New! View global litigation for patent families

US20100161855A1 - Lightweight input/output protocol - Google Patents

Lightweight input/output protocol Download PDF

Info

Publication number
US20100161855A1
US20100161855A1 US12717306 US71730610A US2010161855A1 US 20100161855 A1 US20100161855 A1 US 20100161855A1 US 12717306 US12717306 US 12717306 US 71730610 A US71730610 A US 71730610A US 2010161855 A1 US2010161855 A1 US 2010161855A1
Authority
US
Grant status
Application
Patent type
Prior art keywords
server
client
lwio
set
invention
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Abandoned
Application number
US12717306
Inventor
Ahmed H. Mohamed
Anthony F. Voellm
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Microsoft Technology Licensing LLC
Original Assignee
Microsoft Corp
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date

Links

Images

Classifications

    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04LTRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L67/00Network-specific arrangements or communication protocols supporting networked applications
    • H04L67/10Network-specific arrangements or communication protocols supporting networked applications in which an application is distributed across nodes in the network
    • H04L67/1097Network-specific arrangements or communication protocols supporting networked applications in which an application is distributed across nodes in the network for distributed storage of data in a network, e.g. network file system [NFS], transport mechanisms for storage area networks [SAN] or network attached storage [NAS]
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04LTRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L69/00Application independent communication protocol aspects or techniques in packet data networks
    • H04L69/10Streamlined, light-weight or high-speed protocols, e.g. express transfer protocol [XTP] or byte stream
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04LTRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L69/00Application independent communication protocol aspects or techniques in packet data networks
    • H04L69/26Special purpose or proprietary protocols or architectures

Abstract

A method and system for offloading I/O processing from a first computer to a second computer, using RDMA-capable network interconnects, are disclosed. The method and system include a client on the first computer communicating over an RDMA connection to a server on the second computer by way of a lightweight input/output (LWIO) protocol. The protocol generally comprises a network discovery phase followed by an I/O processing phase. During the discovery phase, the client and server determine a minimal list of shared RDMA-capable providers. During the I/O processing phase, the client posts I/O requests for offloading to the second machine over a mutually-authenticated RDMA channel. The I/O model is asymmetric, with read operations being implemented using RDMA and write operations being implemented using normal sends. Read and write requests may be completed in polling mode and in interrupt mode. Buffers are managed by way of a credit mechanism.

