AU2014200239A1 - System and method for multiple sender support in low latency fifo messaging using rdma - Google Patents

Abstract

SYSTEM AND METHOD FOR MULTIPLE SENDER SUPPORT IN LOW LATENCY FIFO MESSAGING USING RDMA System for transmitting and receiving multiple messages between multiple senders and at least one receiver in inter-process communication with low latencies and high throughput using Remote Direct Memory Access (RDMA) through InfiniBand verbs library is described. System comprises message queue library which allows messaging in lockless manner, Network Interface Card enabled for RDMA, and RDMA based remote sender processes to synchronize sub-queues. Transmitting system maps each remote sender process to First in First Out (FIFO) sub-queue of host node and corresponding FIFO sub queue of receiving node, assigns each user to a sub-queue, receives messages and arranges messages in FIFO sub-queue for each user and transmits messages from each FIFO sub queue of host node to corresponding FIFO sub-queue of receiver node. Receiving system receives messages and arranges in FIFO sub-queue for each user and reads messages from each FIFO sub-queue by round-robin technique in FIFO mode.

Description

P/00/011 Regulation 3.2 AUSTRALIA Patents Act 1990 ORIGINAL COMPLETE SPECIFICATION STANDARD PATENT Invention Title: "SYSTEM AND METHOD FOR MULTIPLE SENDER SUPPORT IN LOW LATENCY FIFO MESSAGING USING RDMA" The following statement is a full description of this invention, including the best method of performing it known to me/us: SYSTEM AND METHOD FOR MULTIPLE SENDER SUPPORT IN LOW LATENCY FIFO MESSAGING USING RDMA TECHNICAL FIELD [001] The present subject matter described herein, in general, relates to messaging systems, and more particularly to a system for supporting multiple senders in a low latency messaging using RDMA BACKGROUND [002] The important aspects of a messaging system are the latency and the throughput of messages.With the steady increase in network speeds, messaging systems are now expected to transfer millions of messages in few micro seconds between multiple publishers and subscribers. [003] Various messaging systems used till today are with locking mechanism and suffers from the slow processing speed. In addition to this they merely support messaging from single sender to single receiver. [004] One of the prior art process discloses a message bus in a messaging application for use in a data center provides a communication mechanism to provide low latency messaging with high throughput. However, this application provides mechanism to send messages from a single sender to a single receiver and there is no support for multiple senders sending messages simultaneously without knowing presence of other. [005] Further, another application discloses dynamic subscription and message routing between publishing nodes and subscriber nodes wherein subscribing module, publishing module and other modules are coupled together by a bus in a plurality of nodes. Although this method uses multiple queues, it fails to provide solution for lockless mechanism and simultaneous writing of multiple senders to the queues. [006] One of the prior art process discloses a lockless mechanism for low latency messaging system wherein a storage buffer is provided in a queue to store and retrieve messages pointed by a free pointing element. The storage buffers from the queue are 1 linked to each other in a circular linked list and messages are read accordingly. However, this invention does not provide any means for multiple publisher support. [007] Another prior art process provides a system and method for implementing messaging software using RDMA technology for remote messaging. But it does not provide any support for multiple publishers. Thus prior art processes failed to provide multiple publisher support in seamless manner. The prior art method where messages are stored in queue where multiple publishers are writing to the queue and when one of the writer is writing to the queue, others are locked for writing. Once first writer finishes his writing, there after another writer can write to the queue. This hampers the speed and performance of the messaging system. Therefore it is essential to enhance the speed of messaging irrespective of number of publishers simultaneously writing to the system. SUMMARY [008] This summary is provided to introduce aspects related to systems and methods for transmitting and receiving multiple messages hosted on at least one host node, in an inter-process communication and the aspects are further described below in the detailed description. This summary is not intended to identify essential features of the claimed subject matter nor is it intended for use in determining or limiting the scope of the claimed subject matter. [009] In one implementation, a system for transmitting multiple messages hosted on at least one host node, in an inter-process communication is described. The system comprises a processor, a Network Interface Card (NIC) coupled to the processor, wherein the Network Interface Card is enabled for remote direct memory access (RDMA) to send messages and a messaging library comprising one or more message-send and message receive functions allowing multiple messaging simultaneously in a lockless manner and a memory. The memory is coupled to the processor, wherein the processor is capable of executing a plurality of modules stored in the memory. The plurality of modules further comprises an organizing module, a mapping module and a transmitting module. The mapping module is configured to map each remote sender process to each FIFO sub-queue 2 associated with the host node and mapping each of the remote sender process with corresponding FIFO sub-queue associated with a receiving node by using one or more memory mapping files. The organizing module is configured to receive the messages from at least one user with the host node and to arrange the messages received from the users in the one or more First in First Out (FIFO) sub-queue associated with the host node wherein the FIFO sub-queue is dedicated for each user and is stored in a memory mapping file. The transmitting module is configured to transmit the messages from each FIFO sub-queue associated with the host node to the corresponding FIFO sub-queue associated with the receiving node using the corresponding remote sender process. [0010] In one implementation, a system for receiving multiple messages hosted on at least one host node, in an inter-process communication is described. The system comprises a processor, a network interface card (NIC) coupled to the processor, wherein the network interface card is enabled for remote direct memory access (RDMA) to receive messages, a messaging library comprising one or more message-send and message-receive functions allowing multiple messaging simultaneously in a lockless manner and a memory coupled to the processor, wherein the processor is capable of executing plurality of modules stored in the memory. The plurality of modules comprises a retrieving module and a reading module. The retrieving module is configured to receive the multiple messages transmitted from one or more host nodes with at least one user and to arrange the messages so received in a First in First Out (FIFO) sub-queue, wherein each FIFO sub queue is dedicated for each user and is stored in a memory mapping file. The reading module is configured to read the multiple messages from each of the FIFO sub-queue by using a round-robin technique in a FIFO mode. [0011] In one implementation, a method for transmitting multiple messages hosted on at least one host node, in an inter-process communication is described. The method comprising executing message-send and message-receive functions allowing multiple messaging simultaneously in a lockless manner and using remote direct memory access (RDMA) to transmit multiple user messages.The transmitting further comprises mapping each of a remote sender process to each of a FIFO sub-queue associated with the host node 3 and mapping each of the remote sender process with corresponding FIFO sub-queue associated with a receiving node by using one or more memory mapping files. The method further comprises receiving messages from at least one user and arranging the messages so received, from at least one user, in a First In First Out (FIFO) sub-queue associated with the host node, wherein each FIFO sub-queue is dedicated for each user and is stored in a memory mapping file. The method further comprises transmitting the messages from each FIFO sub-queue associated with the host node to the corresponding FIFO sub-queue associated with the receiving node using the corresponding remote sender process. The method steps of the arranging, the mapping and the transmitting are performed by means of the processor. [0012] In one implementation, a method for receiving multiple messages hosted on at least one host node, in an inter-process communication is described.The method comprises executing message-send and message-receive functions allowing multiple messaging simultaneously in a lockless manner and using remote direct memory access (RDMA) to receive multiple user messages. The method of receiving the messages further comprises receiving the multiple messages transmitted from one or more host nodes with at least one user and arranging the messages so received in a First in First Out (FIFO) sub queue, wherein the FIFO sub-queue is dedicated for each user and is stored in a memory mapping file. The method further comprises reading the multiple messages from each of the FIFO sub-queue by using a round-robin technique in a FIFO mode. The method steps of the receiving, the arranging and the reading are performed by means of the processor. BRIEF DESCRIPTION OF THE DRAWINGS [0013] The detailed description is described with reference to the accompanying figures. In the figures, the left-most digit(s) of a reference number identifies the figure in which the reference number first appears. The same numbers are used throughout the drawings to refer like features and components. 4 [0014] Figure 1 illustrates a network implementation of a system for transmitting and receiving multiple messages hosted on at least one host node, in an inter-process communication shown, in accordance with an embodiment of the present subject matter. [0015] Figure 2 illustrates the system for transmitting multiple messages hosted on at least one host node, in accordance with an embodiment of the present subject matter. [0016] Figure 3 illustrates the system for receiving multiple messages hosted on at least one host node, in accordance with an embodiment of the present subject matter. [0017] Figure 4 illustrates a configuration 1 for implementation of the present invention for transmitting and receiving the multiple messages hosted on at least one host node, in accordance with an exemplary embodiment of the present subject matter. [0018] Figure 5 illustrates a configuration 2 for implementation of the present invention for transmitting and receiving of the multiple messages hosted on at least one host node, in accordance with an exemplary embodiment of the present subject matter. [0019] Figure 6 illustrates a structure of the queue and sub-queue used in the present invention. [0020] Figure 7 illustrates a method for transmitting multiple messages hosted on at least one host node, in accordance with an embodiment of the present subject matter. [0021] Figure 8 illustrates a method for receiving multiple messages hosted on at least one host node, in accordance with an embodiment of the present subject matter. [0022] Figure 9 illustrates test setup for multiple publisher throughput tests. [0023] Figure 10 illustrates maximum throughput statistics for multiple publisher throughput test results. [0024] Figure 11 illustrates test setup for multiple publisher latency tests. [0025] Figure 12 illustrates test results for average round trip latency statistics for multiple publishers. 