InfiniBand

Last updated
InfiniBand Trade Association
Formation1999
TypeIndustry trade group
PurposePromoting InfiniBand
Headquarters Beaverton, Oregon, U.S.
Website infinibandta.org

InfiniBand (IB) is a computer networking communications standard used in high-performance computing that features very high throughput and very low latency. It is used for data interconnect both among and within computers. InfiniBand is also used as either a direct or switched interconnect between servers and storage systems, as well as an interconnect between storage systems. It is designed to be scalable and uses a switched fabric network topology. Between 2014 and June 2016, [1] it was the most commonly used interconnect in the TOP500 list of supercomputers.

Contents

Mellanox (acquired by Nvidia) manufactures InfiniBand host bus adapters and network switches, which are used by large computer system and database vendors in their product lines. [2] As a computer cluster interconnect, IB competes with Ethernet, Fibre Channel, and Intel Omni-Path. The technology is promoted by the InfiniBand Trade Association.

History

InfiniBand originated in 1999 from the merger of two competing designs: Future I/O and Next Generation I/O (NGIO). NGIO was led by Intel, with a specification released in 1998, [3] and joined by Sun Microsystems and Dell. Future I/O was backed by Compaq, IBM, and Hewlett-Packard. [4] This led to the formation of the InfiniBand Trade Association (IBTA), which included both sets of hardware vendors as well as software vendors such as Microsoft. At the time it was thought some of the more powerful computers were approaching the interconnect bottleneck of the PCI bus, in spite of upgrades like PCI-X. [5] Version 1.0 of the InfiniBand Architecture Specification was released in 2000. Initially the IBTA vision for IB was simultaneously a replacement for PCI in I/O, Ethernet in the machine room, cluster interconnect and Fibre Channel. IBTA also envisaged decomposing server hardware on an IB fabric.

Mellanox had been founded in 1999 to develop NGIO technology, but by 2001 shipped an InfiniBand product line called InfiniBridge at 10 Gbit/second speeds. [6] Following the burst of the dot-com bubble there was hesitation in the industry to invest in such a far-reaching technology jump. [7] By 2002, Intel announced that instead of shipping IB integrated circuits ("chips"), it would focus on developing PCI Express, and Microsoft discontinued IB development in favor of extending Ethernet. Sun Microsystems and Hitachi continued to support IB. [8]

In 2003, the System X supercomputer built at Virginia Tech used InfiniBand in what was estimated to be the third largest computer in the world at the time. [9] The OpenIB Alliance (later renamed OpenFabrics Alliance) was founded in 2004 to develop an open set of software for the Linux kernel. By February, 2005, the support was accepted into the 2.6.11 Linux kernel. [10] [11] In November 2005 storage devices finally were released using InfiniBand from vendors such as Engenio. [12]

Of the top 500 supercomputers in 2009, Gigabit Ethernet was the internal interconnect technology in 259 installations, compared with 181 using InfiniBand. [13] In 2010, market leaders Mellanox and Voltaire merged, leaving just one other IB vendor, QLogic, primarily a Fibre Channel vendor. [14] At the 2011 International Supercomputing Conference, links running at about 56 gigabits per second (known as FDR, see below), were announced and demonstrated by connecting booths in the trade show. [15] In 2012, Intel acquired QLogic's InfiniBand technology, leaving only one independent supplier. [16]

By 2014, InfiniBand was the most popular internal connection technology for supercomputers, although within two years, 10 Gigabit Ethernet started displacing it. [1] In 2016, it was reported that Oracle Corporation (an investor in Mellanox) might engineer its own InfiniBand hardware. [2] In 2019 Nvidia acquired Mellanox, the last independent supplier of InfiniBand products. [17]

Specification

Specifications are published by the InfiniBand trade association.