Description

    CROSS-REFERENCE TO RELATED APPLICATIONS
  • [0001]
    This patent application is a divisional of U.S. patent application Ser. No. 10/749,959 entitled “Lightweight Input/Output Protocol,” filed Dec. 31, 2003, the disclosure of which is expressly incorporated herein, in its entirety, by reference.
  • TECHNICAL FIELD
  • [0002]
    The present invention relates generally to systems and methods of remote file access, and more particularly to techniques for offloading input/output processing using Remote Direct Memory Access (RDMA).
  • BACKGROUND
  • [0003]
    In computing environments it is generally desirable to conserve scarce CPU resources. For some such environments, such as networks of application server nodes, such conservation is especially critical. As networks become faster, they make greater demands on CPUs to process packets and perform I/O operations, resulting in slower application performance. This is particularly detrimental for inherently I/O-intensive applications like databases.
  • [0004]
    One approach to remedying this problem is to offload excessive I/O and network processing from the CPU. In a networked environment, using distributed file systems and transport protocols like NFS or SMB/CIFS, it is possible to send I/O requests from a local machine to a remote machine. However, it is not necessarily the case that the local machine will achieve significant processing economies using such approaches.
  • [0005]
    In the single machine context, I/O processing burdens can be alleviated by offloading I/O tasks to a direct memory access (DMA) controller. Remote Direct Memory Access (RDMA) technology is a more recently-developed extension of DMA for multiple networked computers. RDMA allows data to be moved between memory buffers on two communicating machines equipped with RDMA-capable network interface cards (NICs) without having to involve the CPU and operating system of either the source or the destination machine. RDMA can be used to offload I/O processing to a remote machine, thereby enabling the local machine to reclaim CPU cycles for applications. RDMA has been exploited in high-speed, high-bandwidth interconnect technologies, such as the Virtual Interface Architecture (VIA), InfiniBand, and iWarp. These interconnects are particularly designed for high-reliability network connections between clusters of server nodes within a data center or other local file-sharing environment.
  • [0006]
    Protocols defining the communication between a local offloading node and a remote machine must be designed in order for the capabilities associated with RDMA technology to be fully utilized and their benefits effectively achieved. Therefore, there is a need for the lightweight input/output (LWIO) protocol of the present invention.
  • SUMMARY OF THE INVENTION
  • [0007]
    In accordance with one aspect of the present invention, a system for offloading an I/O task from a first computer to a second computer is provided. The system includes a client running on the first computer and a server running on the second computer. The system further includes one or more RDMA channels linking the first computer and the second computer. The client and server communicate in accordance with an LWIO protocol comprising a network discovery phase and an I/O processing phase. The LWIO protocol is used in association with another network protocol, such as SMB/CIFS, leveraging the security and authentication infrastructure of the second protocol. In order to provide a better security model, the I/O model in the protocol is asymmetric: reads are implemented using RDMA, while writes are implemented using send operations.
  • [0008]
    In accordance with another aspect of the present invention, a method for offloading an I/O task from a first computer to a second computer is provided. The method takes advantage of common RDMA-capable communication devices on the two computers and is associated with a lightweight input/output (LWIO) client-server protocol. The protocol generally comprises a discovery phase followed by an I/O processing phase. During the discovery phase, the client and server determine a minimal list of shared RDMA-capable providers. During the I/O processing phase, the client posts I/O requests for offloading to the second machine.
  • [0009]
    During the discovery phase, the client initially obtains a server request resume key from the server. The client then opens a pipe to the server, over which the client sends a negotiate request containing a list of RDMA-capable providers on the first machine. The server sends a negotiate response over the pipe containing a list of available providers on the second machine that match providers on the first machine. The client then creates an RDMA connection to the server over a shared provider. The client and the server mutually authenticate the new connection. The client then registers one or more files for use with the server.
  • [0010]
    I/O processing request messages include a close message, a cancel message, a read message, a write message, a vectored read message, and a vectored write message. The protocol features an asymmetric I/O model for security reasons. Read data is sent to the client using RDMA write operations, while writes are completed using ordinary sends. Read and write requests can be specified by the client to be completed by the server in polling mode or in interrupt mode. If the client indicates that the completion should not be in polling mode, the server completes the I/O processing request by sending a status block to the first computer by way of RDMA transfer. If the client indicates that the completion should be in polling mode, the client may request that it be woken up by the server upon completion of the I/O by way of an interrupt request message.
  • [0011]
    In accordance with another aspect of the present invention, a method for managing buffers in an I/O offload protocol is provided. The method involves the use of a buffer credit mechanism. A server-client credit transaction comprises a three-way handshake initiated and completed by the server. The server sends a delta credit message to the client, including an information field set to a number of credits. If the number is a negative number −N, the client must give up N credits.
  • [0012]
    Other aspects of the invention include the above-mentioned features embodied on computer-readable media as computer program products and data structures.
  • BRIEF DESCRIPTION OF THE DRAWINGS
  • [0013]
    While the appended claims set forth the features of the present invention with particularity, the invention, together with its objects and advantages, may be best understood from the following detailed description taken in conjunction with the accompanying drawings, of which:
  • [0014]
    FIG. 1 is a diagram generally illustrating an exemplary client-server computing environment involving two computers capable of communicating by way of RDMA transfer, within which aspects of the present invention can be incorporated;
  • [0015]
    FIG. 2 is a flow diagram generally illustrating initial steps taken in the discovery phase of the LWIO protocol in accordance with an embodiment of the invention;
  • [0016]
    FIG. 3 is a diagram generally illustrating a representation of an exemplary server request resume key in accordance with an embodiment of the invention;
  • [0017]
    FIG. 4A is a diagram generally illustrating a representation of an exemplary client negotiate request message in accordance with an embodiment of the invention;
  • [0018]
    FIG. 4B is a diagram generally illustrating a representation of an exemplary server negotiate response in accordance with an embodiment of the invention;
  • [0019]
    FIG. 5 is a flow diagram generally illustrating additional steps taken in the discovery phase of the LWIO protocol in accordance with an embodiment of the invention;
  • [0020]
    FIG. 6A is a diagram generally illustrating a representation of an exemplary client authenticate request message in accordance with an embodiment of the invention;
  • [0021]
    FIG. 6B is a diagram generally illustrating a representation of an exemplary server authenticate response in accordance with an embodiment of the invention;
  • [0022]
    FIG. 6C is a diagram generally illustrating a representation of an exemplary server status response completing authentication in accordance with an embodiment of the invention;
  • [0023]
    FIG. 7A is a diagram generally illustrating a representation of an exemplary client register file message in accordance with an embodiment of the invention;
  • [0024]
    FIG. 7B is a diagram generally illustrating a representation of an exemplary server status response completing file registration in accordance with an embodiment of the invention;
  • [0025]
    FIG. 8 is a flow diagram generally illustrating steps taken with respect to completion of an I/O request in polling mode and in non-polling mode, in accordance with an embodiment of the invention;
  • [0026]
    FIG. 9A is a diagram generally illustrating a representation of an exemplary client interrupt request message in accordance with an embodiment of the invention;
  • [0027]
    FIG. 9B is a diagram generally illustrating a representation of an exemplary server status response completing an interrupt request in accordance with an embodiment of the invention;
  • [0028]
    FIG. 10 is a flow diagram generally illustrating steps taken with respect to a server-client credit transaction in accordance with an embodiment of the invention;
  • [0029]
    FIG. 11A is a diagram generally illustrating a representation of an exemplary server delta credit message in accordance with an embodiment of the invention;
  • [0030]
    FIG. 11B is a diagram generally illustrating a representation of an exemplary client-to-server credit message in accordance with an embodiment of the invention;
  • [0031]
    FIG. 11C is a diagram generally illustrating a representation of an exemplary server status response completing a client-server credit transaction in accordance with an embodiment of the invention;
  • [0032]
    FIG. 12A is a diagram generally illustrating a representation of an exemplary client close request message in accordance with an embodiment of the invention;
  • [0033]
    FIG. 12B is a diagram generally illustrating a representation of an exemplary server status response completing a close request in accordance with an embodiment of the invention;
  • [0034]
    FIG. 13A is a diagram generally illustrating a representation of an exemplary client cancel request message in accordance with an embodiment of the invention;
  • [0035]
    FIG. 13B is a diagram generally illustrating a representation of an exemplary server status response completing a cancel request in accordance with an embodiment of the invention;
  • [0036]
    FIG. 14A is a diagram generally illustrating a representation of an exemplary client read request message in the non-polling mode case, in accordance with an embodiment of the invention;
  • [0037]
    FIG. 14B is a diagram generally illustrating a representation of an exemplary server status response completing a read request in the non-polling mode case, in accordance with an embodiment of the invention;
  • [0038]
    FIG. 14C is a diagram generally illustrating a representation of an exemplary client read request message in the polling mode case, in accordance with an embodiment of the invention;
  • [0039]
    FIG. 14D is a diagram generally illustrating a representation of an exemplary server I/O status block completing a read request in the polling mode case, in accordance with an embodiment of the invention;
  • [0040]
    FIG. 15A is a diagram generally illustrating a representation of an exemplary client write request message in the non-polling mode case, in accordance with an embodiment of the invention;
  • [0041]
    FIG. 15B is a diagram generally illustrating a representation of an exemplary server status response completing a write request in the non-polling mode case, in accordance with an embodiment of the invention;
  • [0042]
    FIG. 15C is a diagram generally illustrating a representation of an exemplary client write request message in the polling mode case, in accordance with an embodiment of the invention;
  • [0043]
    FIG. 15D is a diagram generally illustrating a representation of an exemplary server I/O status block completing a write request in the polling mode case, in accordance with an embodiment of the invention;
  • [0044]
    FIG. 16A is a diagram generally illustrating a representation of an exemplary client vectored read request message in the non-polling mode case, in accordance with an embodiment of the invention;
  • [0045]
    FIG. 16B is a diagram generally illustrating a representation of an exemplary server status response completing a vectored read request in the non-polling mode case, in accordance with an embodiment of the invention;
  • [0046]
    FIG. 16C is a diagram generally illustrating a representation of an exemplary client vectored read request message in the polling mode case, in accordance with an embodiment of the invention;
  • [0047]
    FIG. 16D is a diagram generally illustrating a representation of an exemplary server I/O status block completing a vectored read request in the polling mode case, in accordance with an embodiment of the invention;
  • [0048]
    FIG. 17A is a diagram generally illustrating a representation of an exemplary client vectored write request message in the non-polling mode, non-collapsed case, in accordance with an embodiment of the invention;
  • [0049]
    FIG. 17B is a diagram generally illustrating a representation of an exemplary client vectored write request message in the non-polling mode, collapsed case, in accordance with an embodiment of the invention;
  • [0050]
    FIG. 17C is a diagram generally illustrating a representation of an exemplary client vectored write request message in the polling mode, collapsed case, in accordance with an embodiment of the invention;
  • [0051]
    FIG. 17D is a diagram generally illustrating a representation of an exemplary server status response completing a vectored write request in the non-polling mode case, in accordance with an embodiment of the invention; and
  • [0052]
    FIG. 17E is a diagram generally illustrating a representation of an exemplary server I/O status block completing a vectored write request in the polling mode case, in accordance with an embodiment of the invention.
  • DETAILED DESCRIPTION
  • [0053]
    Certain embodiments of the present invention are discussed below with reference to FIGS. 1-17E. However, those skilled in the art will readily appreciate that the detailed description given herein with respect to these figures is for illustrative purposes, and that the invention extends beyond these embodiments.
  • [0054]
    FIG. 1 is a schematic diagram generally illustrating certain features of a representative networked client/server environment within which aspects of the present invention may be incorporated. Depicted in FIG. 1 are two computer machines, labeled Host A 101 and Host B 121. While the invention may be practiced in an environment involving computers of many different types and uses, in one representative scenario Host A 101 functions as an application server machine charged with I/O-intensive work, such as a database server.
  • [0055]
    Each of Host A 101 and Host B 121 include a number of network interface cards (NICs) 109, 111, 113, 133, 135, 137 allowing for networked data communication from one machine to the other. Among these NICs are NICs 109, 111, 135, 137 permitting RDMA data transfer. As illustrated, a non-RDMA network link 119 and an RDMA channel 117 are present between the two hosts 101, 121.
  • [0056]
    Executing on Host A 101 is an LWIO client application 103, associated with an application responsible for processing I/O tasks which interacts with kernel-mode I/O read/write services 105. The LWIO client 103 is used to offload I/O processing from Host A 101 to Host B 121. On Host B 121 an LWIO server 123 is executing. In accordance with the LWIO protocol described herein, the LWIO client 103 communicates with the LWIO server 123. The LWIO client 103 and the LWIO server 123 make use of posted buffers 107, 127, enabling file-associated data to be transferred directly by way of the RDMA channel connection 117. By way of LWIO protocol messages, read and write tasks are offloaded to Host B 121. The server 123 passes on I/O requests to the file system 129, which serves as the interface to the hard disk 131.
  • [0057]
    Typically, two kinds of messages are associated with an RDMA connection 117. The first type is an ordinary network send/receive, generating an interrupt at the destination machine. The second type is an RDMA read/write, in which memory space on the remote machine is accessed without the aid of the remote CPU and thus without having to generate an interrupt. The remote CPU determines the memory regions that are exposed for RDMA but typically is unaware of when an RDMA operation is performed.
  • [0058]
    In an embodiment of the invention described herein, the LWIO protocol is used in association with another network protocol, such as SMB or CIFS, in order to take advantage of the existing security and authentication infrastructure of the other protocol. This helps to minimize the overhead of the LWIO protocol. As illustrated in FIG. 1, the LWIO server 123 on Host B 121 operates above an SMB server 125. An SMB client (not shown) similarly runs on Host A 101 and interacts with the LWIO client application 103.
  • [0059]
    The LWIO protocol comprises two phases: a discovery phase followed by an I/O phase. In data structures associated with an embodiment described herein, data sizes are as follows:
  • [0000]
    BYTE unsigned 8-bit integer
    CHAR 8-bit ASCII character
    UINT16 unsigned 16-bit integer
    UINT32 unsigned 32-bit integer
    UINT64 unsigned 64-bit integer
    INT16 signed 16-bit integer
    INT32 signed 32-bit integer
    INT64 signed 64-bit integer
    WCHAR 16-bit Unicode character
    PVOID32 32-bit pointer
    PVOID64 64-bit pointer
  • [0060]
    FIG. 2 illustrates steps taken in the discovery phase of the LWIO protocol in an embodiment of the invention. With respect to the host on which the LWIO server is executing, at step 201 the LWIO server registers with the SMB/CIFS server running on that host machine. In accordance with this registration, at step 203 the SMB/CIFS server notifies a SMB/CIFS client running on a remote host that the LWIO server is available. At step 205 the LWIO client requests a server request resume key. The resume key is an authentication mechanism that has been disclosed in another application having the same assignee as the present application, “Method and System for Accessing a File (Resume Key),” U.S. patent application Ser. No. ______, filed on Oct. 24, 2003, which is hereby incorporated herein in its entirety by reference.
  • [0061]
    At step 207 the LWIO server passes the server request resume key back to the client. In an embodiment of the invention the server request resume key has the following structure:
  • [0000]
    typedef struct _SRV_RESUME_KEY {
      UINT64 ResumeKey;
      UINT64 Timestamp;
      UINT64 Pid;
    } SRV_RESUME_KEY, *PSRV_RESUME_KEY;
    typedef struct _SRV_REQUEST_RESUME_KEY {
      SRV_RESUME_KEY Key;
      UINT16 ContextLength;
      BYTE Context[1];
    } SRV_REQUEST_RESUME_KEY,
    *PSRV_REQUEST_RESUME_KEY;