5 [0026] Figure 13 illustrates average round trip latency statistics at different throughput rates for multi publisher using RDMA over InfiniBand. DETAILED DESCRIPTION [0027] Systems and methods for transmitting and receiving multiple messages hosted on at least one host node, in an inter-process communication are disclosed. The systems and methods of the present invention provide support for multiple publishers to write simultaneously to an asynchronous lockless message queue whose remote subscriber is connected to the publishers using RDMA technology. The implementation mechanism of the present systems and methods for lockless messaging with multiple publishers comprises assigning each publisher with a dedicated sub-queue by the messaging system internally in a seamless manner that is without manual intervention giving the notion to all the publishers that they are writing to the same message queue. Further, at the time of the RDMA handshake process, the appropriate sub-queues at the publisher and subscriber ends are shared. The subscriber to the queue with multiple publishers will read from all the sub-queues present in the system in a round-robin fashion to get the messages in near FIFO fashion. According to the present invention, all the publishers write to their own sub-queues thus avoiding the need for locks and the subscriber receives the messages in the order in which each publisher inserts it into their own sub-queues. [0028] In accordance with an embodiment, the system and method for remote messaging running on RDMA supported network between two nodes are disclosed in an Indian Patent Application 1745/MUM/2011 assigned to the Applicant. The application 1745/MUM/2011 discloses memory mapping file hosted on first host node configured to synchronize static circular queue of messages with second memory mapped file hosted on the second host node; and at least one remote sender process running on the first host node asynchronously sending at least one batch of message along with corresponding RDMA work requests from the said queue to the second host node. The entire contents of the application 1745/MUM/2011 are incorporated as reference herein and are not repeated for the sake of brevity. 6 [0029] While aspects of described systems and methods for transmitting and receiving multiple messages hosted on at least one host node may be implemented in any number of different computing systems, environments, and/or configurations, the embodiments are described in the context of the following exemplary system. [0030] Referring now to Figure 1, a network implementation 100 of system 102 for transmitting multiple messages hosted on at least one host node is illustrated, in accordance with an embodiment of the present subject matter. In one embodiment, the system 102 receives multiple messages from a plurality of users in an inter process communication and arranges the messages so received. Further, the system 102 transmits the messages to the receiver. Further, a network implementation 100 of system 103 for receiving multiple messages hosted on at least one host node is illustrated, in accordance with an embodiment of the present subject matter. In one embodiment, the system 103 receives the multiple messages transmitted from one or more host nodes with at least one user. In another embodiment, the system 103 reads the multiple messages so received from one or more host nodes. [0031] Although the present subject matter is explained considering that the system 102 and the system 103 is implemented on one or more servers acting as a host node, it may be understood that the system 102 and the system 103 may also be implemented in a variety of computing systems, such as a laptop computer, a desktop computer, a notebook, a workstation, a mainframe computer, a server, a network server, and the like. It will be understood that the system 102 and the system 103may be accessed by multiple users through one or more user devices 104-1, 104-2... 104-N, collectively referred to as user 104 hereinafter, or applications residing on the user devices 104. Examples of the user devices 104 may include, but are not limited to, a portable computer, a personal digital assistant, a handheld device, and a workstation. The user devices 104 are communicatively coupled to the system 102 and the system 103through a network 106. [0032] In one implementation, the network 106 may be a wireless network, a wired network or a combination thereof The network 106 can be implemented as one of the different types of networks, such as intranet, local area network (LAN), wide area 7 network (WAN), the internet, and the like. The network 106 may either be a dedicated network or a shared network. The shared network represents an association of the different types of networks that use a variety of protocols, for example, Hypertext Transfer Protocol (HTTP), Transmission Control Protocol/Internet Protocol (TCP/IP), Wireless Application Protocol (WAP), InfiniBand Protocol, Ethernet Protocol and the like to communicate with one another. Further the network 106 may include a variety of network devices, including routers, bridges, servers, computing devices, storage devices, and the like. [0033] Referring now to Figure 2, the system 102 is illustrated in accordance with an embodiment of the present subject matter. In one embodiment, the system 102 may include at least one processor 202, an input/output (I/O) interface 204, a network interface card (NIC) 206 coupled to the processor and a memory 208. The network interface card is enabled for remote direct memory access (RDMA). Further, the network interface card may be enabled for remote direct memory access (RDMA) to transmit or to receive messages. The at least one processor 202 may be implemented as one or more microprocessors, microcomputers, microcontrollers, digital signal processors, central processing units, state machines, logic circuitries, and/or any devices that manipulate signals based on operational instructions. Among other capabilities, the at least one processor 202 is configured to fetch and execute computer-readable instructions stored in the memory 208. [0034] The I/O interface 204 may include a variety of software and hardware interfaces, for example, a web interface, a graphical user interface, and the like. The I/O interface 204 may allow the system 102 to interact with a user directly or through the client devices 104. Further, the I/O interface 204 may enable the system 102 to communicate with other computing devices, such as web servers and external data servers (not shown). The I/O interface 204 can facilitate multiple communications within a wide variety of networks and protocol types, including wired networks, for example, LAN, cable, etc., and wireless networks, such as WLAN, cellular, or satellite. The I/O interface 204 may include one or more ports for connecting a number of devices to one another or to another server. 8 [0035] The memory 208 may include any computer-readable medium known in the art including, for example, volatile memory, such as static random access memory (SRAM) and dynamic random access memory (DRAM), and/or non-volatile memory, such as read only memory (ROM), erasable programmable ROM, flash memories, hard disks, optical disks, and magnetic tapes. The memory 208 may include modules 210 and data 212. [0036] The modules 210 include routines, programs, objects, components, data structures, etc., which perform particular tasks or implement particular abstract data types. In one implementation, the modules 210 may include an organizing module 214, a mapping module 216, a transmitting module 218 and other modules 220. The other modules 220 may include programs or coded instructions that supplement applications and functions of the system 102. [0037] The data 212, amongst other things, serves as a repository for storing data processed, received, and generated by one or more of the modules 208. The data 212 may also include a system database 222, a messaging library 224, and other data 226. The other data 130 may include data generated as a result of the execution of one or more modules in the other module 218. [0038] Referring now to Figure 3, the system 103 is illustrated in accordance with an embodiment of the present subject matter. In one embodiment, the system 103 may include at least one processor 302, an input/output (I/O) interface 304, a network interface card (NIC) 306 coupled to the processor and a memory 308. The at least one processor 302 may be implemented as one or more microprocessors, microcomputers, microcontrollers, digital signal processors, central processing units, state machines, logic circuitries, and/or any devices that manipulate signals based on operational instructions. Among other capabilities, the at least one processor 302 is configured to fetch and execute computer-readable instructions stored in the memory 308. [0039] The I/O interface 304 may include a variety of software and hardware interfaces, for example, a web interface, a graphical user interface, and the like. The I/O 9 interface 304 may allow the system 103 to interact with a user directly or through the client devices 104. Further, the I/O interface 304 may enable the system 103 to communicate with other computing devices, such as web servers and external data servers (not shown). The I/O interface 304 can facilitate multiple communications within a wide variety of networks and protocol types, including wired networks, for example, LAN, cable, etc., and wireless networks, such as WLAN, cellular, or satellite. The I/O interface 304 may include one or more ports for connecting a number of devices to one another or to another server. [0040] The memory 308 may include any computer-readable medium known in the art including, for example, volatile memory, such as static random access memory (SRAM) and dynamic random access memory (DRAM), and/or non-volatile memory, such as read only memory (ROM), erasable programmable ROM, flash memories, hard disks, optical disks, and magnetic tapes. The memory 308 may include modules 310 and data 312. [0041] The modules 310 include routines, programs, objects, components, data structures, etc., which perform particular tasks or implement particular abstract data types. In one implementation, the modules 310 may include a retrieving module (314), a reading module (316) and other modules 318. The other modules 318 may include programs or coded instructions that supplement applications and functions of the system 103. [0042] The data 312, amongst other things, serves as a repository for storing data processed, received, and generated by one or more of the modules 208. The data 312 may also include a system database 320, a messaging library (322) and other data 132. The other data 130 may include data generated as a result of the execution of one or more modules in the other modules 318. [0043] In one implementation, at first, a user may use the client device 104 to access the system 102 via the I/O interface 204. The user may register them using the I/O interface 204 in order to use the system 102. The working of the system 102 may be explained in detail in Figures 2 and 3 explained below. The system 102may be used for 10 transmitting multiple messages hosted on at least one host node, in an inter-process communication. [0044] In accordance with an embodiment, referring to figure 2, the system 102 comprises the network interface card (NIC) coupled to the processor to enable for remote direct memory access (RDMA). Further, the network interface card may be enabled for remote direct memory access (RDMA) to transmit or to receive messages. The Network Interface Card is capable of executing RDMA commands from remote host. The RDMA is supported on the Network Interface Card (NIC) using iWARP or InfiniBand network to connect at least one host node acting as sender to the host node acting as a receiver.The RDMA protocol is developed to move data from the memory of one computer directly into the memory of another computer with minimal involvement of the processors on the two computers. The RDMA protocol includes information that allows a system to place transferred data directly into its final memory destination thereby doing away with interim copies. This mode of transfer is thus the most efficient mode of communication between systems. RDMA is applicable to InfiniBand and Ethernet. Ethernet or InfiniBand switch that connects user host node and receiver host node. [0045] InfiniBand is a high-performance, low-latency alternative to Ethernet. InfiniBand architecture uses a switched-fabric, channel based design that adapts well to distributed computing environments. Like Ethernet, InfiniBand uses multilayer processing to transfer data between nodes. Each InfiniBand node contains a Host Channel Adapter (HCA) that connects to an InfiniBand network through a bidirectional serial link. [0046] Since RDMA minimizes the role of the processor and transfers data directly to the final memory destination, the establishment of the connection is the most important part. The two processes that are transferring data register a memory location in which the data will be transferred. The size of the memory location should be the same on both the ends of the connection. [0047] The system 102 further comprises the messaging library (224). The messaging library comprises one or more message-send and message-receive functions. 11 Further, the message-send and message-receive functions may be linked and invoked by the message transmitting and receiving application processes. According to an exemplary embodiment, the RDMA protocol is implemented over InfiniBand architecture, the InfiniBand verbs library is used to transmit the messages and receive acknowledgements for the transmit of the messages. [0048] Referring to figure 2, the system 102 comprises the mapping module (214) configured to map each of a remote sender process to each of a FIFO sub-queue associated with the host node. The mapping module is configured to map remote sender process with corresponding FIFO sub-queue associated with a receiving node by using one or more memory mapping files. Remote Sender processes run on the host node acting as a transmitter or a sender. Number of remote sender processes in the system 102, may exceed the number of users by one. The main remote sender process creates remote sender process threads for each sub-queue and initializes their state. Thus there may be number of remote sender processes exceeding the number of users by one. Remote Sender processes are configured for watching incoming messages and updating the memory mapping file from the transmitting host node to the memory mapping files on the receiving host node via RDMA. Remote Sender processes are configured for updating the sub-queues from memory mapping file of the transmitting host node to the sub-queues of the memory mapping files on the receiving host node via RDIA. [0049] In accordance with an embodiment, referring to figure 2, the system 102 comprises the organizing module (216) configured to arrange messages received from a plurality of users. The organizing module is also configured to arrange the messages so received in one or more First in First out (FIFO) sub-queue associated with the host node. Further, the organizing module has each FIFO sub-queue dedicated for each user and is stored in a memory mapping file. The organizing module is configured to arrange the messages so received from each user in the FIFO sub-queue dedicated for each user. The user can be a sender. The user can be a publisher. The organizing module upon receiving the messages invokes a messaging library. The memory mapping file is stored in the memory 208. The number of memory mapping files created exceeds the number of users 12 by one. One memory mapping file for the main queue and one each for the sub-queues created per user. The memory mapping file contains the FIFO sub-queue which is used for sending and receiving messages. [0050] Referring to figure 2, the system 102 comprises the transmitting module configured to transmit the messages from each FIFO sub-queue associated with the host node to the corresponding each FIFO sub-queue associated with the receiving node using the corresponding remote sender processes. There can be one or more users present in the system. There can be each sub-queue dedicated for each user. One or more FIFO sub queues are associated with a main queue where size of the main queue is equal to or greater than the sum of the size of all the sub-queues present in the system.The remote sender processes transmits the message data using the InfiniBand verbs library. [0051] In accordance with another embodiment, the system 102 further comprises one or more receiving node configured to receive the messages from one or more sender host nodes. The system 102 may comprise Ethernet or InfiniBand switch that connects transmitter / sender host node and the receiver host node. [0052] In one implementation, a user may use the client device 104 to access the system 103 via the I/O interface 304. The user may register them using the I/O interface 304 in order to use the system 103. The working of the system 103 may be explained in detail in Figures 4 and 5 explained below. The system 103 may be used for receiving multiple messages hosted on at least one host node, in an inter-process communication. [0053] In accordance with an embodiment, referring to figure 3, the system 103 comprises the network interface card (NIC) coupled to the processor to enable for remote direct memory access (RDMA). Further, the network interface card may be enabled for remote direct memory access (RDMA) to transmit or to receive messages. The Network Interface Card is capable of executing RDMA commands from local or remote host. The RDMA is supported on the Network Interface Card (NIC) using iWARP or InfiniBand network to connect at least one host node acting as sender to the host node acting as a receiver. 13 [0054] The system 103 further comprises the messaging library (322). The messaging library comprises one or more message-send and message-receive functions. Further, the message-send and message-receive functions may be linked and invoked by the message transmitting and receiving application processes. According to an exemplary embodiment, the RDMA protocol is implemented over InfiniBand architecture, the InfiniBand verbs library is used to receive the messages and provide acknowledgements for the receiving of the messages. [0055] In accordance with an embodiment, referring to figure 3, the system 103 for receiving multiple messages hosted on at least one host node, in an inter-process communication is described. Referring to figure 3, the system 103 comprises the retrieving module 314 configured to receive the multiple messages transmitted from one or more host nodes with at least one user. The retrieving module 314 is further configured to arrange the messages so received in a First in First Out (FIFO) sub-queue wherein each FIFO sub-queue is dedicated for each user and is stored in a memory mapping file. The user can be a sender. The user can be a publisher. The retrieving module is further configured to invoke a messaging library when it receives the messages. The receiving of the messages is carried out by one or more receiving processes using the InfiniBand verbs library stored in the messaging library 314. The memory mapping file is stored in the memory 308. The number of memory mapping files created exceeds the number of users by one. One memory mapped file for the main queue and one each for the sub-queues created per user. The memory mapping file contains one or more FIFO sub-queue which is used for sending and receiving messages. One or more FIFO sub-queues stored in memory mapping file are associated with a main queue where size of the main queue is equal to or greater than the sum of the size of all the sub-queues present in the system. [0056] Referring to figure 3, the system 103 comprises the reading module 316 configured to read the multiple messages from each of the FIFO sub-queue by using a round-robin technique in a FIFO mode. The reading module facilitates the user of the system to read the messages sent by the sender. The sender can be a publisher and a user can be receiver or can be a subscriber. The system 103 further comprises one or more 14 transmitting node configured to transmit the messages from one or more host nodes. The system 103 further may comprise Ethernet or InfiniBand switch that connects sender host node and receiver host node. [0057] In accordance with an embodiment of the present invention, referring to figure 4 and figure 5 implementation of the present invention is explained. Traditionally the low latency messaging systems support single publishers and one or more subscribers. However, when the messaging system has to support multiple writers, the framework of the present invention implements the same in a seamless and lockless manner. The system 102 of the present invention comprises the main queue which is divided into as many sub queues as the maximum number of users/senders provided at the time of system configuration. The sum of the sizes of the sub-queues does not exceed that of the main queue. All the other characteristics of the sub-queue are inherited from the main queue. [0058] As each user connects to the main queue, a sub-queue is assigned to each user to write. The writing or sending of messages to the sub-queue occurs