Performance

Original names for speeds were single-data rate (SDR), double-data rate (DDR) and quad-data rate (QDR) as given below. [12] Subsequently, other three-letter acronyms were added for even higher data rates. [18]

InfiniBand unidirectional data rates
Year [19] Line codeSignaling rate (Gbit/s) Throughput (Gbit/s) [20] Adapter latency (μs) [21]
1x4x8x12x
SDR2001, 2003 NRZ 8b/10b [22] 2.52816245
DDR2005541632482.5
QDR20071083264961.3
FDR102011 64b/66b 10.3125 [23] 1040801200.7
FDR201114.0625 [24] [18] 13.6454.54109.08163.640.7
EDR2014 [25] 25.78125251002003000.5
HDR2018 [25] PAM4 53.125 [26] 50200400600<0.6 [27]
NDR2022 [25] 256b/257b [lower-roman 1] 106.25 [28] 1004008001200t.b.d.
XDR2024 [29] t.b.dt.b.d20020080016002400
GDRt.b.d.400400160032004800
Notes
  1. ↑ Using Reed-Solomon forward error correction

Each link is duplex. Links can be aggregated: most systems use a 4 link/lane connector (QSFP). HDR often makes use of 2x links (aka HDR100, 100 Gb link using 2 lanes of HDR, while still using a QSFP connector). 8x is called for with NDR switch ports using OSFP (Octal Small Form Factor Pluggable) connectors "Cable and Connector Definitions".

InfiniBand provides remote direct memory access (RDMA) capabilities for low CPU overhead.

Topology

InfiniBand uses a switched fabric topology, as opposed to early shared medium Ethernet. All transmissions begin or end at a channel adapter. Each processor contains a host channel adapter (HCA) and each peripheral has a target channel adapter (TCA). These adapters can also exchange information for security or quality of service (QoS).

Messages

InfiniBand transmits data in packets of up to 4 KB that are taken together to form a message. A message can be:

Physical interconnection

InfiniBand switch with CX4/SFF-8470 connectors Infinibandport.jpg
InfiniBand switch with CX4/SFF-8470 connectors

In addition to a board form factor connection, it can use both active and passive copper (up to 10 meters) and optical fiber cable (up to 10 km). [30] QSFP connectors are used.

The InfiniBand Association also specified the CXP connector system for speeds up to 120 Gbit/s over copper, active optical cables, and optical transceivers using parallel multi-mode fiber cables with 24-fiber MPO connectors.[ citation needed ]

Software interfaces

Mellanox operating system support is available for Solaris, FreeBSD, [31] [32] Red Hat Enterprise Linux, SUSE Linux Enterprise Server (SLES), Windows, HP-UX, VMware ESX, [33] and AIX. [34]

InfiniBand has no specific standard application programming interface (API). The standard only lists a set of verbs such as ibv_open_device or ibv_post_send, which are abstract representations of functions or methods that must exist. The syntax of these functions is left to the vendors. Sometimes for reference this is called the verbs API. The de facto standard software is developed by OpenFabrics Alliance and called the Open Fabrics Enterprise Distribution (OFED). It is released under two licenses GPL2 or BSD license for Linux and FreeBSD, and as Mellanox OFED for Windows (product names: WinOF / WinOF-2; attributed as host controller driver for matching specific ConnectX 3 to 5 devices) [35] under a choice of BSD license for Windows. It has been adopted by most of the InfiniBand vendors, for Linux, FreeBSD, and Microsoft Windows. IBM refers to a software library called libibverbs, for its AIX operating system, as well as "AIX InfiniBand verbs". [36] The Linux kernel support was integrated in 2005 into the kernel version 2.6.11. [37]

Ethernet over InfiniBand

Ethernet over InfiniBand, abbreviated to EoIB, is an Ethernet implementation over the InfiniBand protocol and connector technology. EoIB enables multiple Ethernet bandwidths varying on the InfiniBand (IB) version. [38] Ethernet's implementation of the Internet Protocol Suite, usually referred to as TCP/IP, is different in some details compared to the direct InfiniBand protocol in IP over IB (IPoIB).

