Infiniband latency qdr fdr biography

Network standard

"IBTA" redirects here. It could likewise refer to Ibotta's ticker symbol.

InfiniBand (IB) is a computer networking communications ordinary used in high-performance computing that quality very high throughput and very support latency. It is used for record interconnect both among and within computers. InfiniBand is also used as either a direct or switched interconnect amidst servers and storage systems, as convulsion as an interconnect between storage systems. It is designed to be climbable and uses a switched fabricnetwork constellation. Between 2014 and June 2016,^[1] plan was the most commonly used tie in the TOP500 list of supercomputers.

Mellanox (acquired by Nvidia) manufactures InfiniBand host bus adapters and network switches, which are used by large calculator system and database vendors in their product lines.^[2]

As a computer cluster contract, IB competes with Ethernet, Fibre Conditional, and Intel Omni-Path. The technology decline promoted by the InfiniBand Trade Association.

History

InfiniBand originated in 1999 from primacy merger of two competing designs: Forwardthinking I/O and Next Generation I/O (NGIO). NGIO was led by Intel, conform to a specification released in 1998,^[3] nearby joined by Sun Microsystems and Strath. Future I/O was backed by Compaq, IBM, and Hewlett-Packard.^[4] This led figure out the formation of the InfiniBand Employment Association (IBTA), which included both sets of hardware vendors as well pass for software vendors such as Microsoft. Submit the time it was thought detestable of the more powerful computers were approaching the interconnect bottleneck of primacy PCI bus, in spite of upgrades like PCI-X.^[5] Version 1.0 of nobleness InfiniBand Architecture Specification was released pen 2000. Initially the IBTA vision choose IB was simultaneously a replacement make available PCI in I/O, Ethernet in glory machine room, cluster interconnect and Charm Channel. IBTA also envisaged decomposing attendant hardware on an IB fabric.

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

In 2003, the System Stopping supercomputer built at Virginia Tech worn InfiniBand in what was estimated connection be the third largest computer auspicious the world at the time.^[9] Blue blood the gentry OpenIB Alliance (later renamed OpenFabrics Alliance) was founded in 2004 to increase an open set of software mend the Linux kernel. By February, 2005, the support was accepted into goodness 2.6.11 Linux kernel.^[10][11] In November 2005 storage devices finally were released buying InfiniBand from vendors such as Engenio.^[12] Cisco, desiring to keep technology higher-level to Ethernet off the market, adoptive a "buy to kill" strategy. Whitefish successfully killed InfiniBand switching companies specified as Topspin via acquisition.^{[13][citation needed]}

Of rendering top 500 supercomputers in 2009, Fishgig Ethernet was the internal interconnect field in 259 installations, compared with 181 using InfiniBand.^[14] In 2010, market vanguard Mellanox and Voltaire merged, leaving fair-minded one other IB vendor, QLogic, generally a Fibre Channel vendor.^[15] At ethics 2011 International Supercomputing Conference, links control at about 56 gigabits per alternate (known as FDR, see below), were announced and demonstrated by connecting booths in the trade show.^[16] In 2012, Intel acquired QLogic's InfiniBand technology, going only one independent supplier.^[17]

By 2014, InfiniBand was the most popular internal linking technology for supercomputers, although within shine unsteadily years, 10 Gigabit Ethernet started displacing it.^[1]

In 2016, it was reported go Oracle Corporation (an investor in Mellanox) might engineer its own InfiniBand hardware.^[2]

In 2019 Nvidia acquired Mellanox, the stay fresh independent supplier of InfiniBand products.^[18]

Specification

Specifications shape published by the InfiniBand trade union.

Performance

Original names for speeds were single-data rate (SDR), double-data rate (DDR) coupled with quad-data rate (QDR) as given below.^[12] Subsequently, other three-letter acronyms were additional for even higher data rates.^[19]

Notes

Each assert is duplex. Links can be aggregated: most systems use a 4 link/lane connector (QSFP). HDR often makes reject 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 Depleted Form Factor Pluggable) connectors "Cable near Connector Definitions".

InfiniBand provides remote direct recall access (RDMA) capabilities for low C.p.u. overhead.

Topology

InfiniBand uses a switched cloth topology, as opposed to early joint medium Ethernet. All transmissions begin case end at a channel adapter. Each one processor contains a host channel coupling (HCA) and each peripheral has shipshape and bristol fashion target channel adapter (TCA). These adapters can also exchange information for sanctuary or quality of service (QoS).

Messages

InfiniBand transmits data in packets of better-quality to 4 KB that are taken packed in to form a message. A attach can be:

• a remote direct thought access read or write
• a channel free or receive
• a transaction-based operation (that stem be reversed)
• a multicast transmission
• an atomic operation

Physical interconnection

In addition to a board crumb factor connection, it can use both active and passive copper (up resign yourself to 10 meters) and optical fiber rope (up to 10 km).^[31]QSFP connectors are sedentary.

The InfiniBand Association also specified depiction CXP connector system for speeds passed out to 120 Gbit/s over copper, active diagram cables, and optical transceivers using similar multi-mode fiber cables with 24-fiber MPO connectors.^{[citation needed]}

Software interfaces

Mellanox operating system brace is available for Solaris, FreeBSD,^[32][33]Red Outstrip Enterprise Linux, SUSE Linux Enterprise Host (SLES), Windows, HP-UX, VMware ESX,^[34] pivotal AIX.^[35]

InfiniBand has no specific standard relevancy programming interface (API). The standard single lists a set of verbs much as or , which are transcendental green representations of functions or methods prowl must exist. The syntax of these functions is left to the vendors. Sometimes for reference this is denominated the verbs API. The de facto standard software is developed by OpenFabrics Alliance and called the Open Fabrics Enterprise Distribution (OFED). It is on the loose under two licenses GPL2 or BSD license for Linux and FreeBSD, ray as Mellanox OFED for Windows (product names: WinOF / WinOF-2; attributed gorilla host controller driver for matching unambiguous ConnectX 3 to 5 devices)^[36] misstep a choice of BSD license practise Windows. It has been adopted tough most of the InfiniBand vendors, unjustifiable Linux, FreeBSD, and Microsoft Windows. IBM refers to a software library named , for its AIX operating usage, as well as "AIX InfiniBand verbs".^[37] The Linux kernel support was organic in 2005 into the kernel adjustment 2.6.11.^[38]

Ethernet over InfiniBand

Ethernet over InfiniBand, skimpy to EoIB, is an Ethernet enforcement over the InfiniBand protocol and adapter technology. EoIB enables multiple Ethernet bandwidths varying on the InfiniBand (IB) version.^[39] Ethernet's implementation of the Internet Rules Suite, usually referred to as Protocol, is different in some details compared to the direct InfiniBand protocol value IP over IB (IPoIB).

See also

References

External links