seamlessly just like the user/sender experience of writing to the main queue. Thus the interface for the user/sender remains unchanged. The system on the receiving node for the reader of the queue executes a round robin read on all the sub-queues and read the data and the interface for the reader also remains unchanged. [0059] Referring to figure 4, according to an exemplary embodiment, the implementation of the present invention in configuration 1 is explained by way of an example. Let us consider that there are the senders Sender 1, Sender 2 and Sender 3 using the system 102 installed on the same host node. The senders are sending the messages to the same system installed on a Server 1. A configuration file is created on the Server 1 which specifies the number of senders to the queue. By way of an example, the minimum no of senders may be 2 the maximum no of senders may be 10. The maximum number of senders can be configured as per requirements at the time of development. For explanation purpose, let us take the number of senders as 3. A main queue is created by specifying the size of messages, the size of the queue, and the port on which the receiver should listen for the incoming messages from the senders. By way of an example, let us consider the queue 15 is created for a size of 300 messages. Since there are 3 senders, 3 smaller sub-queues are created each of size 100 messages. The IP Address of the receiver node is specified along with the port number on which the receiver is waiting for the sender to connect and receive the messages. [0060] Further, the implementation of the invention on the receiving node is explained accordance with an exemplary embodiment. Let us consider a receiver on the receiving Server 2 wherein system 103 is installed. A configuration file is created on the Server 2 which specifies the number of senders to the queue. By way of an example, the minimum no of senders may be 2 the maximum no of senders may be 10. As explained above, let us take the number of senders as 3 same as above. A main queue is created by specifying the size of messages, the size of the queue, and the port on which the receiver should listen for the incoming messages from the senders. By way of an example, let us consider the queue is created for a size of 300 messages same as explained above. Since there are 3 senders, 3 smaller sub-queues are created each of size 100 messages. [0061] Referring to figure 4, implementation of the present invention over RDMA supported network is explained. The implementation may involve two processes in the RDIA transfer, the Remote Sender Process (RS) and the Receiver Process (R). Remote Sender Processes (RS) and the Receiver Process can share their respective queue files as part of the memory registration process. The Sender Process (S) can write to the local memory mapped queue file. The Remote Sender Process (RS) may transfer the data using the InfiniBand verbs library (ibvpost sendo. [0062] The Receiver Process on the receiver host node may wait until it receives the updates from the remote sender process. After receiving of the messages further the receiver process may read the messages and updates the remote sender process about the message it has read (using ibvpost sendo. The Sender Process is thus notified that the receiver process has finished reading the messages. The Remote Sender process may be used to optimize the communication between the sender and receiver by clubbing work requests together and minimizing the steps required to finish a transfer. 16 [0063] Referring to figure 5, there can be multiple senders using the system 102 installed on different host nodes such as servers. Referring to figure 5, by way of an example there may be four users using system for transmitting multiple messages installed on different host nodes such as Server 1 and Server 2 respectively for Sender 1 and Sender 2 using Server 1 and Sender 3 and Sender 4 using the Server 2. Remote sender processes RS1, RS2 and RS3 and RS4 may set up RDMA connections directly with the receiver process S. The receiver process may take part in memory registration and may set up the RDMA connection with all the remote sender processes that connect for example RS1 and RS2 as shown in figure 5. [0064] By way of an example, referring to figure 5, in configuration 2, a configuration file is created on the Server 1 and Server 2 which specifies the number of senders to the queue. By way of an example, the minimum no of senders may be 2 the maximum no of senders may be 10. A main queue is created by specifying the size of messages, the size of the queue, and the port on which the receiver should listen for the incoming messages from the senders. By way of an example, let us consider, as explained above, there are 2 senders on each server, so a main queue of size 200 messages is created on each server having two sub-queues each of size 100 messages. Similar structures of main queue holding sub-queues are created on the receiver node that is on the Server 3 to receive and arrange the messages. This is important as the two sub-queues on either end have to be of the same size for RDMA transfer to succeed. As part of the queue creation process the IP Address of the receiver is specified along with the port number on which the receiver is waiting for senders to connect. Further, it is to be noted that TP Address has to belong to a NIC card which supports RDMA transfer. Further, as explained above the transfer of the messages takes place over RDMA supported network as explained in above paragraphs 0061 and 0062. [0065] Referring to figure 4 and figure 5, in accordance with an exemplary embodiment, the detailed explanation of the implementation of transfer of messages is provided. The process execution is started first at the receiver end before any of the senders or remote sender process execution is started. The receiver side process knows the 17 number of senders that are going to connect to it for its queue (as mentioned in paragraph 0060 above). The receiver initializes at some state and waits for each sender's remote sender process to connect with it. The receiver listens on the port specified at queue creation. On the sender's system, a remote sender process is started for the multiple senders queue. The remote sender process checks the number of senders on the system for the queue and spawns that many processes. [0066] So referring to figure 4, for Configuration 1, on the sender's Server 1, a remote sender process is started, which in turn starts of three remote sender processes. Each remote sender process picks up a sub-queue and connects with the receiver process. In the handshaking process, the memory addresses of the data area of each of the sub queues on either end are exchanged. The header area of the queue on either end is also exchanged. Referring to figure 6, explains the structure of the queue and sub-queue used in the present invention. Referring to figure 5, for configuration 2, the remote sender process started starts of two remote sender processes on each sender's server so total four remote sender processes started for four senders. [0067] Referring to figure 4, the sender's processes S1 and S2 are started with the name of the queue and the number of messages they want to send or publish. As each sender is started, it is assigned a sub-queue. The sender starts inserting messages into the sub-queue through respective sender processes, for example sender process S1 for Sender 1 and Sender process S2 for Sender 2. Further, the remote sender process picks up the available messages in respective assigned sub-queue and transfers it to the corresponding sub-queue on the receiver's system. As each transfer happens, the receiver process is interrupted to complete the data transfer. The receiver process then reads from all the sub queues in a round robin fashion to read the messages from all the senders. If more senders than the number configured attempt to write to the queue, an error is returned and the extra sender exits. If there are lesser senders, then the remote sender process for the sub-queue continues to wait for a sender to start. The receiver checks each sub-queue and moves on if there are no messages in it. When the receiver reads the messages from a sub-queue, it updates the header area of the sub-queue denoting how many messages are read. That 18 information is in turn transferred to the sender's system in the corresponding sub-queue's header area. This way the sender knows that its messages have been read and it continues writing fresh messages in its sub-queue. Once a sender exits, the sub-queue becomes available for assignment to another sender. [0068] Still referring to figure 4 and figure 5, in accordance with an embodiment of the present invention, the lockless mechanism is explained for multiple senders sending messages using system 102 and system 103 over the RDMA network in an inter process communication. The lockless mechanism is provided by way of allowing the plurality of senders to write messages to their assigned sub-queues without knowing presence of each other. The writing or sending of messages to the sub-queue occurs seamlessly just as the sender experience of writing to the main queue. Thus the interface for the sender remains unchanged. The system 103 on the receiving node for the receiver of the queue executes a round robin read on all the sub-queues to read the data and the interface for the receiver also remains unchanged. Thus simultaneous writing of multiple senders without providing any locking to the queue and simultaneous reading of the messages on the receiver end is achieved through the system and method provided by the present invention. [0069] Referring to figure 6, the concept of main queue and sub-queue is described in accordance with an embodiment. At the time of queue creation, the configuration file is checked to identify whether the queue is a multi-senders queue. The configuration file also mentions the maximum number of senders that the queue may support. Further, at the time of creation of the queue, the maximum size of messages, the length of the queue, the IP Address and port of the receiver among other parameters is also mentioned. [0070] The main queue is created with above mentioned parameters. The number of senders is extracted from the configuration file and the same numbers of sub-queues are created. The sub-queues inherit all the properties of the main queue but the length of the sub-queue is calculated as described below: Length of sub-queue = length of main queue/number of senders 19 By way of an example, when the main queue is of length 300 and there are 3 senders, the length of each sub-queue is 100. By way of an example, the main queue is called MULTWRITE, and there are three senders, the sub-queues are named as MULT_WRITE_0, MULTWRITE_1 and MULTWRITE_2. [0071] Further, the main queue also maintains an array of structures in which the name of each sub-queue and its status (whether currently allocated to a sender or unallocated) is stored. By maintaining the above mentioned information in the main queue, track of sub-queue is maintained and the sub-queue