    FIG. 3 provides an illustrative representation of the server request resume key 219. ResumeKey 221, Timestamp 223, and Pid 225 are generated on the server and are opaque to the client. Context 229 is an array containing a UNC name that is used by the LWIO client to contact the server. ContextLength 227 is the number of bytes in Context 229.
  • Network Discovery
  • [0062]
    When the client application receives the server request resume key 219, it retrieves the server UNC name from the Context field 229. Returning to FIG. 2, at step 209 the client opens a pipe to the LWIO server. The pipe is used for automatic discovery of RDMA-capable devices that are available in the network, in a manner described further below. This is an important and useful feature of the present invention; address resolution mechanisms like ARP are generally absent from VIA networks and similar networks.
  • [0063]
    The client next queries the server for a list of its RDMA-capable devices (“providers”) that are available for use with the LWIO protocol. The querying is accomplished by way of a negotiate request, which the client constructs and sends to the server over the newly-opened pipe at step 211. In an embodiment of the invention, the negotiate request has the following structure:
  • [0000]
    typedef struct {
      LWIO_CONTROL_HEADER;
      WCHAR   ClientName[LWIO_MAX_HOST_NAME];
      UUID   Key;
      UINT16   ResponseLength;
      UINT16   ProviderCount;
      LwioAddressBlk_t   ProviderList[1];
    } LwioNegotiateRequest_t;
    typedef struct {
      CHAR ProtocolId[4];
      UINT32 RevId;
      UINT16 Opcode;
      UINT16 Length;
    } LWIO_CONTROL_HEADER;
    typedef struct _GUID {
      UINT32 Data1;
      UINT16 Data2;
      UINT16 Data3;
      BYTE Data4[8];
    } GUID, UUID;
    typedef struct {
    WCHAR Name[LWIO_MAX_PROVIDER_NAME];
    UINT16 InstanceCount;
    LWIO_NET_ADDRESS InstanceTable[1];
    } LwioAddressBlk_t;
    typedef struct _LWIO_NET_ADDRESS {
    UINT16 HostAddressLen;
    UINT16 DiscriminatorLen;
    BYTE HostAddressFollowedByDiscriminator[1];
    } LWIO_NET_ADDRESS;
  • [0064]
    FIG. 4A provides an illustrative representation of the negotiate request packet 231 in an embodiment of the invention. The negotiate request includes a control header 233, a fixed-length Unicode client name field 235, a client UUID 237 used as a key, a local buffer size 239 for receiving a response, and the list of providers 241. In the control header 233, the ProtocolId ‘LWIO’ 243 is stored as the first four bytes of the header.
  • [0000]
    RevId 245 holds a currently defined value 0x1001, LWIO_REV_ID. Opcode 247 holds a currently defined value 0xfe, LWIO_CONTROL_OPCODE_NEGOTIATE. Length 249 is the size in bytes of the complete packet to be sent to the server, including all opcode-specific data.
  • [0065]
    ClientName 235 is used by the server to identify the client. Key 237 is used in a subsequent network-specific authentication procedure, as described below. ResponseLength 239 is the size of the buffer for receiving a negotiate response from the server, as described below. ProviderCount 251 is the number of providers associated with the client machine and about which the client is informing the server. The provider list 241 contains the list of ProviderCount providers.
  • [0066]
    In an element of the provider list 241, Name 253 is the name of the provider. In order for compatible networks to be detected, the client and the server should preferably use the same name for the same provider. InstanceCount 255 is the number of devices of a particular provider type. The instance table 257 is a table of network/discriminator pairs, in which a pair serves to describe, in a device-specific way, how to form a remote connection. HostAddressLen 259 is the length of the network-specific host address 263. DiscriminatorLen 261 is the length of the network-specific discriminator 265. Following these length fields are the HostAddressLen bytes of the host address 263 and the DiscriminatorLen bytes of the discriminator 265.
  • [0067]
    Returning to FIG. 2, having received the negotiate request with the client's list of providers, at step 213 the server determines which RDMA-capable communication devices it has in common with the client. At step 215 the server sends a negotiate response to the client over the pipe, including a list of shared providers. In an embodiment of the invention, the negotiate response has the following structure:
  • [0000]
    typedef struct {
      LWIO_CONTROL_HEADER;
      WCHAR SrvName[LWIO_MAX_HOST_NAME];
      UUID Key;
      UINT16 ProviderCount;
      LwioAddressBlk_t ProviderList[1];
    } LwioNegotiateResponse_t;
  • [0068]
    FIG. 4B provides an illustrative representation of the negotiate response 267 in an embodiment of the invention. The control header 269 is as in the negotiate request, except that Length 271 now reflects the size of the response message 267. SrvName 273 holds the name of the server. Key 275 is a server-generated GUID for use by the client. As explained further below, the client sends the Key back to the server in an authenticate request over a new connection using one of the common communication devices. ProviderCount 277 is the number of providers in the provider list 279. The provider list 279 contains a list of providers common to the server and the client. There is no guarantee that the client can actually connect to these providers.
  • [0069]
    Returning to FIG. 2, at this point the server and the client have shared communication device information, and the minimal list of common providers has been determined. At step 217 the client creates one or more RDMA connections to the LWIO server over one or more of the shared devices. In an embodiment of the invention, as described herein, the following opcodes are defined for client-to-server communication:
  • [0000]
    #define LWIO_OPCODE_READ 0x0
    #define LWIO_OPCODE_WRITE 0x1
    #define LWIO_OPCODE_VEC_READ 0x2
    #define LWIO_OPCODE_VEC_WRITE 0x3
    #define LWIO_OPCODE_CLOSE 0x4
    #define LWIO_OPCODE_CANCEL 0x5
    #define LWIO_OPCODE_AUTH 0x6
    #define LWIO_OPCODE_REGISTER 0x7
    #define LWIO_OPCODE_CREDIT 0x8
    #define LWIO_OPCODE_INTERRUPT 0x9

    The following defined flags are used as modifiers in client-to-server communication:
  • [0000]
    #define LWIO_HDR_FLAG_INTERRUPT 0x80
    #define LWIO_HDR_FLAG_CONTROL 0x40
    #define LWIO_HDR_FLAG_COLLAPSE_IO 0x20

    The corresponding client-to-server messages in the LWIO protocol feature a common header structure. The common header has the following format in an embodiment of the invention:
  • [0000]
    typedef struct {
      UINT32 Length;
      union {
       UINT32 Status;
       struct {
        BYTE Opcode;
        BYTE Flags;
        BYTE Credits;
        BYTE Marker;
       };
       };
       struct {
       UINT16 Fid;
       UINT16 Sequence;
       UINT32 Tid;
       };
       UINT64 Offset;
       // data buffer block
       struct {
       PVOID64 DataVa;
       union {
        UINT32 DataMh;
        struct {
         UINT16 NumPages;
         UINT16 PageSize;
        } Vec;
       };
       };
      // io status block
      union {
       struct {
        UINT32 IosMh;
        PVOID64 IosVa;
       };
       struct {
        UINT32 ImmediateCookie;
        UINT64 Cookie;
       };
      };
    } LWIO_COMMON_HEADER;
  • Connection Authentication
  • [0070]
    FIG. 5 illustrates steps taken by the client and the server in an embodiment of the invention, during the remainder of the initial phase of the LWIO protocol. At step 601 the client establishes a connection to the server over a shared communication device, as explained above. The client and the server now mutually authenticate the new connection. At step 603 the client sends an authentication request message (LWIO_OPCODE_AUTH) to the server. Authentication is done in order to prevent server-side and client-side spoofing. If the authentication is not timely completed, the connection is terminated.
  • [0071]
    FIG. 6A provides an illustrative representation of the client authenticate request message in an embodiment of the invention. The authenticate message 617 comprises the common header 619 followed by an LWIO_AUTH_PARAMS structure 621. In the header 619, Length 623 is set to the number of bytes sent to the server (the size of the common header 619 plus the size of the LWIO_AUTH_PARAMS 621). Opcode 625 is set to LWIO_OPCODE_AUTH (0x6). Flags 627 is set to LWIO_HDR_FLAG_INTERRUPT. Cookie 629, in this and the other client protocol messages, is set to a value chosen by the client and is sent back in the server reply. The Cookie value is typically used to match a request with a server reply. DataVa 631 is set to the address to which the server should RDMA the server authentication parameters. DataMh 633 holds the RDMA memory handle associated with DataVa 631.
  • [0072]
    In an embodiment of the invention, the LWIO_AUTH_PARAMS structure has the following format:
  • [0000]
    #define LWIO_AUTH_OPTION_END 0
    #define LWIO_AUTH_OPTION_KEY 1
    #define LWIO_AUTH_OPTION_SESSION_ID 2
    #define LWIO_AUTH_OPTION_SIGNATURE 3
    #define LWIO_AUTH_OPTION_KEY_LENGTH 16
    #define LWIO_AUTH_OPTION_SESSION_ID_LENGTH 8
    #define LWIO_AUTH_OPTION_SIGNATURE_LENGTH 16
    typedef struct {
      UCHAR OptionCode;
      UCHAR OptionLen;
      BYTE OptionData[1];
    } LWIO_AUTH_OPTIONS, *LPLWIO_AUTH_OPTIONS;
    typedef struct {
      CHAR Magic[4]; // ‘LWIO’
      UINT16 RevId;
      UINT16 Endian;
      UINT16 PageSize;
      UINT16 BaseSequence;
      UINT32 MaxRdmaWindowSize;
      UINT32 MaxSendBufferSize;
      UINT32 MaxRecvBufferSize;
      UINT16 HeaderSize;
      UINT16 Credits;
      UINT16 RdmaReadSupported;
      LWIO_AUTH_OPTIONS Options[1];
    } LWIO_AUTH_PARAMS, *LPLWIO_AUTH_PARAMS;
  • [0073]
    In the authenticate message 617, an LWIO_AUTH_PARAMS 621 forms the second part of the packet. Magic 635 is set to ‘LWIO’. RevId 637 is set to LWIO_REV_ID. Endian 639 is set to sizeof(ULONG_PTR). PageSize 641 is set to the CPU page size (4 k on 32-bit machines and 8 k on 64-bit machines). BaseSequence 643 is set to 0. MaxRdmaWindowSize 645 is intended to be set to the maximum number of bytes that the client can accept in an RDMA transfer; in the depicted embodiment it is set to 64 k. MaxSendBufferSize 647 is intended to be set to the number of bytes that the client can send to the server in a single request; in the depicted embodiment it is set to 1 k. MaxRecvBufferSize 649 is intended to be set to the number of bytes that the client has posted to receive data from the server; in the depicted embodiment it is set to 16 bytes. HeaderSize 651 is set to the number of bytes in the LWIO control header 619. Credits 652 is set to the initial number of buffer credits that the client wishes to have. The use of credits is explained further below. The server may or may not satisfy the client's request. RdmaReadSupported 653 is set to 0 if the client does not support RDMA read operations and is set to 1 if the client does support RDMA read.
  • [0074]
    Part of the LWIO_AUTH_PARAMS structure is a set of one or more options. The options are used to make authentication more flexible. Each option has an option code, length and data, except for the last option in the list, LWIO_AUTH_OPTION_END, which has the option code only, serving as a null option terminating the list of options. In the authenticate message, the client sends the server the following options: Key (LWIO_AUTH_OPTION_KEY) and a signature (LWIO_AUTH_OPTION_SIGNATURE). Key 655 is set to the key previously returned by the server in the negotiate response. Signature 657 is an MD5 signing of the LWIO_AUTH_PARAMS 621 excluding the signature.
  • [0075]
    Returning to FIG. 5, at step 605, if the Key sent in the authenticate message matches the key that was returned in the negotiate response over the pipe, the server RDMAs to the client as an authenticate response an LWIO_AUTH_PARAMS structure, including an eight-byte SessionId, to the DataVa address and associated DataMh memory handle provided by the client in the authenticate message. At step 607 the server sends an LWIO_MSG_STATUS_RESPONSE to complete the authentication.
  • [0076]
    FIG. 6B provides an illustrative representation of the LWIO_AUTH_PARAMS structure 659 returned by the server in an embodiment of the invention. Magic 661 is set to ‘LWIO’. RevId 663 is set to LWIO_REV_ID. Endian 665 is set to sizeof(ULONG_PTR). PageSize 667 is set to the CPU page size. BaseSequence 669 is intended to be set to (client BaseSequence+1). MaxRdmaWindowSize 671 is intended to be set to the maximum number of bytes that the client can accept in an RDMA transfer; in the depicted embodiment it is set to 512 k. MaxSendBufferSize 673 is intended to be set to the number of bytes that the server sends to the client in a single response; in the depicted embodiment it is set to 16 bytes. MaxRecvBufferSize 675 is intended to be set to the number of bytes that the server has pre-posted to receive data from the client; in the depicted embodiment it is set to 8 k. HeaderSize 677 is set to the number of bytes in the common header. Credits 679 is set to the initial number of credits that the server has available for the client. RdmaReadSupported 681 is set to 0 if the server does not support RDMA read and is set to 1 if the server does support RDMA read. The server sends the following options: Key (LWIO_AUTH_OPTION_KEY) 683, SessionId (LWIO_AUTH_OPTION_SESSION_ID) 685, and a Signature (LWIO_AUTH_OPTION_SIGNATURE) 687. Key 683 is set to the Key that the client had sent previously in the Negotiate Request. The SessionId 685 value is used by the client in registering client files with the server, as explained below. Signature 687 is an MD5 signing of the LWIO_AUTH_PARAMS excluding the Signature.
  • [0077]
    In an embodiment of the invention, the LWIO_MSG_STATUS_RESPONSE structure has the following format:
  • [0000]
    typedef struct _LWIO_IO_STATUS_BLOCK {
      UINT32 Information;
      UINT32 Status;
    } LWIO_IO_STATUS_BLOCK, *LPLWIO_IO_STATUS_BLOCK;
    typedef struct _LWIO_MSG_STATUS_RESPONSE {
      UINT64 Cookie;
      LWIO_IO_STATUS_BLOCK Ios;
    } LWIO_MSG_STATUS_RESPONSE,
    *LPLWIO_MSG_STATUS_RESPONSE;

