Descripciones mejoradas de Syndetics:

This book explains how VI provides a new interface between applications and switched interconnect hardware. Just as Virtual Memory allowed personal computers to break through physical memory limitations, VI solves the long-standing problem of efficiently interfacing general-purpose computers to high-speed switched networks. This book shows how implementing VI can lower overhead by two orders of magnitude and reduce latency for I/O and communications traffic. The Virtual Interface Architecture provides guidance for both developers and users of large clustered systems: Software designers can discover ways to exploit the VI Architecture's capabilities and gain a competitive advantage offered by InfiniBand, which includes VI Architecture concepts. Software designers can learn to develop the necessary drivers, middleware, and applications for VI Architecture-based systems. Chief Technical Officers and system architects of corporate enterprises and Internet Service Providers can gain the background information to plan for deployment of VI Architecture in corporate systems.VI Architecture is related to InfiniBand Architecture, a specification for I/O fabric that includes electrical and mechanical configurations of the physical media, supported operations and semantics, and management interface. InfiniBand does not include an API but does include many concepts of the VI Architecture. The authors were members of the core architecture team that developed VI Architecture within Intel. Both are significant contributors to the VI Architecture, a published specification developed by Intel, Compaq, and Microsoft.

Tabla de contenidos provista por Syndetics

• Preface(p. ix)
• Chapter 1 Introduction(p. 1)
• What's the Problem?(p. 1)
• How VIA Solves the Problem(p. 8)
• Functional Partitioning(p. 10)
• Separation of Control and Data(p. 10)
• Fundamental Concepts of VIA(p. 11)
• Performance Case Study(p. 12)
• Chapter 2 Historical Background and Architectural Roots(p. 17)
• Two Parallel Architectures(p. 17)
• Shared Memory and Multiprocessing(p. 18)
• Distributed Memory and Message Passing(p. 21)
• Combining SMP and Message Passing(p. 24)
• Prior Research(p. 25)
• Cornell U-Net(p. 25)
• Princeton SHRIMP Project(p. 26)
• Scheduled Transfer (ST)(p. 26)
• AM and FM(p. 27)
• Enter VIA: A New Industry Standard(p. 27)
• Chapter 3 Principles of Operation(p. 31)
• The VI Stack(p. 31)
• Data Movement Operations(p. 33)
• Send/Receive(p. 33)
• RDMA-Write(p. 34)
• RDMA-Read(p. 35)
• Using RDMA(p. 35)
• Initiating Data Movement Operations(p. 36)
• Descriptors(p. 36)
• Work Queues(p. 37)
• The Receive Assumption(p. 38)
• Doorbells(p. 38)
• Completions(p. 42)
• Polling for Work Queue Completion(p. 42)
• Blocking On Work Queue Completion(p. 43)
• Completion Queues(p. 44)
• Resource Management(p. 45)
• Opening the VI NIC(p. 45)
• Registering Memory(p. 46)
• Creating Completion Queues(p. 47)
• Creating VIs(p. 47)
• Connections(p. 48)
• Reliability Levels(p. 50)
• Formats of Descriptor Segments(p. 52)
• Control Segment Fields(p. 52)
• Descriptor Address Segment(p. 57)
• Descriptor Data Segment(p. 57)
• Chapter 4 The VIA Specification(p. 59)
• The Partners(p. 59)
• About RDMA-Read(p. 61)
• The VIA Specification(p. 62)
• The VIPL 'Example'(p. 63)
• The VI Developer's Guide(p. 64)
• Chapter 5 Memory Translation and Protection(p. 67)