can be allocated to a sender as it joins the system and opens the queue. Further, when a sender finishes inserting messages and disconnects from the queue, the appropriate sub-queue is marked as unallocated. The sub-queues also contain the ID of the main queue for ease of access. [0072] When a sender wants to use the queue, it opens the queue MULTWRITE. The organizing module of the system 102 looks up for the sub-queue that is available and if available, organizing module of the system 102 allocates that to the sender. For instance in the example above if four senders attempt to open the main queue, sender 1 is allocated IULTWRITE_0, sender 2 is allocated MULTWRITE_1, sender 3 is allocated IULTWRITE_2 and the fourth sender gets an error message as all three sub-queues are allocated. All messages written by the senders are inserted into its sub-queue. Thus the main queue is not used for messages it is primarily used for book keeping. After the sender finishes inserting its messages, it disconnects from the framework. At that time, the sub-queue it was using is marked as unallocated. [0073] According to an embodiment of the present invention, referring to figure 6, the structure of main queue and sub-queue is described. The structure of the main queue consists of a header portion. The header portion contains the queue name, queue ID, message size, queue size and other characteristics. The header portion further consists of sub structure called as RH is the Receiver Header. Receiver Header contains the data pointing element and the deletion counter. The data pointing element points to the next message to be read by the receiver. The deletion counter is the number of elements or messages that has been read by the receiver. The sub structure RH is important for each 20 sub-queue, as this memory is shared with the receiver and the receiver directly updates this sub structure in the sender's structure through RDMA ibvpost send() commands. The receiver updates all the RH structures of each of the sub-queues belonging to different senders. Further, SQ stands for sub-queue. This is the name of the sub-queue created under the main queue. 0 stands for occupied. This is true if the sub-queue has been assigned to a publisher. NO stands for not occupied. This is false if the sub-queue has not been allocated to a publisher. [0074] Still referring to figure 6, the fields mentioned in above paragraph belong to the structure of main queue and sub-queue. The queue comprises an array of structures. The size of the array is equal to the maximum number of senders to be supported. The array of structures is filled in the main queue and remains empty in the sub-queue. The reason is that the main queue is used to keep track of the sub-queues and of which sub queue can be allocated to a sender. The "w" portion of the sub-queue contains the free pointing element and the insertion counter. These two variables are updated by the remote sender process and are used to notify the receiver about the number of messages that have been transferred. The free pointing element points to the message location from where the remote sender has transferred the messages that is written by the sender and the insertion counter is the number of messages transferred by the remote sender from the queue. The "w" structure is primarily a sub structure of the main queue header. It is copied by the respective sender and sent along with the messages to the receiver to denote how many messages are available to be read. The "DATA" portion of the sub-queue indicated in figure 6 is the message inserted by the sender. Each sender inserts its messages into its own sub-queue. [0075] Still referring figure 6, in accordance with an exemplary embodiment, by way of an example the main queue is called as QTRANSFER. Let us assume that there are three publishers or senders to the main queue, so three sub-queues are created and they are called as Q TRANSFER_0, Q TRANSFER_1 and Q TRANSFER_2. The user issues a command to create a queue by the name Q TRANSFER. As part of the call, the main queue is created and then the sub-queues are created. Each sub-queue is one third the 21 length of the main queue. All other properties of the main queue are inherited by the sub queues. [0076] In accordance with an embodiment, the detail information related to layout and working of main queue (queue), the memory mapping file, the remote sender process and the transmission and receiving of messages over RDMA supported network can be referred from the application 1745/MUM/2012. The layout and working of the sub-queue is similar to the main queue (queue). [0077] According to an embodiment of the present invention, the system 102 and system 103 comprises multiple senders and a single receiver using the framework with RDIA. For each sender process, a remote sender process is associated. By way of present example, for three senders on a single system, there will be seven active processes including three sender's processes, three remote sender processes and one receiver process. Further, the system also comprises a dormant remote sender process. According to the system 102, 103 and method 700 and 800 of the present invention, the receiver process is started first. Further, the Remote sender process is started on the sender's server. Let us call this as the main remote sender process. The main remote sender process starts of remote sender processes for each sender on the server. Thus if the number of senders to be supported on the server is three, three remote sender processes are started by the main remote sender process. The main remote sender process then becomes dormant. The three remote sender processes are each allocated a sub-queue for which it will transfer the messages. Then each remote sender process contacts the receiver process, creates channels to communicate with each other and exchanges the memory locations of the respective sub-queues. Thus the receiver is contacted by the three remote sender processes and communication channels are sets up and memory locations are shared with all three remote sender processes. [0078] Further, each remote sender process waits for the sender's processes to start inserting messages into the respective sub-queues by checking the position of the RDMA free pointing element and the free pointing element and the value of the RDMA insertion counter and the insertion counter of the respective sub-queues. If the values are equal, then 22 no new message has been inserted. The sender processes insert the messages and updates the RDMA free pointing element and the RDMA insertion counter of each respective sub queues. When a change is accounted in the RDMA free pointing element and the RDMA insertion counter, the remote sender processes constructs their respective data to be transferred i.e. the messages inserted by the senders in the respective sub-queues and sends it to the respective receiver sub-queues using the ibvpost send() call from the message library (224) . The process of transmission is described in detail in patent application 1745/MUM/201 1. Further, the remote sender processes updates the values of the free pointing element and the insertion counter of the respective sub-queues as part of the process and sends the updated values as part of the data transfer to the respective receiver sub-queues in the "w" structure of the sub-queues as mentioned in Figure 6. [0079] According to another embodiment of the present invention, the receiver process on the system 103, checks the free pointing element and insertion counter of the "w" structure at its current position in each sub-queue of the sender's system 102 and compares it with the data pointing element and the deletion counter of the sub-queue of its own end i.e. system 103. When any of the remote sender processes initiates RDMA data transfer, the receiver process is interrupted to complete the transfer. The new transferred data contains the updated values of the free pointer element and insertion counter in the "w" structure of the sub-queue. Accounting the change in the values of the free pointer element and insertion counter, the receiver process reads the messages and updates the data pointing element and the deletion counter in the receiver structure (denoted as RH in figure 6) of the sub-queue. RDMA data transfer is initiated to the sender's server to update the above values in the sender's sub-queue structure. Further details about the receiving of the message data can be referred from the previously file application 1745/MUM/201 1. [0080] Further, each of the senders is notified about the number of messages read by the receiver when each of the senders receives the RDMA data transfer of the receiver structure RH. The senders can then proceed to write further messages. The senders wait if the position next to the one pointed by the RDMA free pointer is equal to the data pointing element and if the RDMA insertion counter and the deletion counter is equal in value. 23 This indicates that the receiver has not yet read the messages previously inserted and there are no more payload locations for the senders to fill with messages. Once the data pointing element and the deletion counter is updated by the receiver and transferred to the sender's structure via RDIA, the sender can proceed to write. Each sender checks the RDIA free pointing element and RDMA insertion counter of respective sub-queue and compares their value with the data pointing element and deletion counter in the RH structure of their respective sub-queue. Once the data pointing element and the deletion counter is updated by the receiver and transferred to each of the corresponding sender's structure via RDMA, the senders can proceed to write. [0081] Referring now to Figure 7, a method 700 for transmitting multiple messages hosted on at least one host node, in an inter-process communication is shown, in accordance with an embodiment of the present subject matter. The method 700 may be described in the general context of computer executable instructions. Generally, computer executable instructions can include routines, programs, objects, components, data structures, procedures, modules, functions, etc., that perform particular functions or implement particular abstract data types. The method 700 may also be practiced in a distributed computing environment where functions are performed by remote processing devices that are linked through a communications network. In a distributed computing environment, computer executable instructions may be located in both local and remote computer storage media, including memory storage devices. [0082] The order in which the method 700 is described is not intended to be construed as a limitation, and any number of the described method blocks can be combined in any order to implement the method 700 or alternate methods. Additionally, individual blocks may be deleted from the method 700 without departing from the spirit and scope of the subject matter described herein. Furthermore, the method can be implemented in any suitable hardware, software, firmware, or combination thereof However, for ease of explanation, in the embodiments described below, the method 700 may be considered to be implemented in the above described media system 102. 