    FIG. 6C provides an illustrative representation of the LWIO_MSG_STATUS_RESPONSE 689 returned by the server to complete the authentication in an embodiment of the invention. Cookie 691 is set to the cookie value set by the client in the header of the authenticate message. Information 693 is set to the number of bytes of LWIO_AUTH_PARAMS plus eight bytes. Status 695 is set to 0x0 (signifying success) or 0xC0000022 (signifying “access denied”).
  • File Registration
  • [0078]
    Returning to FIG. 5, at step 609, when the new connection has been mutually authenticated by the client and the server, the client begins registering files for use with the server. File operations for a file are not processed over a link until the client has registered the file for use with the server.
  • [0079]
    FIG. 7A provides an illustrative representation of the register file message sent by the client to the server in an embodiment of the invention. The registration message 701 comprises the common header 703 followed by an LWIO_FID_PARAMS structure 705. Length 707 is set to the number of bytes sent to the server (the size of the header 703 plus the size of the LWIO_FID_PARAMS 705). Opcode 709 is set to LWIO_OPCODE_REGISTER (0x7). Flags 711 is set to LWIO_HDR_FLAG_INTERRUPT. In this client message and subsequent client messages, Credits 713 is set to the number of pending I/O requests on the client. The Credits field serves as a hint to the server to allocate more credits to the connection, thus allowing additional outstanding I/O requests, as explained further below. The number of outstanding client requests at any one time cannot exceed the “Credits” value. As before, Cookie 715 is set to a client-specified value.
  • [0080]
    In an embodiment of the invention, the LWIO_FID_PARAMS structure has the following format:
  • [0000]
    typedef struct {
      SRV_RESUME_KEY ResumeKey;
      INT64 SessionId;
      UINT32 FlagsAndAttributes;
    } LWIO_FID_PARAMS, *LPLWIO_FID_PARAMS;

