.

In step S 204 , TOE Driver 142 allocates host memory 106 for anonymous buffer pool 103 based on the default value for “indicateWindowSize”.

In steps S 206 , TOE Driver 142 assigns certain anonymous buffers 107 to TOE Module 132 .

In step S 208 , initialization of adapter 114 is complete and system 100 is ready for network communication, i.e. to receive and send network data.

FIG. 3 illustrates a process for establishing and managing a network connection, according to one embodiment. The process starts in step S 300 when host system 102 using adapter 114 establishes a network connection.

In step S 302 , host system 102 notifies Adapter 114 via TOE Driver 142 to process the connection (i.e. the connection is offloaded).

In step S 304 , TOE Driver 142 creates the appropriate structures for managing the offloaded connection.

In step S 306 , TOE Driver 142 passes the “indicatewindowsize” to adapter 114 in general and to TOE module 132 in particular.

In step S 308 , TOE Driver 142 also indicates to Adapter 114 that no data has been received i.e. the “indicatedBytes” value is zero (based on counter 148 value).

In step S 310 , TOE Module 132 processes the network connection and notifies TOE Driver 142 upon its completion.

In step S 312 , TOE Driver 142 notifies host system 102 network stack that the connection has been completed and in step S 314 , the process ends.

FIGS. 4A-4H show process flow diagrams for moving network data to a host system, according to one embodiment.

The process starts in step S 400 , when TOE Driver 142 receives the “IndicateData” parameter from adapter 114 for a connection.

In step S 402 , TOE Driver 142 queues (or assigns) Anonymous Buffers 107 for the connection. In one embodiment, every connection has a context and data is queued for every connection.

In step S 404 , TOE Driver 142 updates counter 148 by an amount of data that it has received from adapter 114 at any given time. It is assumed that adapter 114 has previously received data from the network prior to passing it to TOE Driver 142 in step S 400 .

In step S 406 , TOE Driver 142 determines if application 138 has completed a given task at any given instance, application 138 has been notified of data availability, or if a named buffer from buffer pool 105 has been returned (i.e. cleared). If yes, then the process exits in step S 408 . Otherwise in step S 410 , TOE Driver 142 determines if a Named Buffer 109 is available from Application 138 for the connection. If a Named Buffer 109 is available, then in step S 412 TOE Driver 142 copies data from Anonymous Buffer 107 to Named Buffer 109 , otherwise the process moves to step S 428 , described below.

In step S 414 , TOE Driver 142 releases Anonymous Buffer 107 and in step S 416 , the released Anonymous Buffer 107 is returned to buffer pool 103 that is available for TOE Module 132 .

In step S 418 , TOE Driver 142 determines if a Named Buffer 109 for application 138 is full. If yes, then in step S 420 , the Named Buffer 109 is returned to buffer pool 105 . If Named Buffer 109 is not full the process moves to step S 426 , as described below.

In step S 422 , TOE Driver 142 checks if a Push Timer is running for the returned Named Buffer 109 . If yes, then in step S 424 , TOE Driver 142 stops the Push Timer, otherwise the process continues to step S 426 .

In step S 426 , TOE Driver 142 determines whether all the received data has been copied to a named Buffer 109 . If TOE Driver 142 determines that not all buffered data is copied to Named Buffer 109 (which means, that there is data but there are no Named Buffers 109 available), the process moves to step S 428 . If all the buffers are copied (which means, there are available buffers but there is no more data), the process moves to step S 456 , described below.

In step S 428 , TOE Driver 142 determines that Application 138 has been notified of data availability or Named Buffer 109 has been returned. If TOE Driver 142 determines that Application 138 has been notified, the process exits in step S 440 . If not, then in step S 430 , TOE Driver 142 notifies Application 138 of data availability.

In step S 432 , if application 138 provides a Return Code, then in step S 436 , TOE Driver 142 releases Anonymous Buffers 107 , otherwise the process continues in step S 434 .

In step S 434 , if TOE Driver 142 determines that Named Buffer 109 is available, then the process moves to step S 412 , otherwise the process exits in step S 440 .

In step S 438 , TOE Driver 142 releases Anonymous Buffer 107 to pool 103 making it available for TOE Module 132 .

In step S 442 , if TOE Driver 142 determines that there is more queued data, then in step S 444 , TOE Driver 142 checks if there is a Named Buffer 109 available for the queued data. If there is no queued data in step S 442 , then the process moves to step S 446 .

In step S 444 , if there is no Named Buffer 109 available then the process loops back to step S 428 . If a Named Buffer is available, then the process continues to step S 412 .