24 [0083] Referring to figure 7, according to an embodiment of the present invention, a method (700) for transmitting multiple messages hosted on at least one host node, in an inter-process communication is described. The method 700 comprises executing message send and message-receive functions allowing multiple messaging simultaneously in a lockless manner by using remote direct memory access (RDMA) to transmit multiple user messages. In one implementation, the message-send and message-receive functions may be stored in messaging library and may be invoked by the modules stored in the memory and are executed by the processor. Remote direct memory access (RDMA) is supported by the Network Interface Card. [0084] In step 702, each of a remote sender process may be mapped to each of a FIFO sub-queue associated with the host node and in step 704, each remote sender process may be mapped with corresponding FIFO sub-queue associated with a receiving node by using one or more memory mapping files. In one implementation, the mapping of each of the remote sender process to each of the FIFO sub-queue associated with the host node and mapping of each remote sender process with corresponding FIFO sub-queue associated with the receiving node by using one or more memory mapping files may be carried out by the mapping module 214. [0085] In step 706, messages received from at least one user may be arranged in the one or more First in First Out (FIFO) sub-queue associated with the host node wherein each FIFO sub-queue may be dedicated for each user and may be stored in a memory mapping file. In one implementation, the receiving of the messages and arranging of the messages in First in First Out (FIFO) sub-queue associated with the host node may be carried out by the organizing module 216. [0086] In step 708, the messages from each FIFO sub-queue associated with the host node may be transmitted to the corresponding FIFO sub-queue associated with the receiver node by using the corresponding remote sender process. In one implementation, the transmission of the messages from each FIFO sub-queue associated with the host node to the corresponding each FIFO sub-queue associated with the receiver node using the corresponding remote sender process may be carried out by the transmitting module 218. 25 The method 700 steps such as 702, 704, 706 and 708 comprising the arranging, the mapping and the transmitting is performed by means of the processor 202. The number of memory mapping files created may exceed the number users by one and the number of remote sender processes may exceed the number of users by one. The method 700 is executed on the RDMA supported network by at least one network interface card (NIC). [0087] Referring now to Figure 8, a method 800 for receiving multiple messages hosted on at least one host node, in an inter-process communication is shown, in accordance with an embodiment of the present subject matter. The method 800 may be described in the general context of computer executable instructions. Generally, computer executable instructions can include routines, programs, objects, components, data structures, procedures, modules, functions, etc., that perform particular functions or implement particular abstract data types. The method 800 may also be practiced in a distributed computing environment where functions are performed by remote processing devices that are linked through a communications network. In a distributed computing environment, computer executable instructions may be located in both local and remote computer storage media, including memory storage devices. [0088] The order in which the method 800 is described is not intended to be construed as a limitation, and any number of the described method blocks can be combined in any order to implement the method 800 or alternate methods. Additionally, individual blocks may be deleted from the method 800 without departing from the spirit and scope of the subject matter described herein. Furthermore, the method can be implemented in any suitable hardware, software, firmware, or combination thereof However, for ease of explanation, in the embodiments described below, the method 800 may be considered to be implemented in the above described media system 103. [0089] Referring to figure 8, according to an embodiment of the present invention, a method (800) for receiving multiple messages hosted on at least one host node, in an inter-process communication is described. The method 800 comprises executing message send and message-receive functions allowing multiple messaging simultaneously in a lockless manner by using remote direct memory access (RDMA) to receive multiple user 26 messages. In one implementation, the message-send and message-receive functions may be stored in messaging library and may be invoked by the modules stored in the memory and are executed by the processor. Remote direct memory access (RDMA) is supported by the Network Interface Card. [0090] In step 802, the multiple messages transmitted from one or more host nodes with at least one user may be received and in step 804 the messages so received may be arranged in a First in First Out (FIFO) sub-queue, wherein the FIFO sub-queue may be dedicated for each user and may be stored in a memory mapping file. In one implementation, the receiving of the multiple messages transmitted from one or more host nodes with at least one user and arranging of the messages so received in a First in First Out (FIFO) sub-queue, wherein the FIFO sub-queue may be dedicated for each user may be carried out by the retrieving module 314. [0091] In step 806, the multiple messages from each of the FIFO sub-queue may be read by using a round-robin technique in a FIFO mode. In one implementation, the reading of the multiple messages from each of the FIFO sub-queue by using a round-robin technique in a FIFO mode may be carried out by the reading module 316. The method 800 steps 802, 804, 806 comprising the receiving, the arranging and the reading are performed by means of the processor.The number of memory mapping files created may exceed the number users by one. The method 800 may be executed on the RDMA supported network by at least one network interface card (NIC). [0092] In accordance with an exemplary embodiment, the performance results of the system 102 and system 103 for multiple sender support in low latency FIFO messaging is provided. The system is called Custom Built Queue (CBQ) The throughput test results are provided. Throughput is the maximum speed with which messages can be exchanged between the publishers and subscribers or senders and receivers. Referring to figure 9, the test set up for the throughput test is shown. By way of an example, the Custom Built Queue (CBQ) has been implemented in C. Since the trading application used for throughput testing was implemented in Java, JNI (Java Native Interface) was used to call the native C code functions from the Java application. Referring to figure 10, 27 the maximum throughput performance of the Multiple Publisher Custom Built Queue (CBQ) using both Java and C application programs for three publishers is provided. The publishers can be senders. The subscribers can be receivers. [0093] According to an exemplary embodiment of the present invention and by way of an example, in the throughput test, the publishers at Server 1 insert messages of size 512 bytes into the Multiple Publisher Custom Built Queue (CBQ). There is no think time between simultaneous messages. The system 103 at subscriber end on server 2 calculates the time to receive one million messages and calculates the throughput in messages per second (msgs/sec). The size of the Custom Built Queue (CBQ) is 30 messages. Publishers implemented in Java use ByteArray messages and those in C use string messages. The tests were conducted with all processes (3 publishers + 1 subscriber) on the same system (IPC), and with the publishers and subscribers (3 publishers + 1 subscriber) on different servers as shown in figure 9, connected by a 1 Gbps link (TCP IG), a 10 Gbps link (TCP-10G) and an RDMA link. Referring to figure 10, the throughput test results are provided. [0094] In accordance with an exemplary embodiment, the latency test results of the system 102 and system 103 for multiple sender support in low latency FIFO messaging is provided. Latency is the time taken for a message to travel to a receiver and the response to come back to the sender. Referring to figure 11, the test setup for the latency test is shown. By way of an example, three publishers on server 1 insert messages into the Multi Publisher Custom Built Queue (CBQ). Publisher can be a sender of messages. A timestamp is embedded into the message just before inserting the message. The subscriber on server 2 reads the message from the Multi Publisher CBQ and in turn publishes it on a point to point CBQ connected to a subscriber on server 1. This is a loopback. The subscriber can be a receiver of the messages. The subscriber on server 1 reads the message and calculates the difference in time between when the message was sent and when the message was received. This is the latency of the message. Referring to figure 12, an average value of the round trip latency for one million messages is calculated 28 and reported in figure 12 for all four deployments. The messages are sent at a throughput of 512 byte messages @ 50000 msgs/sec. The CBQ size is 30 as stated before. [0095] In accordance with another exemplary embodiment, various RDMA throughput and corresponding latency statistics is described. Referring to figure 13, average round trip latency statistics for the Multiple Publisher CBQ using RDMA over InfiniBand by way of system 102 and system 103 is shown. The latency test was conducted for different publisher throughput rates - 20,000 msgs/sec, 50,000 msgs/sec, 100,000 msgs/sec and maximum throughput. Max throughput indicates the test in which the publishers have zero think time between consecutive messages. For the max throughput test, the average value of the throughput of an individual publisher is mentioned in the graph. Measurements have been taken for 1, 3, 6 and 9 publishers. The publishers were distributed across two servers in the tests for six and nine publishers. The publishers have been implemented in C. The publishers can be the senders of the messages. [0096] Still referring to figure 13, it is observed that the average round trip latency for throughput rates of 20,000 msgs/sec and 50,000 msgs/sec is about 12 pis and is almost constant as the number of subscribers increase. As per the shown metric from figure 13, the use of RDMA over InfiniBand as the transport mechanism in the Multi Sender CBQ for the trading application is supporting and highly efficient. [0097] According to an exemplary embodiment, the hardware configuration of the servers of the system 102 and system 103 on which the throughput, latency and RDMA 29 tests were conducted is provided below. S TF" EnEsa ........................................... . . . . . . . . .. . . C e r e s