    In the LWIO_FID_PARAMS 705 of the register file message 701, ResumeKey 717 is set to the server request resume key that was returned over the initial file access channel. SessionId 719 is set to the SessionId that was returned by the server during the connection authentication stage. FlagsAndAttributes 721 is set to the Win32 Create Flags used initially to open the file.
  • [0081]
    Returning to FIG. 5, at step 611 the server responds with an LWIO_MSG_STATUS_RESPONSE to complete the file registration. FIG. 7B provides an illustrative representation of the LWIO_MSG_STATUS_RESPONSE 723 sent by the server in an embodiment of the invention. Information 725 is set to the Fid (File ID) to be used when sending I/O requests. Status 727 is set to 0x0 (success) or another NTSTATUS code on failure. Cookie 729 is set to the cookie value that the client set in the header of the register file message.
  • I/O Processing
  • [0082]
    At this point client connections are established and files have been registered, and the I/O processing phase of the LWIO protocol begins. One key feature of embodiments of the LWIO protocol is an asymmetric I/O model for reads and writes. Read operations are implemented using RDMA, while writes are implemented using send operations. Writes are not implemented using RDMA in order to provide a better security model. If the server exposes its address space over the NIC for RDMA it introduces a data corruption vulnerability that can be exploited by a malicious client. In this scenario, the malicious client issues, in a loop, RDMA write operations on a given server virtual address. Because the server address space is finite and at some point server virtual addresses must be reused, the malicious client eventually catches the server using the same virtual address for a different connection, causing the data to be written into a server buffer that might be associated with a different client. The asymmetric I/O model in the LWIO protocol guards against this possibility. This feature is a principal difference between the LWIO protocol and other RDMA-based file transfer protocols, such as DAFS.
  • [0083]
    Returning to FIG. 5, at step 613, the client begins posting I/O processing requests. Server-to-client completions of I/O requests are either in non-polling mode or polling mode. In non-polling mode, I/O completions are interrupt-based, using ordinary send/receive messages. In polling mode, I/O completions use RDMA and are not interrupt-based.
  • [0084]
    The flow diagram of FIG. 8 generally illustrates, from the general perspective of the LWIO server, steps taken in an embodiment of the invention with respect to completing an I/O request in polling mode or non-polling mode. A client I/O request specifies whether the server should send back a post-send (interrupting the CPU) or an RDMA message. At step 801, the server determines whether an LWIO_HDR_FLAG_INTERRUPT flag is set in the common header of the client I/O request message. If this flag is set, at step 803 the server completes the client request by way of an LWIO_MSG_STATUS_RESPONSE using an ordinary send. If the LWIO_HDR_FLAG_INTERRUPT flag is not set (polling mode), then the server completes the client request by RDMAing an LWIO_IO_STATUS_BLOCK to the client, as indicated at step 805.
  • Wakeup of Client in Polling Mode
  • [0085]
    In polling mode, the client may wish to sleep while waiting for an I/O completion from the server. Completions in this case are sent by way of RDMA to the client, so a mechanism is needed to wake up the client to notify it that a completion has occurred. If the client wishes to be woken up, it sends an interrupt request (LWIO_OPCODE_INTERRUPT) message to the server, received by the server at step 807 of FIG. 8. A server that receives an interrupt request will not send a response until an I/O request has completed on the server (step 809). The completion is sent to the client at step 811 by way of an ordinary send, interrupting the client. Only one interrupt message can be outstanding for a given client connection.
  • [0086]
    FIG. 9A provides an illustrative representation of the interrupt request message sent by the client to the server in an embodiment of the invention. The message comprises the common header 815. Opcode 817 is set to LWIO_OPCODE_REGISTER (0x9). Flags 819 is set to (LWIO_HDR_FLAG_INTERRUPT|LWIO_HDR_FLAG_CONTROL) (0xC0). Credits 821 is set to the number of pending I/O requests on the client, and Cookie 823 is set to a client-specified value.
  • [0087]
    The server responds to the interrupt request message after another I/O request has been processed. FIG. 9B provides an illustrative representation of the LWIO_MSG_STATUS_RESPONSE message 825 sent by server in an embodiment of the invention. Information 827 is set to 0. Status 829 is set to 0x0 (success) or another NTSTATUS code on failure. Cookie 831 is set to the Cookie value in the header of the interrupt request sent by the client.
  • Credits
  • [0088]
    As has been noted, all client-to-server I/O requests include a credits field in the header. The credits field is a hint to the server regarding the number of outstanding I/O requests that the client would like to send to the server. It is the responsibility of the server to manage credits. Credits provide a novel solution to the problem of flushing buffers. If the client currently has N credits, it is required to post N+1 receive buffers in order for the server to send the client a credit message. The server has only one outstanding credit request along a client connection at any one time. Credit messages are always sent in interrupt mode.
  • [0089]
    A credit transaction comprises a server-initiated three-way handshake between client and server. FIG. 10 generally illustrates the steps comprising the credit transaction in an embodiment of the invention. At step 1001 the server sends a delta credit request message along a client connection.
  • [0090]
    FIG. 11A provides an illustrative representation of the server delta credit message in an embodiment of the invention. This message takes the form of an LWIO_MSG_STATUS_RESPONSE 1011. Credits correspond to buffers. Information 1013 is set to the number of credits that the client should give up (a negative number) or the number of credits (extra buffers) that the server has newly allocated for the client's use (a positive number). Status 1015 is set to LWIO_NOTIFY_CREDIT (0x1). Cookie 1017 is set to 0.
  • [0091]
    Returning to FIG. 10, the client receives the credit message from the server. The client is required to respond with an LWIO_OPCODE_CREDIT message to the server on the same connection. This message signifies either the releasing of a single credit or notifying the server of the number of the newly-allocated credits that the client has used. If the Information field in the server credit message contains a negative number, −N (step 1003), the client sends N LWIO_OPCODE_CREDIT messages (one for each credit that it is required to give up), indicated as step 1005. If the Information field is positive, then the client sends only one LWIO_OPCODE_CREDIT message, indicated as step 1007.
  • [0092]
    FIG. 11B provides an illustrative representation of the LWIO_OPCODE_CREDIT message sent by the client in an embodiment of the invention. The LWIO_OPCODE_CREDIT message 1019 comprises a common header 1021. Opcode 1023 is set to LWIO_OPCODE_CREDIT (0x8). Flags 1025 is set to LWIO_HDR_FLAG_INTERRUPT (0x80). Credits 1027 is set to the number of pending I/O requests on the client. Cookie 1031 is set to a client-specified value. If the client received a positive delta credit message, the upper 32 bits of Offset 1029 are set to the number of credits allocated by the server that the client did not use. Once the client returns a value greater than zero in this field, the server normally does not send another positive update message until at least one negative update is sent. Typically, the client returns zero.
  • [0093]
    As noted above, if the client received a negative (−N) delta credit message, the client is required to send N credit messages to the server, one for each credit that it is giving up. The upper 32 bits of Offset 1029 in this case are accordingly set to −N, −(N−1), . . . , −1. When the server receives the client credit message with the upper 32 bits of Offset 1029 set to −1, the server assumes that the client has finished processing the server credit message and is eligible to receive new credit messages.
  • [0094]
    Returning to FIG. 10, the server completes the three-way handshake by sending an LWIO_MSG_STATUS_RESPONSE message to the client, indicated as step 1009. FIG. 11C provides an illustrative representation of the LWIO_MSG_STATUS_RESPONSE 1033 sent by the server in an embodiment of the invention. Information 1037 is set to 0. If the upper 32 bits of Offset in the header of the LWIO_OPCODE_CREDIT message sent by the client was greater than or equal to zero, Status 1039 is set to 0x0, signifying success. If the upper 32 bits of Offset were set to a negative number, the server sets Status 1039 to LWIO_CREDIT_NOTIFY in order to allow the client to retire the credit. Cookie 1035 is set to the Cookie value set by the client in the common header of the LWIO_OPCODE_CREDIT message.
  • Close
  • [0095]
    The close message is used to stop I/O processing for a particular Fid that was exchanged during the registration stage. Once the server responds, any new requests will fail until the Fid is recycled. FIG. 12A provides an illustrative representation of the close message sent by the client in an embodiment of the invention. The close message 1041 comprises a common header 1043. Opcode 1045 is set to LWIO_OPCODE_CLOSE (0x4). Flags 1047 is set to LWIO_HDR_FLAG_INTERRUPT (0x80). Credits 1049 is set to the number of pending I/O requests on the client. Cookie 1053 is set to a client-specified value. Fid 1051 is set to the File Id of the file that is to be closed.
  • [0096]
    The server responds with an LWIO_MSG_STATUS_RESPONSE. FIG. 12B provides an illustrative representation of the close completion LWIO_MSG_STATUS_RESPONSE 1055 returned by the server in an embodiment of the invention. Information 1059 is set to 0. Status 1061 is set to 0, indicating success. Cookie 1057 is set to the Cookie value that was set in the client close request.
  • Cancel
  • [0097]
    The cancel message is used to stop I/O processing for a particular Fid that was exchanged during the registration stage. When the cancel is issued, the server completes the request. However, I/O requests that cannot be canceled may still proceed on the server. FIG. 13A provides an illustrative representation of the cancel message sent by the client in an embodiment of the invention. The cancel message 1063 comprises a common header 1065. Opcode 1067 is set to LWIO_OPCODE_CANCEL (0x5). Flags 1069 is set to LWIO_HDR_FLAG_INTERRUPT (0x80). Credits 1071 is set to the number of pending I/O requests on the client. Cookie 1075 is set to a client-specified value. Fid 1073 is set to the File Id on which the cancel is being issued.
  • [0098]
    The server completes the cancel with an LWIO_MSG_STATUS_RESPONSE message. FIG. 13B provides an illustrative representation of the cancel completion LWIO_MSG_STATUS_RESPONSE 1077 returned by the server in an embodiment of the invention. Information 1081 is set to 0. Status 1083 is set to 0, indicating success. Cookie 1079 is set to the Cookie value that was set in the client cancel request.
  • Read
  • [0099]
    The read message is used to obtain data from a particular Fid that was exchanged during the registration stage. For a read request smaller than one kilobyte, if the user buffer is not registered with the NIC, the data is received into an internal pre-registered buffer, and a copy is performed into the user buffer once the data is received from the server. This is done because it is more efficient to copy small amounts of data rather than to register small user buffers. For large reads the user buffer is registered and the data is received directly by way of RDMA write. The amount of data read pursuant to a single read request is limited by the server MaxRdmaWindowSize.
  • [0100]
    FIGS. 14A and 14C provide illustrative representations of the read message sent by the client in an embodiment of the invention, with FIG. 14A giving the non-polling case and FIG. 14C giving the polling case. The read message 1401 comprises a common header 1403. Length 1405 is set to the number of bytes to be read from the associated file. Opcode 1407 is set to LWIO_OPCODE_READ (0x0). Offset 1417 is set to the byte location at which the file read is to begin. Marker 1413 is set to 0xFF. Flags 1409, 1427 is set to 0x0 in the polling case 1427 or LWIO_HDR_FLAG_INTERRUPT (0x80) in the non-polling case 1409. Credits 1411 is set to the number of pending I/O requests on the client. Fid 1415 is set to the File Id on which to issue the I/O. DataVa 1419 is set to the address to which the read data is to be RDMAed, and DataMh 1421 is set to the associated memory handle.
  • [0101]
    In the non-polling case, ImmediateCookie 1423 and Cookie 1425 are set to client-specified values. The server can complete the read request in this case with an LWIO_MSG_STATUS_RESPONSE by way of a normal send, or with an RDMA with immediate data if the read is successful. The immediate data of the RDMA write is accordingly set to the ImmediateCookie value of the read request. In the polling case, IosVa 1431 is set to the location to which the server response status (LWIO_IO_STATUS_BLOCK) is RDMAed, and IosMh 1429 is set to the associated memory handle.
  • [0102]
    In the non-polling case, the server first RDMAs the read data. The server then can respond with an LWIO_MSG_STATUS_RESPONSE, or the server can send immediate data with the RDMA read data, in which case the immediate data is set to the ImmediateCookie value of the read request. FIG. 14B provides an illustrative representation of the LWIO_MSG_STATUS_RESPONSE 1433 returned by the server in the non-polling case in an embodiment of the invention.
  • [0000]
    Information 1437 is set to the number of bytes read. Status 1439 is set to 0, indicating success, or to another NTSTATUS, indicating failure. Cookie 1435 is set to the Cookie value set by the client in the header of the read message.
  • [0103]
    In the polling case, the server first RDMAs the read data. The server then RDMAs an LWIO_IO_STATUS_BLOCK to the client. FIG. 14D provides an illustrative representation of the LWIO_IO_STATUS_BLOCK 1441 returned by the server in an embodiment of the invention. Information 1443 is set to the number of bytes read. Status 1445 is set to 0, indicating success, or another NTSTATUS, indicating failure.
  • Write
  • [0104]
    The write message is used to place data into a particular Fid that was exchanged during the file registration. All write data is sent using ordinary send operations. The amount of data written is limited by the server MaxRecvBufferSize. If the client sends more data than this, the connection is terminated.
  • [0105]
    FIGS. 15A and 15C provide illustrative representations of the write message sent by the client in an embodiment of the invention, with FIG. 15A giving the non-polling case and FIG. 15C giving the polling case. The write message 1501 includes a common header 1503. Length 1505 is set to the number of bytes of data to be written. Opcode 1507 is set to LWIO_OPCODE_WRITE (0x1). Offset 1517 is set to the byte location at which to begin writing the file data. Flags 1509, 1529 is set to 0x0 in the polling case 1529 or LWIO_HDR_FLAG_INTERRUPT (0x80) in the non-polling case 1509. Marker 1513 is set to 0xFF. Credits 1511 is set to the number of pending I/O requests on the client. Fid 1515 is set to the File Id on which to issue the I/O. The data to be written 1527 immediately follows the common header 1503 of the write message.
  • [0106]
    In the non-polling case, Cookie 1525 is set to a client-specified value. In the polling case, IosVa 1533 is set to the location at which the server response status (LWIO_IO_STATUS_BLOCK) is RDMAed, and IosMh 1531 is set to the associated memory handle.
  • [0107]
    In the non-polling case, the server responds to the write message with an LWIO_MSG_STATUS_RESPONSE. FIG. 15B provides an illustrative representation of the LWIO_MSG_STATUS_RESPONSE 1535 returned by the server in an embodiment of the invention. Information 1539 is set to the number of bytes written. Status 1541 is set to 0, indicating success, or to another NTSTATUS, indicating failure. Cookie 1537 is set to the Cookie value set by the client in the header of the write message. In the polling case, the server RDMAs an LWIO_IO_STATUS_BLOCK. FIG. 15D provides an illustrative representation of the LWIO_IO_STATUS_BLOCK 1543 returned by the server in an embodiment of the invention. Information 1545 is set to the number of bytes written. Status 1547 is set to 0, indicating success, or to another NTSTATUS, indicating failure.
  • Vectored Read
  • [0108]
    The vectored read is used to obtain data from a particular Fid that was exchanged during the registration stage and to scatter the data on a page basis to multiple segments on the requester. All data read is sent to the requester by way of RDMA writes, with one RDMA write from the server for each read segment. The data read from disk is contiguous. The amount of data read is limited by the maximum number of destination pages that can be described in a single request. This limit is the server MaxRecvBufferSize divided by sizeof(LWIO_RDMA_REGION). The structure of LWIO_RDMA_REGION is given below.
  • [0109]
    FIGS. 16A and 16C provide illustrative representations of the vectored read message sent by the client in an embodiment of the invention, with FIG. 16A giving the non-polling case and FIG. 16C giving the polling case. The read message 1401 comprises a common header 1603 followed by one or more LWIO_RDMA_REGION segments 1605, 1607. In the header 1603, Length 1609 is set to the number of bytes of data to be read from the file. Opcode 1611 is set to LWIO_OPCODE_VEC_READ (0x2). Offset 1621 is set to the byte location at which to begin reading the file data. Flags 1613, 1631 is set to 0x0 in the polling case 1631, or LWIO_HDR_FLAG_INTERRUPT (0x80) in the non-polling case 1613. Marker 1617 is set to 0xFF. Credits 1615 is set to the number of pending I/O requests on the client. Fid 1619 is set to the File Id on which to issue the I/O. NumPages 1623 is set to the number of LWIO_RDMA_REGIONs that follow the common header 1603. PageSize 1625 is set to the local page size in bytes.
  • [0110]
    In the non-polling case, ImmediateCookie 1627 and Cookie 1629 are set to client-specified values. The server can complete the vectored read request in this case with an LWIO_MSG_STATUS_RESPONSE by way of a normal send, or with an RDMA with immediate data if the read is successful. The immediate data of the RDMA write is accordingly set to the ImmediateCookie 1627 value of the read request. In the polling case, IosVa 1635 is set to the location at which the server response status (LWIO_IO_STATUS_BLOCK) is RDMAed, and IosMh 1633 is set to the associated memory handle.
  • [0111]
    The common header 1603 is immediately followed by a sufficient number of LWIO_RDMA_REGION segments 1605, 1607 to cover the length of the request. All intermediate segments must be one page in size. The final segment may be smaller than a page, but it must be a multiple of the backend disk sector size. In an embodiment of the invention, the LWIO_RDMA_REGION has the following format:
  • [0000]
    typedef volatile struct {
      PVOID64 DataVa;
      UINT32 DataMh;
      UINT32 Length;
    } LWIO_RDMA_REGION;