Ethernet over InfiniBand performance
TypeLanesBandwidth (Gbit/s)Compatible Ethernet type(s)Compatible Ethernet quantity
SDR12.5GbE to 2.5 GbE2 × GbE to 1 × 2.5 GbE
410GbE to 10 GbE10 × GbE to 1 × 10 GbE
820GbE to 10 GbE20 × GbE to 2 × 10 GbE
1230GbE to 25 GbE30 × GbE to 1 × 25 GbE + 1 × 5 GbE
DDR15GbE to 5 GbE5 × GbE to 1 × 5 GbE
420GbE to 10 GbE20 × GbE to 2 × 10 GbE
840GbE to 40 GbE40 × GbE to 1 × 40 GbE
1260GbE to 50 GbE60 × GbE to 1 × 50 GbE + 1 × 10 GbE
QDR110GbE to 10 GbE10 × GbE to 1 × 10 GbE
440GbE to 40 GbE40 × GbE to 1 × 40 GbE

See also

Related Research Articles

Myrinet, ANSI/VITA 26-1998, is a high-speed local area networking system designed by the company Myricom to be used as an interconnect between multiple machines to form computer clusters.

<span class="mw-page-title-main">PCI Express</span> Computer expansion bus standard

PCI Express, officially abbreviated as PCIe or PCI-e, is a high-speed serial computer expansion bus standard, designed to replace the older PCI, PCI-X and AGP bus standards. It is the common motherboard interface for personal computers' graphics cards, sound cards, hard disk drive host adapters, SSDs, Wi-Fi and Ethernet hardware connections. PCIe has numerous improvements over the older standards, including higher maximum system bus throughput, lower I/O pin count and smaller physical footprint, better performance scaling for bus devices, a more detailed error detection and reporting mechanism, and native hot-swap functionality. More recent revisions of the PCIe standard provide hardware support for I/O virtualization.

<span class="mw-page-title-main">HIPPI</span>

HIPPI, short for High Performance Parallel Interface, is a computer bus for the attachment of high speed storage devices to supercomputers, in a point-to-point link. It was popular in the late 1980s and into the mid-to-late 1990s, but has since been replaced by ever-faster standard interfaces like Fibre Channel and 10 Gigabit Ethernet.

<span class="mw-page-title-main">Quadrics (company)</span>

Quadrics was a supercomputer company formed in 1996 as a joint venture between Alenia Spazio and the technical team from Meiko Scientific. They produced hardware and software for clustering commodity computer systems into massively parallel systems. Their highpoint was in June 2003 when six out of the ten fastest supercomputers in the world were based on Quadrics' interconnect. They officially closed on June 29, 2009.

<span class="mw-page-title-main">Small Form-factor Pluggable</span> Modular communications interface

Small Form-factor Pluggable (SFP) is a compact, hot-pluggable network interface module format used for both telecommunication and data communications applications. An SFP interface on networking hardware is a modular slot for a media-specific transceiver, such as for a fiber-optic cable or a copper cable. The advantage of using SFPs compared to fixed interfaces is that individual ports can be equipped with different types of transceivers as required, with the majority including optical line terminals, network cards, switches and routers.

TCP offload engine (TOE) is a technology used in some network interface cards (NIC) to offload processing of the entire TCP/IP stack to the network controller. It is primarily used with high-speed network interfaces, such as gigabit Ethernet and 10 Gigabit Ethernet, where processing overhead of the network stack becomes significant. TOEs are often used as a way to reduce the overhead associated with Internet Protocol (IP) storage protocols such as iSCSI and Network File System (NFS).

In computing, remote direct memory access (RDMA) is a direct memory access from the memory of one computer into that of another without involving either one's operating system. This permits high-throughput, low-latency networking, which is especially useful in massively parallel computer clusters.

<span class="mw-page-title-main">IBM BladeCenter</span> Blade server architecture by IBM

The IBM BladeCenter was IBM's blade server architecture, until it was replaced by Flex System in 2012. The x86 division was later sold to Lenovo in 2014.