................

d. .... C owesgdMjjgg cnlge [009] Acodn aceead Etemoie h et wokCrnomto msagusing RDMAh dh hogptaecibed inM thepesets ieniondce has seoviea technica ad a te i but h n o f. k-7 ~ ~? 098 odingmutpl anendemsplar emoimen itetorkoC ard formation usingwhichtwen hogtlatenorcse and in RDM t weast tondes.poiedblw PIQVp,2 O5GT IB DR ______________ IGigqE] ( b0:1U Ethernet 10 htf~ NE2Q: 10G~b Accelerated Ethernet Adapt er (5%6RWRNICI ADVANTAGES [00991 System and Method for multiple senders support in low latency FTFO messaging using RDMA described in the present invention has several technical advantages including but not limited to the realization of *Provides multiple senders support for an inter-processes communication between at least two processes running on at least two nodes. 30 * Provides lockless mechanism for multiple senders to send messages for an inter-processes communication between at least two processes running on at least two nodes. * Supports high throughput message rates for multiple simultaneous users. * Delivers messages with low average latency Requires minimum infrastructure and easy to deploy on the existing systems. The written description describes the subject matter herein to enable any person skilled in the art to make and use the embodiments of the invention. The scope of the subject matter embodiments is defined by the claims and may include other modifications that occur to those skilled in the art. Such other modifications are intended to be within the scope of the claims if they have similar elements that do not differ from the literal language of the claims or if they include equivalent elements with insubstantial differences from the literal language of the claims. 31