    The first LWIO_RDMA_REGION corresponds to the first PageSize bytes read, the second LWIO_RDMA_REGION corresponds to the second PageSize bytes read, and so on. DataVa 1637 is set to the location marking the beginning of the page in which the read data is to be placed. DataMh 1639 is set to the memory handle of the DataVa 1637. Length 1641 is set to the PageSize 1625 for all regions except for the final region, for which Length may be smaller but must be a multiple of the backend disk sector size.
  • [0112]
    In the non-polling case, the server first RDMAs the read data. The server then can respond with an LWIO_MSG_STATUS_RESPONSE, or the server can send immediate data with the RDMA read data, in which case the immediate data is set to the ImmediateCookie value of the read request. FIG. 16B provides an illustrative representation of the LWIO_MSG_STATUS_RESPONSE 1643 returned by the server in the non-polling case in an embodiment of the invention. Information 1647 is set to the number of bytes read. Status 1649 is set to 0, indicating success, or to another NTSTATUS, indicating failure. Cookie 1645 is set to the Cookie value set by the client in the header of the vectored read message.
  • [0113]
    In the polling case, first the server RDMAs the read data, and then the server RDMAs an LWIO_IO_STATUS_BLOCK. FIG. 16D provides an illustrative representation of the LWIO_IO_STATUS_BLOCK 1651 returned by the server in an embodiment of the invention. Information 1653 is set to the number of bytes read. Status 1655 is set to 0, indicating success, or another NTSTATUS, indicating failure.
  • Vectored Write
  • [0114]
    The vectored write message is used to perform a gather write into a particular Fid that was exchanged during the file registration. All write data is sent using ordinary send operations. The amount of data written is limited by the server MaxRecvBufferSize. If the client sends more data than this, the connection is terminated.
  • [0115]
    FIGS. 17A, 17B and 17C provide illustrative representations of the vectored write message sent by the client in an embodiment of the invention, with FIG. 17A illustrating the non-polling, non-collapse case, FIG. 17B illustrating the non-polling, collapse case, and FIG. 17C illustrating the polling, collapse case.
  • [0000]
    The write message 1701 includes a common header 1703, immediately followed by the data to be written 1705. In the common header 1703, Length 1707 is set to the number of bytes of data being written. Opcode 1709 is set to LWIO_OPCODE_WRITE (0x3). Offset 1719 is set to the byte location at which to begin writing the file data. Marker 1715 is set to 0xFF. Credits 1713 is set to the number of pending I/O requests on the client. Fid 1717 is set to the File Id on which to issue the I/O.
  • [0116]
    Flags 1711, 1721, 1727 is set to 0x0, signifying polling 1727, or else to LWIO_HDR_FLAG_INTERRUPT (0x80) 1711. In the latter case, flags can also include LWIO_HDR_FLAG_COLLAPSE 1721 to indicate that all pages in the write contain the same data, so that only a single page of data has been sent. This is an optimization intended to minimize the transfer of redundant data. LWIO_HDR_FLAG_COLLAPSE can only be used if the registered file flags include FILE_NO_INTERMEDIATE_BUFFERING (0x8) and the PageSizes exchanged during the authentication stage are even multiples of each other. In the case of a collapsed I/O, NumPages 1723 is set to the number of pages of data spanned by the I/O. The last page may be partial due to the Length parameter. PageSize 1725 is set to the local page size in bytes. In the polling case, IosVa 1731 is set to the location at which the server response status (LWIO_IO_STATUS_BLOCK) is to be RDMAed. IosMh 1729 is the associated memory handle.
  • [0117]
    In the non-polling case, for both non-collapsed and collapsed I/O, the server responds to the write message with an LWIO_MSG_STATUS_RESPONSE.
  • [0000]
    FIG. 17D provides an illustrative representation of the LWIO_MSG_STATUS_RESPONSE 1733 returned by the server in an embodiment of the invention. Information 1737 is set to the number of bytes written. Status 1739 is set to 0, indicating success, or to another NTSTATUS, indicating failure. Cookie 1735 is set to the Cookie value set by the client in the header of the write message.
  • [0118]
    In the polling case, for both non-collapsed and collapsed I/O, the server RDMAs an LWIO_IO_STATUS_BLOCK. FIG. 17E provides an illustrative representation of the LWIO_IO_STATUS_BLOCK 1741 returned by the server in an embodiment of the invention. Information 1743 is set to the number of bytes written. Status 1745 is set to 0, indicating success, or to another NTSTATUS, indicating failure.
  • CONCLUSION
  • [0119]
    While illustrative embodiments of the invention have been illustrated and described, it will be appreciated that various changes can be made without departing from the invention. Similarly, any process steps described herein may be interchangeable with other steps in order to achieve the same result. In addition, the illustrative examples described above are not intended to be exhaustive or to limit the invention to the precise forms disclosed. On the contrary, the intention is to cover all modifications, alternative constructions, and equivalents falling within the spirit and scope of the invention.

Claims (1)

  1. 1. A method for managing buffers in an input/output offload protocol, comprising:
    sending, by a server to a client, a delta credit message including an information field set to a number of credits, wherein, if the number is a negative number −N, the server requires the client to retire N credits;
    if the number of credits is a negative number −N, sending, by the client to the server, N credit messages, and otherwise sending, by the client to the server, one credit message; and
    for each credit message sent by the client, sending, by the server to the client, a status response message.
US12717306 2003-12-31 2010-03-04 Lightweight input/output protocol Abandoned US20100161855A1 (en)

Priority Applications (2)

Application Number Priority Date Filing Date Title
US10749959 US7698361B2 (en) 2003-12-31 2003-12-31 Lightweight input/output protocol
US12717306 US20100161855A1 (en) 2003-12-31 2010-03-04 Lightweight input/output protocol

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
US12717306 US20100161855A1 (en) 2003-12-31 2010-03-04 Lightweight input/output protocol

Publications (1)

Publication Number Publication Date
US20100161855A1 true true US20100161855A1 (en) 2010-06-24

Family

ID=34701133

Family Applications (2)

Application Number Title Priority Date Filing Date
US10749959 Expired - Fee Related US7698361B2 (en) 2003-12-31 2003-12-31 Lightweight input/output protocol
US12717306 Abandoned US20100161855A1 (en) 2003-12-31 2010-03-04 Lightweight input/output protocol

Family Applications Before (1)

Application Number Title Priority Date Filing Date
US10749959 Expired - Fee Related US7698361B2 (en) 2003-12-31 2003-12-31 Lightweight input/output protocol

Country Status (8)

Country Link
US (2) US7698361B2 (en)
EP (1) EP1700264A4 (en)
JP (1) JP4564970B2 (en)
KR (1) KR101084897B1 (en)
CN (1) CN101375263B (en)
CA (1) CA2501521A1 (en)
RU (2) RU2388039C2 (en)
WO (1) WO2005067430A3 (en)

Cited By (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20080025290A1 (en) * 2006-07-27 2008-01-31 Sharon Barkai Distributed edge network
US20090109968A1 (en) * 2007-10-30 2009-04-30 Ariel Noy Grid router
CN102708062A (en) * 2012-04-10 2012-10-03 深圳市佳信捷电子有限公司 Multimedia storage method and multimedia storage device based on data block
WO2013002980A2 (en) * 2011-06-29 2013-01-03 Microsoft Corporation Transporting operations of arbitrary size over remote direct memory access
US8631277B2 (en) 2010-12-10 2014-01-14 Microsoft Corporation Providing transparent failover in a file system
US8788579B2 (en) 2011-09-09 2014-07-22 Microsoft Corporation Clustered client failover
US8825885B2 (en) 2005-05-25 2014-09-02 Microsoft Corporation Data communication protocol
US8856582B2 (en) 2011-06-30 2014-10-07 Microsoft Corporation Transparent failover
US9111081B2 (en) 2012-06-26 2015-08-18 International Business Machines Corporation Remote direct memory access authentication of a device
US9344369B2 (en) 2008-08-21 2016-05-17 Hewlett Packard Enterprise Development Lp System and methods for distributed quality of service enforcement