• Overview(p. 67)
• TPT Format(p. 68)
• Registering Memory(p. 69)
• Memory Handles(p. 69)
• Translating Addresses(p. 70)
• Doorbell Tokens(p. 71)
• Protection Tags(p. 72)
• Page Sizes Greater Than 4K and 64-bit Addresses(p. 73)
• TPT Limitations(p. 74)
• Chapter 6 VIPL Made Simple(p. 77)
• Application Overview(p. 77)
• Client Program(p. 78)
• Server Program(p. 89)
• PuzzleCommon.h(p. 103)
• Chapter 7 Industry Activity(p. 107)
• Standards Activities(p. 107)
• VIDF(p. 107)
• FC-VI(p. 107)
• VHP(p. 108)
• DAT Collaborative(p. 108)
• DAFS(p. 109)
• VI Hardware Implementations(p. 109)
• Compaq ServerNet II(p. 109)
• Emulex/GigaNet(p. 109)
• Troika Networks(p. 110)
• QLogic(p. 110)
• VI Software Stacks(p. 111)
• M-VIA(p. 111)
• MVICH(p. 111)
• Microsoft Winsock Direct(p. 112)
• Chapter 8 Applications(p. 113)
• Scalable Databases(p. 113)
• IBM DB2(p. 114)
• Oracle Database(p. 116)
• Microsoft SQL Server(p. 117)
• Storage(p. 117)
• High Performance Computing(p. 118)
• Future VIA Applications(p. 119)
• Chapter 9 InfiniBand Architecture(p. 123)
• InfiniBand Overview(p. 123)
• Differences Between VIA and IBA(p. 124)
• Scope(p. 124)
• Verbs(p. 124)
• Atomic Operations(p. 125)
• Memory Protection Model(p. 125)
• Descriptors and Work Queues(p. 126)
• Completion Model(p. 126)
• Reliability Levels(p. 127)
• Other(p. 128)
• Chapter 10 Conclusions(p. 129)
• Successes(p. 129)
• Areas for Improvement(p. 130)
• Limited Number of VIs(p. 130)
• Thread Pool Model(p. 131)
• Manager Thread Code(p. 132)
• Worker Thread Code(p. 132)
• Connection Orientation(p. 134)
• Memory Registration(p. 135)
• Receive Assumption, Lack of Flow Control(p. 136)
• Insufficient Standardization(p. 136)
• The Future(p. 137)
• Appendix A VIPL Calls(p. 139)
• Hardware Connection(p. 139)
• VipOpenNic(p. 139)
• VipCloseNic(p. 140)
• Endpoint Creation and Destruction(p. 141)
• VipCreateVi(p. 141)
• VipDestroyVi(p. 142)
• Connection Management(p. 143)
• VipConnectWait(p. 143)
• VipConnectAccept(p. 145)
• VipConnectReject(p. 146)
• VipConnectRequest(p. 147)
• VipDisconnect(p. 149)
• VipConnectPeerRequest(p. 150)
• VipConnectPeerDone(p. 152)
• VipConnectPeerWait(p. 153)
• Memory Protection and Registration(p. 155)
• VipCreatePtag(p. 155)
• VipDestroyPtag(p. 156)
• VipRegisterMem(p. 157)
• VipDeregisterMem(p. 158)
• Data Transfer and Completion Operations(p. 159)
• VipPostSend(p. 159)
• VipSendDone(p. 160)
• VipSendWait(p. 161)
• VipPostRecv(p. 163)
• VipRecvDone(p. 164)
• VipRecvWait(p. 165)
• VipCQDone(p. 166)
• VipCQWait(p. 167)
• VipSendNotify(p. 169)
• VipRecvNotify(p. 171)
• VipCQNotify(p. 173)
• Notify Semantics(p. 175)
• Completion Queue Management(p. 175)
• VipCreateCQ(p. 175)
• VipDestroyCQ(p. 176)
• VipResizeCQ(p. 177)
• Querying Information(p. 178)
• VipQueryNic(p. 178)
• VipSetViAttributes(p. 179)
• VipQueryVi(p. 180)
• VipSetMemAttributes(p. 181)
• VipQueryMem(p. 182)
• VipQuerySystemManagementInfo(p. 183)
• Error Handling(p. 185)
• VipErrorCallback(p. 185)
• Name Service(p. 188)
• VipNSInit(p. 188)
• VipNSGetHostByName(p. 189)
• VipNSGetHostByAddr(p. 191)
• VipNSShutdown(p. 192)
• References(p. 195)
• Publications, Studies, and Projects(p. 195)
• Web Sites(p. 196)
• Index(p. 199)

Notas de autor provistas por Syndetics

Don Cameron is a Senior Staff Architect in the Intel Enterprise Architecture Lab
Greg Regnier is a Principal Engineer in the Intel Enterprise Architecture Lab