Claims (24)

1. A system for transmitting multiple messages hosted on at least one host node, in an inter-process communication, the system comprising: a processor; a Network Interface Card (NIC) coupled to the processor, wherein the Network Interface Card is enabled for remote direct memory access (RDMA) to send messages; a message library comprising one or more message-send and message receive functions allowing multiple messaging simultaneously in a lockless manner; and a memory coupled to the processor, wherein the processor is capable of executing a plurality of modules stored in the memory, the plurality of modules comprising: a mapping module configured to map each of a remote sender process to each of a FIFO sub-queue associated with the host node and mapping each remote sender process with corresponding FIFO sub-queue associated with a receiving node by using one or more memory mapping files; an organizing module configured to arrange messages received from at least one user in the one or more First In First Out (FIFO) sub queue associated with the host node; wherein the FIFO sub-queue is dedicated for each user and is stored in a memory mapping file; and a transmitting module configured to transmit the messages from each FIFO sub-queue associated with the host node to the corresponding each FIFO sub-queue associated with the receiving node using the corresponding remote sender process. 32

2. The system of claim 1, wherein one or more FIFO sub-queues are associated with a main queue where size of the main queue is equal to or greater than the sum of the size of all the sub-queues present in the system.

3. The system of claim 1, wherein the organizing module upon receiving the messages invokes a messaging library.

4. The system of claim 1, wherein the user can be a sender.

5. The system of claim 1, wherein the sender process transmits the message data using the InfiniBand verbs library.

6. The system of claim 1, wherein the number of memory mapping files created exceeds the number of users by one and number of remote sender processes exceeds the number of users by one.

7. The system of claim 1, wherein the RDMA is supported on the Network Interface Card (NIC) using iWARP or InfiniBand network to connect at least one host node acting as sender to the host node acting as a receiver.

8. The system of claim 1, further comprises one or more receiving node configured to receive the messages from one or more sender host nodes.

9. The system of claim 1, further comprises Ethernet or InfiniBand switch that connects user host node and receiver host node.

10. A system for receiving multiple messages hosted on at least one host node, in an inter-process communication, the system comprising: a processor; a network interface card (NIC) coupled to the processor, wherein the network interface card is enabled for remote direct memory access (RDMA) to receive messages; 33 a messaging library comprising one or more message-send and message receive functions allowing multiple messaging simultaneously in a lockless manner; and a memory coupled to the processor, wherein the processor is capable of executing plurality of modules stored in the memory, the plurality of modules comprising: a retrieving module configured to receive the multiple messages transmitted from one or more host nodes with at least one user, the messages so received are arranged in a First In First Out (FIFO) sub-queue, wherein each FIFO sub-queue is dedicated for each user and is stored in a memory mapping file; and a reading module configured to read the multiple messages from each of the FIFO sub-queue by using a round-robin technique in a FIFO mode.

11. The system of claim 10, wherein one or more FIFO sub-queues are associated with a main queue where size of the main queue is equal to or greater than the sum of the size of all the sub-queues present in the system.

12. The system of claim 10, wherein the retrieving module upon receiving the messages invokes a messaging library.

13. The system of claim 10, wherein the receiving of the messages is carried out by receiving processes using the InfiniBand verbs library.

14. The system of claim 10, wherein the number of memory mapping files created exceeds the number of users by one.

15. The system of claim 10, wherein the RDMA is supported on the NIC using iWARP or InfiniBand network to connect at least one host node acting as sender to the host node acting as a receiver. 34

16. The system of claim 10, further comprises one or more transmitting node configured to transmit the messages from one or more host nodes.

17. The system of claim 10, further comprises Ethernet or InfiniBand switch that connects sender host node and receiver host node.

18. A method for transmitting multiple messages hosted on at least one host node, in an inter-process communication, the method comprising: executing message-send and message-receive functions allowing multiple messaging simultaneously in a lockless manner; transmitting multiple user messages using remote direct memory access (RDMA); the transmitting further comprises: mapping each of a remote sender process to each of a FIFO sub queue associated with the host node and mapping each remote sender process with corresponding FIFO sub-queue associated with a receiving node by using one or more memory mapping files; and arranging messages received from at least one user, in the one or more First In First Out (FIFO) sub-queue associated with the host node, wherein each FIFO sub-queue is dedicated for each user and is stored in a memory mapping file, and transmitting the messages from each FIFO sub-queue associated with the host node to the corresponding each FIFO sub-queue associated with the receiving node using the corresponding remote sender process; wherein the mapping, the arranging and the transmitting is performed by means of the processor.

19. The method of claim 18, wherein the number of memory mapping files created exceeds the number users by one and the number of remote sender processes exceeds the number of users by one. 35

20. The method of claim 18, wherein the method is executed on the RDMA supported network by at least one network interface card (NIC).

21. A method for receiving multiple messages hosted on at least one host node, in an inter-process communication, the method comprising: executing message-send and message-receive functions allowing multiple messaging simultaneously in a lockless manner; receiving multiple user messages using remote direct memory access (RDMA); the receiving further comprises: receiving the multiple messages transmitted from one or more host nodes with at least one user, the messages so received are arranged in a First In First Out (FIFO) sub-queue, wherein the FIFO sub-queue is dedicated for each user and is stored in a memory mapping file; and reading the multiple messages from each of the FIFO sub-queue by using a round-robin technique in a FIFO mode; wherein the receiving, the arranging and the reading is performed by means of the processor.

22. The method of claim 21, wherein the number of memory mapping files created exceeds the number of users by one.

23. The method of claim 21, wherein the user can be a sender.

24. The method of claim 21, wherein the method is executed on the RDMA supported network by at least one network interface card (NIC). 36