Families Citing this family (29)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7529778B1 (en) 2001-12-12 2009-05-05 Microsoft Corporation System and method for providing access to consistent point-in-time file versions
US7698361B2 (en) * 2003-12-31 2010-04-13 Microsoft Corporation Lightweight input/output protocol
US7617256B2 (en) * 2004-07-19 2009-11-10 Microsoft Corporation Remote file updates through remote protocol
US20060168094A1 (en) * 2005-01-21 2006-07-27 International Business Machines Corporation DIRECT ACCESS OF SCSI BUFFER WITH RDMA ATP MECHANISM BY iSCSI TARGET AND/OR INITIATOR
US7549037B1 (en) * 2005-09-01 2009-06-16 Symantec Operating Corporation Efficient off-host backup of a file set clone
US8151116B2 (en) * 2006-06-09 2012-04-03 Brigham Young University Multi-channel user authentication apparatus system and method
US7836143B2 (en) 2007-04-25 2010-11-16 International Business Machines Corporation Message communications of particular message types between compute nodes using DMA shadow buffers
US8325633B2 (en) 2007-04-26 2012-12-04 International Business Machines Corporation Remote direct memory access
US7827024B2 (en) 2007-05-09 2010-11-02 International Business Machines Corporation Low latency, high bandwidth data communications between compute nodes in a parallel computer
US7966618B2 (en) * 2007-05-29 2011-06-21 International Business Machines Corporation Controlling data transfers from an origin compute node to a target compute node
US8037213B2 (en) 2007-05-30 2011-10-11 International Business Machines Corporation Replenishing data descriptors in a DMA injection FIFO buffer
US7921428B2 (en) 2007-06-18 2011-04-05 International Business Machines Corporation Multi-registration of software library resources
US8478834B2 (en) 2007-07-12 2013-07-02 International Business Machines Corporation Low latency, high bandwidth data communications between compute nodes in a parallel computer
US8018951B2 (en) 2007-07-12 2011-09-13 International Business Machines Corporation Pacing a data transfer operation between compute nodes on a parallel computer
US7805546B2 (en) * 2007-07-27 2010-09-28 International Business Machines Corporation Chaining direct memory access data transfer operations for compute nodes in a parallel computer
US8959172B2 (en) 2007-07-27 2015-02-17 International Business Machines Corporation Self-pacing direct memory access data transfer operations for compute nodes in a parallel computer
US8521821B2 (en) * 2009-03-17 2013-08-27 Brigham Young University Encrypted email based upon trusted overlays
US8949453B2 (en) 2010-11-30 2015-02-03 International Business Machines Corporation Data communications in a parallel active messaging interface of a parallel computer
US8806030B2 (en) * 2010-12-06 2014-08-12 Microsoft Corporation Multichannel connections in file system sessions
FR2973185B1 (en) * 2011-03-22 2013-03-29 Sagem Defense Securite Method and connection device has a high-security network
US8949328B2 (en) 2011-07-13 2015-02-03 International Business Machines Corporation Performing collective operations in a distributed processing system
RU2589398C2 (en) * 2011-10-27 2016-07-10 Хуавэй Текнолоджиз Ко., Лтд. Data-fast-distribution method and device
US8930962B2 (en) 2012-02-22 2015-01-06 International Business Machines Corporation Processing unexpected messages at a compute node of a parallel computer
US20150012607A1 (en) * 2013-07-08 2015-01-08 Phil C. Cayton Techniques to Replicate Data between Storage Servers
US20150326684A1 (en) * 2014-05-07 2015-11-12 Diablo Technologies Inc. System and method of accessing and controlling a co-processor and/or input/output device via remote direct memory access
CN104202391A (en) * 2014-08-28 2014-12-10 浪潮(北京)电子信息产业有限公司 RDMA (Remote Direct Memory Access) communication method between non-tightly-coupled systems of sharing system address space
RU2599937C1 (en) * 2015-03-26 2016-10-20 Общество с ограниченной ответственностью "Научно-производственное предприятие "Цифровые решения" Method for prevention of faults in a local computer network caused by onset of incorrect units
US20170034268A1 (en) * 2015-07-31 2017-02-02 Netapp, Inc. Systems, methods and devices for rdma read/write operations
RU2638781C2 (en) * 2016-09-29 2017-12-15 Общество с ограниченной ответственностью "ГРЭК" Method for arranging direct access to memory in transmission of information between physical objects

Citations (32)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10612P (en) *
US10612A (en) * 1854-03-07 Seaman
US5499371A (en) * 1993-07-21 1996-03-12 Persistence Software, Inc. Method and apparatus for automatic generation of object oriented code for mapping relational data to objects
US6014681A (en) * 1997-07-15 2000-01-11 Microsoft Corporation Method for saving a document using a background save thread
US20010023460A1 (en) * 1997-10-14 2001-09-20 Alacritech Inc. Passing a communication control block from host to a local device such that a message is processed on the device
US20020063885A1 (en) * 2000-11-30 2002-05-30 Rabindranath Dutta System and method for advertisements in web-based printing
US20020091852A1 (en) * 2001-01-11 2002-07-11 International Business Machines Corporation Method and apparatus for providing a reliable protocol for transferring data
US20020099980A1 (en) * 2001-01-25 2002-07-25 Olarig Sompong P. Computer system having configurable core logic chipset for connection to a fault-tolerant accelerated graphics port bus and peripheral component interconnect bus
US20020144037A1 (en) * 2001-03-29 2002-10-03 Bennett Joseph A. Data fetching mechanism and method for fetching data
US20020152327A1 (en) * 2001-04-11 2002-10-17 Michael Kagan Network interface adapter with shared data send resources
US20020172195A1 (en) * 2001-03-23 2002-11-21 Pekkala Richard E. Apparatus amd method for disparate fabric data and transaction buffering within infiniband device
US20020184362A1 (en) * 2001-05-31 2002-12-05 International Business Machines Corporation System and method for extending server security through monitored load management
US20030061402A1 (en) * 2001-09-26 2003-03-27 Satyendra Yadav Method and apparatus enabling both legacy and new applications to access an InfiniBand fabric via a socket API
US20030065856A1 (en) * 2001-10-03 2003-04-03 Mellanox Technologies Ltd. Network adapter with multiple event queues
US20030145230A1 (en) * 2002-01-31 2003-07-31 Huimin Chiu System for exchanging data utilizing remote direct memory access
US20040003069A1 (en) * 2002-06-28 2004-01-01 Broadcom Corporation Selective early drop method and system
US20040010612A1 (en) * 2002-06-11 2004-01-15 Pandya Ashish A. High performance IP processor using RDMA
US20040019714A1 (en) * 2002-07-29 2004-01-29 Kelley Richard A. Transaction credit control for serial I/O systems
US20040030822A1 (en) * 2002-08-09 2004-02-12 Vijayan Rajan Storage virtualization by layering virtual disk objects on a file system
US6718370B1 (en) * 2000-03-31 2004-04-06 Intel Corporation Completion queue management mechanism and method for checking on multiple completion queues and processing completion events
US20040093451A1 (en) * 2002-11-13 2004-05-13 International Business Machines Corporation Connector that enables aggregation of independent system resources across PCI/PCI-X bus and interlocked method for aggregating independent system resources across PCI/PCI-X bus
US20040117438A1 (en) * 2000-11-02 2004-06-17 John Considine Switching system
US6832297B2 (en) * 2001-08-09 2004-12-14 International Business Machines Corporation Method and apparatus for managing data in a distributed buffer system
US20050078601A1 (en) * 2003-10-14 2005-04-14 Broadcom Corporation Hash and route hardware with parallel routing scheme
US20050122986A1 (en) * 2003-12-05 2005-06-09 Alacritech, Inc. TCP/IP offload device with reduced sequential processing
US20050137966A1 (en) * 2003-12-19 2005-06-23 Munguia Peter R. Flow control credit synchronization
US20050149632A1 (en) * 2003-12-19 2005-07-07 Iready Corporation Retransmission system and method for a transport offload engine
US6976174B2 (en) * 2001-01-04 2005-12-13 Troika Networks, Inc. Secure multiprotocol interface
US7103626B1 (en) * 2000-05-24 2006-09-05 Hewlett-Packard Development, L.P. Partitioning in distributed computer system
US7152128B2 (en) * 2001-08-24 2006-12-19 Intel Corporation General input/output architecture, protocol and related methods to manage data integrity
US20080275989A1 (en) * 2003-12-05 2008-11-06 Ebersole Dwayne E Optimizing virtual interface architecture (via) on multiprocessor servers and physically independent consolidated nics
US7698361B2 (en) * 2003-12-31 2010-04-13 Microsoft Corporation Lightweight input/output protocol

Family Cites Families (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6219761B1 (en) * 1998-05-06 2001-04-17 International Business Machines Corporation Load/store assist engine
US6356863B1 (en) * 1998-09-08 2002-03-12 Metaphorics Llc Virtual network file server
WO2001024059A3 (en) * 1999-09-29 2002-06-06 Cyberset Technologies Inc Method and apparatus for an active file system
JP4192416B2 (en) * 2000-06-08 2008-12-10 株式会社日立製作所 Computer system and method Data transfer
WO2002046866A9 (en) 2000-11-02 2003-07-03 Pirus Networks Switching system
US6748502B2 (en) * 2001-01-12 2004-06-08 Hitachi, Ltd. Virtual volume storage