40 Gigabit Ethernet (40GbE) and 100 Gigabit Ethernet (100GbE) are groups of computer networking technologies for transmitting Ethernet frames at rates of 40 and 100 gigabits per second (Gbit/s), respectively. These technologies offer significantly higher speeds than 10 Gigabit Ethernet. The technology was first defined by the IEEE 802.3ba-2010 standard and later by the 802.3bg-2011, 802.3bj-2014, 802.3bm-2015, and 802.3cd-2018 standards. The first succeeding Terabit Ethernet specifications were approved in 2017.

<span class="mw-page-title-main">OpenFabrics Alliance</span> Organization

The OpenFabrics Alliance is a non-profit organization that promotes remote direct memory access (RDMA) switched fabric technologies for server and storage connectivity. These high-speed data-transport technologies are used in high-performance computing facilities, in research and various industries.

The Cray CX1 is a deskside workstation designed by Cray Inc., based on the x86-64 processor architecture. It was launched on September 16, 2008, and was discontinued in early 2012. It comprises a single chassis blade server design that supports a maximum of eight modular single-width blades, giving up to 96 processor cores. Computational load can be run independently on each blade and/or combined using clustering techniques.

<span class="mw-page-title-main">HPE BladeSystem</span> Line of blade server machines by Hewlett Packard Enterprise

BladeSystem is a line of blade server machines from Hewlett Packard Enterprise that was introduced in June 2006.

<span class="mw-page-title-main">IPtronics</span> Danish fabless semiconductor company

IPtronics was a fabless semiconductor company headquartered in Copenhagen, Denmark. Its products include integrated circuits for parallel optical interconnect applications intended for the computer, storage and communication industries. IPtronics' design center is certified by STMicroelectronics, which is also their semiconductor foundry partner. In June 2013, IPtronics was acquired by Mellanox Technologies.

NVM Express (NVMe) or Non-Volatile Memory Host Controller Interface Specification (NVMHCIS) is an open, logical-device interface specification for accessing a computer's non-volatile storage media usually attached via the PCI Express bus. The initial NVM stands for non-volatile memory, which is often NAND flash memory that comes in several physical form factors, including solid-state drives (SSDs), PCIe add-in cards, and M.2 cards, the successor to mSATA cards. NVM Express, as a logical-device interface, has been designed to capitalize on the low latency and internal parallelism of solid-state storage devices.

<span class="mw-page-title-main">LIO (SCSI target)</span> Open-source version of SCSI target

In computing, Linux-IO (LIO) Target is an open-source implementation of the SCSI target that has become the standard one included in the Linux kernel. Internally, LIO does not initiate sessions, but instead provides one or more Logical Unit Numbers (LUNs), waits for SCSI commands from a SCSI initiator, and performs required input/output data transfers. LIO supports common storage fabrics, including FCoE, Fibre Channel, IEEE 1394, iSCSI, iSCSI Extensions for RDMA (iSER), SCSI RDMA Protocol (SRP) and USB. It is included in most Linux distributions; native support for LIO in QEMU/KVM, libvirt, and OpenStack makes LIO also a storage option for cloud deployments.

<span class="mw-page-title-main">Dell M1000e</span> Server computer

The Dell blade server products are built around their M1000e enclosure that can hold their server blades, an embedded EqualLogic iSCSI storage area network and I/O modules including Ethernet, Fibre Channel and InfiniBand switches.

RDMA over Converged Ethernet (RoCE) or InfiniBand over Ethernet (IBoE) is a network protocol which allows remote direct memory access (RDMA) over an Ethernet network. It does this by encapsulating an InfiniBand (IB) transport packet over Ethernet. There are multiple RoCE versions. RoCE v1 is an Ethernet link layer protocol and hence allows communication between any two hosts in the same Ethernet broadcast domain. RoCE v2 is an internet layer protocol which means that RoCE v2 packets can be routed. Although the RoCE protocol benefits from the characteristics of a converged Ethernet network, the protocol can also be used on a traditional or non-converged Ethernet network.

