An Integrated Tutorial on InfiniBand, Verbs, and MPI

MacArthur, Patrick; Liu, Qian; Russell, Robert D.; Mizero, Fabrice; Veeraraghavan, Malathi; Dennis, John M.

doi:10.1109/comst.2017.2746083

Cited by 20 publications

(11 citation statements)

References 80 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The InfiniBand Trade Association was aware of this, when it said that the "present [router] specification does not cover the routing protocol nor the messages exchanged between routers." True routers are optional in InfiniBand networks; InfiniBand has been successfully utilized without routers in a production environment between two distant clusters [10]. Nonetheless, InfiniBand is fundamentally a data center technology that is not typically deployed in a network demilitarized zone unless firewalls or other similar access-controlled layers are placed in front of the InfiniBand fabric [12].…”

Section: Infinibandmentioning

confidence: 99%

“…Like the Transmission Control Protocol/Internet Protocol (TCP/IP) stack, the InfiniBand protocol stack is based on the seven-layer Open Systems Interconnection (OSI) model. Layer-2 addressing is done via a local identifier (LID), which is dynamically assigned by the subnet manager [10]. The local identifier is a 16-bit value, so a single subnet can support up to 65 K hosts.…”

Section: Infiniband Terminologymentioning

confidence: 99%

“…InfiniBand is more efficient at avoiding congestion due to its priority-based flow control. The Ethernet pause frame "stops all traffic indiscriminately" whereas InfiniBand "strictly avoids packet loss by employing link-by-link flow control, which prevents a data packet from being sent from one end of a link if there is insufficient space to receive the packet at the other end of [the] link" [10]. More importantly, the InfiniBand receiver host channel adapter drops all outof-order packets because it is an error condition as far as the InfiniBand receiver is concerned.…”

Section: Infiniband Terminologymentioning

confidence: 99%

“…Infini-Band connections are established via queue pairs. MacArthur et al [10] state that queue pairs are "comparable to port numbers in TCP and UDP, and make it possible to multiplex many independent flows to the same destination [host channel adapter]." The union ibv gid *gid output parameter of the ibv query gid() function or the return value of the ibv get device guid() method could be changed later.…”

Section: Address Spoofingmentioning

confidence: 99%

See 3 more Smart Citations

Vulnerability Assessment of InfiniBand Networking

Schmitt

Graham

Sweeney

et al. 2019

IFIP Advances in Information and Communication Technology

View full text Add to dashboard Cite

InfiniBand is an input/output interconnect technology for high performance computing clusters -it is employed in more than one-quarter of the world's 500 fastest computer systems. Although InfiniBand was created to provide extremely low network latency with high quality of service, the cyber security aspects of InfiniBand have yet to be investigated thoroughly. The InfiniBand architecture was designed as a data center technology that is logically separated from the Internet, so defensive mechanisms such as packet encryption were not implemented. The security community does not appear to have taken an interest in InfiniBand, but this is likely to change as attackers branch out from traditional computing devices. This chapter discusses the security implications of InfiniBand features and presents a technical cyber vulnerability assessment.

show abstract

Section: Infinibandmentioning

confidence: 99%

Section: Infiniband Terminologymentioning

confidence: 99%

Section: Infiniband Terminologymentioning

confidence: 99%

Section: Address Spoofingmentioning

confidence: 99%

See 2 more Smart Citations

Vulnerability Assessment of InfiniBand Networking

Schmitt

Graham

Sweeney

et al. 2019

IFIP Advances in Information and Communication Technology

View full text Add to dashboard Cite

show abstract

“…Current research has improved the efficiency of erasure codes in data recovery in several ways, focusing on encoding design, network transmission optimization, and decoding acceleration. With continuous breakthroughs in encoding design and the application of high-performance network technologies such as InfiniBand and RDMA in storage systems [6,7], issues related to encoding design and network bottlenecks have gradually been resolved, making the decoding efficiency of erasure codes a key factor affecting data recovery efficiency.…”

Section: Introductionmentioning

confidence: 99%

Symmetrical Data Recovery: FPGA-Based Multi-Dimensional Elastic Recovery Acceleration for Multiple Block Failures in Ceph Systems

Lei,

Wang,

Chen

et al. 2024

Symmetry

View full text Add to dashboard Cite

In the realm of Ceph distributed storage systems, ensuring swift and symmetrical data recovery during severe data corruption scenarios is pivotal for data reliability and system stability. This paper introduces an innovative FPGA-based Multi-Dimensional Elastic Recovery Acceleration method, termed AMDER-Ceph. Utilizing FPGA technology, this method is a pioneer in accelerating erasure code data recovery within such systems symmetrically. By harnessing the parallel computing power of FPGAs and optimizing Cauchy matrix binary operations, AMDER-Ceph significantly enhances data recovery speed and efficiency symmetrically. Our evaluations in real-world Ceph environments show that AMDER-Ceph achieves up to 4.84 times faster performance compared with traditional methods, especially evident in the standard 4 MB block size configurations of Ceph systems.

show abstract