Patent Citations (34)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10612P (en) *
US10612A (en) * 1854-03-07 Seaman
US5499371A (en) * 1993-07-21 1996-03-12 Persistence Software, Inc. Method and apparatus for automatic generation of object oriented code for mapping relational data to objects
US6014681A (en) * 1997-07-15 2000-01-11 Microsoft Corporation Method for saving a document using a background save thread
US20010023460A1 (en) * 1997-10-14 2001-09-20 Alacritech Inc. Passing a communication control block from host to a local device such that a message is processed on the device
US6718370B1 (en) * 2000-03-31 2004-04-06 Intel Corporation Completion queue management mechanism and method for checking on multiple completion queues and processing completion events
US7103626B1 (en) * 2000-05-24 2006-09-05 Hewlett-Packard Development, L.P. Partitioning in distributed computer system
US20040117438A1 (en) * 2000-11-02 2004-06-17 John Considine Switching system
US20020063885A1 (en) * 2000-11-30 2002-05-30 Rabindranath Dutta System and method for advertisements in web-based printing
US6976174B2 (en) * 2001-01-04 2005-12-13 Troika Networks, Inc. Secure multiprotocol interface
US20020091852A1 (en) * 2001-01-11 2002-07-11 International Business Machines Corporation Method and apparatus for providing a reliable protocol for transferring data
US20020099980A1 (en) * 2001-01-25 2002-07-25 Olarig Sompong P. Computer system having configurable core logic chipset for connection to a fault-tolerant accelerated graphics port bus and peripheral component interconnect bus
US20020172195A1 (en) * 2001-03-23 2002-11-21 Pekkala Richard E. Apparatus amd method for disparate fabric data and transaction buffering within infiniband device
US20020144037A1 (en) * 2001-03-29 2002-10-03 Bennett Joseph A. Data fetching mechanism and method for fetching data
US20020152327A1 (en) * 2001-04-11 2002-10-17 Michael Kagan Network interface adapter with shared data send resources
US20020184362A1 (en) * 2001-05-31 2002-12-05 International Business Machines Corporation System and method for extending server security through monitored load management
US6832297B2 (en) * 2001-08-09 2004-12-14 International Business Machines Corporation Method and apparatus for managing data in a distributed buffer system
US7152128B2 (en) * 2001-08-24 2006-12-19 Intel Corporation General input/output architecture, protocol and related methods to manage data integrity
US20030061402A1 (en) * 2001-09-26 2003-03-27 Satyendra Yadav Method and apparatus enabling both legacy and new applications to access an InfiniBand fabric via a socket API
US20030065856A1 (en) * 2001-10-03 2003-04-03 Mellanox Technologies Ltd. Network adapter with multiple event queues
US20030145230A1 (en) * 2002-01-31 2003-07-31 Huimin Chiu System for exchanging data utilizing remote direct memory access
US20040030806A1 (en) * 2002-06-11 2004-02-12 Pandya Ashish A. Memory system for a high performance IP processor
US20040010612A1 (en) * 2002-06-11 2004-01-15 Pandya Ashish A. High performance IP processor using RDMA
US20040010545A1 (en) * 2002-06-11 2004-01-15 Pandya Ashish A. Data processing system using internet protocols and RDMA
US20040003069A1 (en) * 2002-06-28 2004-01-01 Broadcom Corporation Selective early drop method and system
US20040019714A1 (en) * 2002-07-29 2004-01-29 Kelley Richard A. Transaction credit control for serial I/O systems
US20040030822A1 (en) * 2002-08-09 2004-02-12 Vijayan Rajan Storage virtualization by layering virtual disk objects on a file system
US20040093451A1 (en) * 2002-11-13 2004-05-13 International Business Machines Corporation Connector that enables aggregation of independent system resources across PCI/PCI-X bus and interlocked method for aggregating independent system resources across PCI/PCI-X bus
US20050078601A1 (en) * 2003-10-14 2005-04-14 Broadcom Corporation Hash and route hardware with parallel routing scheme
US20050122986A1 (en) * 2003-12-05 2005-06-09 Alacritech, Inc. TCP/IP offload device with reduced sequential processing
US20080275989A1 (en) * 2003-12-05 2008-11-06 Ebersole Dwayne E Optimizing virtual interface architecture (via) on multiprocessor servers and physically independent consolidated nics
US20050137966A1 (en) * 2003-12-19 2005-06-23 Munguia Peter R. Flow control credit synchronization
US20050149632A1 (en) * 2003-12-19 2005-07-07 Iready Corporation Retransmission system and method for a transport offload engine
US7698361B2 (en) * 2003-12-31 2010-04-13 Microsoft Corporation Lightweight input/output protocol

Cited By (20)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8850025B2 (en) 2005-05-25 2014-09-30 Microsoft Corporation Data communication coordination with sequence numbers
US8825885B2 (en) 2005-05-25 2014-09-02 Microsoft Corporation Data communication protocol
US9438696B2 (en) 2005-05-25 2016-09-06 Microsoft Technology Licensing, Llc Data communication protocol
US9332089B2 (en) 2005-05-25 2016-05-03 Microsoft Technology Licensing, Llc Data communication coordination with sequence numbers
US9071661B2 (en) 2005-05-25 2015-06-30 Microsoft Technology Licensing, Llc Data communication coordination with sequence numbers
US20080025290A1 (en) * 2006-07-27 2008-01-31 Sharon Barkai Distributed edge network
US8929372B2 (en) * 2007-10-30 2015-01-06 Contextream Ltd. Grid router
US20090109968A1 (en) * 2007-10-30 2009-04-30 Ariel Noy Grid router
US9344369B2 (en) 2008-08-21 2016-05-17 Hewlett Packard Enterprise Development Lp System and methods for distributed quality of service enforcement
US8631277B2 (en) 2010-12-10 2014-01-14 Microsoft Corporation Providing transparent failover in a file system
WO2013002980A3 (en) * 2011-06-29 2013-03-14 Microsoft Corporation Transporting operations of arbitrary size over remote direct memory access
WO2013002980A2 (en) * 2011-06-29 2013-01-03 Microsoft Corporation Transporting operations of arbitrary size over remote direct memory access
US9331955B2 (en) 2011-06-29 2016-05-03 Microsoft Technology Licensing, Llc Transporting operations of arbitrary size over remote direct memory access
JP2014528107A (en) * 2011-06-29 2014-10-23 マイクロソフト コーポレーション The transfer operation of any size by a remote direct memory access
US8856582B2 (en) 2011-06-30 2014-10-07 Microsoft Corporation Transparent failover
US9462039B2 (en) 2011-06-30 2016-10-04 Microsoft Technology Licensing, Llc Transparent failover
US8788579B2 (en) 2011-09-09 2014-07-22 Microsoft Corporation Clustered client failover
CN102708062A (en) * 2012-04-10 2012-10-03 深圳市佳信捷电子有限公司 Multimedia storage method and multimedia storage device based on data block
US9197629B2 (en) 2012-06-26 2015-11-24 International Business Machines Corporation Remote direct memory access authentication of a device
US9111081B2 (en) 2012-06-26 2015-08-18 International Business Machines Corporation Remote direct memory access authentication of a device

Also Published As

Publication number Publication date Type
CA2501521A1 (en) 2005-06-30 application
CN101375263B (en) 2012-06-20 grant
EP1700264A2 (en) 2006-09-13 application
WO2005067430A3 (en) 2008-04-24 application
EP1700264A4 (en) 2011-06-15 application
RU2388039C2 (en) 2010-04-27 grant
US7698361B2 (en) 2010-04-13 grant
RU2009109685A (en) 2010-09-27 application
US20050198113A1 (en) 2005-09-08 application
WO2005067430A2 (en) 2005-07-28 application
KR101084897B1 (en) 2011-11-17 grant
CN101375263A (en) 2009-02-25 application
KR20060112178A (en) 2006-10-31 application
JP4564970B2 (en) 2010-10-20 grant
JP2007527057A (en) 2007-09-20 application
RU2005120695A (en) 2006-01-20 application

Similar Documents

Publication Publication Date Title
US6948004B2 (en) Host-fabric adapter having work queue entry (WQE) ring hardware assist (HWA) mechanism
US7835380B1 (en) Multi-port network interface device with shared processing resources
US6904519B2 (en) Method and computer program product for offloading processing tasks from software to hardware
US7620057B1 (en) Cache line replacement with zero latency
US6401099B1 (en) Asynchronous binding of named objects
US7688838B1 (en) Efficient handling of work requests in a network interface device
US6728265B1 (en) Controlling frame transmission
US6751665B2 (en) Providing window updates from a computer to a network interface device
US7359920B1 (en) Communication protocol for synchronization of personal information management databases
US6775719B1 (en) Host-fabric adapter and method of connecting a host system to a channel-based switched fabric in a data network
US7065706B1 (en) Network router configured for executing network operations based on parsing XML tags in a received XML document
US6370599B1 (en) System for ascertaining task off-load capabilities of a device and enabling selected capabilities and when needed selectively and dynamically requesting the device to perform the task
US7299266B2 (en) Memory management offload for RDMA enabled network adapters
US20050204058A1 (en) Method and apparatus for data re-assembly with a high performance network interface
US7496689B2 (en) TCP/IP offload device
US20040133848A1 (en) System and method for providing and displaying information content
US20020191599A1 (en) Host- fabrec adapter having an efficient multi-tasking pipelined instruction execution micro-controller subsystem for NGIO/infinibandTM applications
US20020156927A1 (en) TCP/IP offload network interface device
US7254696B2 (en) Functional-level instruction-set computer architecture for processing application-layer content-service requests such as file-access requests
US20060034283A1 (en) Method and system for providing direct data placement support
US8010707B2 (en) System and method for network interfacing
US6721806B2 (en) Remote direct memory access enabled network interface controller switchover and switchback support
US20030097481A1 (en) Method and system for performing packet integrity operations using a data movement engine
US20070041383A1 (en) Third party node initiated remote direct memory access
US7409432B1 (en) Efficient process for handover between subnet managers

Legal Events

Date Code Title Description
AS Assignment

Owner name: MICROSOFT TECHNOLOGY LICENSING, LLC, WASHINGTON

Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNOR:MICROSOFT CORPORATION;REEL/FRAME:034766/0001

Effective date: 20141014