<span class="mw-page-title-main">Mellanox Technologies</span> Israeli-American multinational supplier of computer networking products

Mellanox Technologies Ltd. was an Israeli-American multinational supplier of computer networking products based on InfiniBand and Ethernet technology. Mellanox offered adapters, switches, software, cables and silicon for markets including high-performance computing, data centers, cloud computing, computer data storage and financial services.

<span class="mw-page-title-main">QPACE2</span> Massively parallel and scalable supercomputer

QPACE 2 is a massively parallel and scalable supercomputer. It was designed for applications in lattice quantum chromodynamics but is also suitable for a wider range of applications..

<span class="mw-page-title-main">Summit (supercomputer)</span> Supercomputer developed by IBM

Summit or OLCF-4 is a supercomputer developed by IBM for use at Oak Ridge Leadership Computing Facility (OLCF), a facility at the Oak Ridge National Laboratory, capable of 200 petaFLOPS thus making it the 5th fastest supercomputer in the world after Frontier (OLCF-5), Fugaku, LUMI, and Leonardo, with Frontier being the fastest. It held the number 1 position from November 2018 to June 2020. Its current LINPACK benchmark is clocked at 148.6 petaFLOPS.

References

  1. 1 2 "Highlights– June 2016". Top500.Org. June 2016. Retrieved September 26, 2021. InfiniBand technology is now found on 205 systems, down from 235 systems, and is now the second most-used internal system interconnect technology. Gigabit Ethernet has risen to 218 systems up from 182 systems, in large part thanks to 176 systems now using 10G interfaces.
  2. 1 2 Timothy Prickett Morgan (February 23, 2016). "Oracle Engineers Its Own InfiniBand Interconnects". The Next Platform. Retrieved September 26, 2021.
  3. ↑ Scott Bekker (November 11, 1998). "Intel Introduces Next Generation I/O for Computing Servers". Redmond Channel Partner. Retrieved September 28, 2021.
  4. ↑ Will Wade (August 31, 1999). "Warring NGIO and Future I/O groups to merge". EE Times. Retrieved September 26, 2021.
  5. ↑ Pentakalos, Odysseas. "An Introduction to the InfiniBand Architecture". O'Reilly. Retrieved 28 July 2014.
  6. ↑ "Timeline". Mellanox Technologies. Retrieved September 26, 2021.
  7. ↑ Kim, Ted. "Brief History of InfiniBand: Hype to Pragmatism". Oracle. Archived from the original on 8 August 2014. Retrieved September 28, 2021.
  8. ↑ Computerwire (December 2, 2002). "Sun confirms commitment to InfiniBand". The Register. Retrieved September 26, 2021.
  9. ↑ "Virginia Tech Builds 10 TeraFlop Computer". R&D World. November 30, 2003. Retrieved September 28, 2021.
  10. ↑ Sean Michael Kerner (February 24, 2005). "Linux Kernel 2.6.11 Supports InfiniBand". Internet News. Retrieved September 28, 2021.
  11. ↑ OpenIB Alliance (January 21, 2005). "OpenIB Alliance Achieves Acceptance By Kernel.org". Press release. Retrieved September 28, 2021.
  12. 1 2 Ann Silverthorn (January 12, 2006), "Is InfiniBand poised for a comeback?", Infostor, 10 (2), retrieved September 28, 2021
  13. ↑ Lawson, Stephen (November 16, 2009). "Two rival supercomputers duke it out for top spot". Computerworld. Retrieved September 29, 2021.
  14. ↑ Raffo, Dave. "Largest InfiniBand vendors merge; eye converged networks". Archived from the original on 1 July 2017. Retrieved 29 July 2014.
  15. ↑ Mikael Ricknäs (June 20, 2011). "Mellanox Demos Souped-Up Version of InfiniBand". CIO. Archived from the original on April 6, 2012. Retrieved September 30, 2021.