In step S 446 , TOE Driver 142 determines if “indicateWindow” is full, i.e. the amount of data that can be received has been received. This is inferred when “indicatedBytes”=“indicateWindowSize”. If yes, then in step S 448 , TOE Driver 142 resets counter 148 by setting “indicatedBytes”=0, otherwise the process exits in step S 454 .

In step S 450 , TOE Driver 142 determines if a Named Buffer 109 has been posted (or allocated) for TOE Module 132 , if yes, then in step S 454 , the process ends, otherwise in step S 452 , TOE Driver 142 issues an “indicateAcknowledge” parameter for TOE Module 132 . The “indicateAcknowledge” parameter instructs TOE Module 132 to open a new “indicateWindow” which enables TOE Module 132 to resume sending data to TOE Driver 142 .

In step S 456 , TOE Driver 142 determines whether “indicatedBytes”=“indicateWindowSize”. If yes, then in step 458 , TOE Driver 142 resets counter 148 and the process ends in step S 464 , otherwise the process moves to step S 460 .

In step S 460 , TOE Driver 142 determines if a Named Buffer 109 is partially full. If not, then the process ends in step S 464 , otherwise, TOE Driver 142 determines if a Named Buffer 109 is in Push Mode. If a Named Buffer 109 is not in a Push Mode, the process ends in step S 464 .

The term “Push Mode” as used herein applies to Named Buffers. A Named Buffer is determined to be in a Push Mode when an Application sets a PUSH MODE flag in an application programming interface (API) that is used to pass the Named Buffer to the TOE Driver 142 . When the Push Mode flag is set TOE Driver 142 and adapter 114 monitor received TCP segments that have been DMAed (using DMA engine 126 ) or copied to a Named buffer. If the TOE Driver 114 or adapter 114 detect that the TCP PUSH flag is set in a processed TCP Segment, TOE Driver 142 or adapter 114 return the Named Buffer within a PUSH Timer timeout period regardless of how much data has been placed into the Named Buffer. If the Push Mode flag is not set, then adapter 114 and TOE Driver 142 do not monitor the TCP Push flag in received TCP segments.

If the Named Buffer 109 is in a Push Mode, then in step S 466 , TOE Driver 142 determines if a push bit is set in a copied TCP segment. If the bit is set, then in step S 470 , the Named Buffer 109 is returned. If the push bit is not set, then in step S 468 , TOE Driver 142 determines if the push timer is running. If the Push Timer is not running, then the timer is started in step S 476 and the process ends in step S 478 . If the push timer is running then the process ends in step S 478 .

If the push timer is running in step S 472 , then the timer is stopped in step S 474 and the process ends in step S 478 . If the push timer is not running in step S 472 , then the process ends in step S 478 .

FIG. 4F shows a process flow diagram for handling Named Buffers, according to one embodiment. The process starts in step S 480 , when Application 138 posts (or allocates) a Named Buffer 109 for TOE Driver 142 . The term allocates or posts as used throughout this specification means that Named Buffer 109 is assigned to store data for a connection.

In step S 482 , TOE Driver 142 queues a Named Buffer 109 for TOE Module 132 for a connection.

In step S 484 , TOE Driver 142 determines if some Named Buffers have been submitted to TOE Module 132 . If yes, then the process continues to step S 412 , otherwise TOE Driver 142 provides a Named Buffer 109 for TOE Module 132 , starting from the head of a Named Buffer queue (or pool 103 ).

In step S 488 , if TOE Driver 142 determines that Named Buffer Size is not larger than “indicateWindowSize”, then the process moves to step S 412 , otherwise TOE Driver 142 in step S 490 , notifies TOE Module 132 of Named Buffer 109 's size and address where data can be placed.

In step S 492 , TOE Module 132 returns Named Buffer 109 to TOE Driver 142 .

In step S 494 , if TOE Driver 142 determines if the push timer is running. If yes, then in step S 496 , TOE Driver 142 stops the push timer. Once the push timer expires or is stopped, in step S 498 , Named Buffer 109 is returned and the process ends in step S 499 .

The foregoing adaptive aspects reduce latency because the “indicateWindow” parameter allows data to be buffered while application 138 is determining whether to accept data or post a buffer. This reduces latency for data availability to a socket application. This results in better application performance. If the application is transaction based and operates primarily with small amounts of data the foregoing embodiments significantly increase performance by reducing latency that is incurred on every transaction.

In another embodiment, the foregoing aspects also assist applications that operate primarily on large amounts of data by reducing latency and by providing simultaneous transfer of data by the TOE Driver (via data copy) and the TOE Device (via DMA engine 126 ).

Although the present invention has been described with reference to specific embodiments, these embodiments are illustrative only and not limiting. Many other applications and embodiments of the present invention will be apparent in light of this disclosure and the following claims.