  16. ↑ Michael Feldman (January 23, 2012). "Intel Snaps Up InfiniBand Technology, Product Line from QLogic". HPCwire. Retrieved September 29, 2021.
  17. ↑ "Nvidia to Acquire Mellanox for $6.9 Billion". Press release. March 11, 2019. Retrieved September 26, 2021.
  18. 1 2 "FDR InfiniBand Fact Sheet". InfiniBand Trade Association. November 11, 2021. Retrieved September 30, 2021.
  19. ↑ Panda, Dhabaleswar K.; Sayantan Sur (2011). "Network Speed Acceleration with IB and HSE" (PDF). Designing Cloud and Grid Computing Systems with InfiniBand and High-Speed Ethernet. Newport Beach, CA, USA: CCGrid 2011. p. 23. Retrieved 13 September 2014.
  20. ↑ "InfiniBand Roadmap: IBTA - InfiniBand Trade Association". Archived from the original on 2011-09-29. Retrieved 2009-10-27.
  21. ↑ http://www.hpcadvisorycouncil.com/events/2014/swiss-workshop/presos/Day_1/1_Mellanox.pdf // Mellanox
  22. ↑ "InfiniBand Types and Speeds".
  23. ↑ "Interfaces". NVIDIA Docs. Retrieved 2023-11-12. FDR10 is a non-standard InfiniBand data rate, where each lane of a 4X port runs a bit rate of 10.3125 Gbit/s with a 64b/66b encoding, resulting in an effective bandwidth of 40 Gbit/s. FDR10 supports 20% more bandwidth over QDR due to better encoding rate.
  24. ↑ "324-Port InfiniBand FDR SwitchX® Switch Platform Hardware User Manual" (PDF). nVidia. 2018-04-29. section 1.2. Retrieved 2023-11-12. InfiniBand FDR and FDR10 Overview [...] FDR, standard InfiniBand data rate, where each lane of a 4X port runs a bit rate of 14.0625 Gbit/s with a 64b/66b encoding, resulting in an effective bandwidth of 54.54 Gbit/s. The FDR physical layer is an IBTA specified physical layer using different block types, deskew mechanism and framing rules. The SX6518 switch also supports FDR10, a non-standard InfiniBand data rate, where each lane of a 4X port runs a bit rate of 10.3125 Gbit/s with a 64b/66b encoding, resulting in an effective bandwidth of 40 Gbit/s.
  25. 1 2 3 "InfiniBand Roadmap - Advancing InfiniBand". InfiniBand Trade Association.
  26. ↑ "Introduction". NVIDIA Docs. Retrieved 2023-11-12.
  27. ↑ https://www.mellanox.com/files/doc-2020/pb-connectx-6-vpi-card.pdf [ bare URL PDF ]
  28. ↑ "Introduction". NVIDIA Docs. Retrieved 2023-11-12.
  29. ↑ "NVIDIA Announces New Switches Optimized for Trillion-Parameter GPU Computing and AI Infrastructure". NVIDIA Newsroom. Retrieved 2024-03-19.
  30. ↑ "Specification FAQ". ITA. Archived from the original on 24 November 2016. Retrieved 30 July 2014.
  31. ↑ "Mellanox OFED for FreeBSD". Mellanox. Retrieved 19 September 2018.
  32. ↑ Mellanox Technologies (3 December 2015). "FreeBSD Kernel Interfaces Manual, mlx5en". FreeBSD Man Pages. FreeBSD. Retrieved 19 September 2018.
  33. ↑ "InfiniBand Cards - Overview". Mellanox. Retrieved 30 July 2014.
  34. ↑ "Implementing InfiniBand on IBM System p (IBM Redbook SG24-7351-00)" (PDF).
  35. ↑ Mellanox OFED for Windows - WinOF / WinOF-2
  36. ↑ "Verbs API". IBM AIX 7.1 documentation. 2020. Retrieved September 26, 2021.
  37. ↑ Dotan Barak (March 11, 2014). "Verbs programming tutorial" (PDF). OpenSHEM, 2014. Mellanox. Retrieved September 26, 2021.
  38. ↑ "10 Advantages of InfiniBand". NADDOD. Retrieved January 28, 2023.
This page is based on this Wikipedia article
Text is available under the CC BY-SA 4.0 license; additional terms may apply.
Images, videos and audio are available under their respective licenses.