• Journal of KIISE:Computer Systems and Theory
• /
• v.31 no.11
• /
• pp.635-642
• /
• 2004

The traditional communication protocols such as TCP/IP are not suitable for PC cluster systems because of their high software processing overhead. To eliminate this overhead, industry leaders have defined the Virtual Interface Architecture (VIA). VIA provides two different data transfer mechanisms, a traditional Send/Receive model and the Remote Direct Memory Access (RDMA) model. RDMA is extremely efficient way to reduce software overhead because it can bypass the OS and use the network interface controller (NIC) directly for communication, also bypass the CPU on the remote host. In this paper, we have implemented VIA-based RDMA mechanism in hardware. Compared to the traditional Send/Receive model, the RDMA mechanism improves latency and bandwidth. Our RDMA mechanism can also communicate without using remote CPU cycles. Our experimental results show a minimum latency of 12.5${\mu}\textrm{s}$ and a maximum bandwidth of 95.5MB/s. As a result, our RDMA mechanism allows PC cluster systems to have a high performance communication method.

• KIPS Transactions on Software and Data Engineering
• /
• v.7 no.8
• /
• pp.297-306
• /
• 2018

In this paper, we address how to apply Apache Storm, a distributed stream processing framework, to InfiniBand, a high performance communication device. An easy way to run Storm on InfiniBand is to simply use IPoIP (IP over InfiniBand). However, this method causes a serious CPU load on the node, which is caused by frequent context switches and buffer copies. To solve this problem, we propose a new communication method using InfiniBand's Remote Direct Memory Access (RDMA) function in Storm. First, we design and implement RJ-Netty (RDMA/JXIO Netty), a new framework that replaces Netty, the legacy framework, to exploit RDMA functionality. Second, we reimplement the related classes so that Storm can use both existing Netty and new RJ-Netty. Third, we extend the JXIO server functionality so as to support multi-threading to maximize the performance of RJ-Netty. Experimental results show that the proposed RJ-Netty significantly reduces CPU load while improving message throughput compared to IPoIB as well as Ethernet. This paper is the first attempt to run Apache Storm on InfiniBand, and we believe that it is an excellent research result that improves the performance of Storm by using InfiniBand RDMA.

• Proceedings of the Korean Information Science Society Conference
• /
• 2005.11a
• /
• pp.997-999
• /
• 2005

기가비트의 속도를 넘는 고속 네트워크상에서 TCP/IP를 사용할 경우 호스트 CPU에서 TCP/IP를 처리하는데 많은 부하가 발생한다. 이러한 문제를 해결하기 위해 최근 네트워크 어댑터에서 TCP/IP를 처리하는 TCP/IP Offload Engine(TOE)에 대한 연구가 활발히 진행되고 있다. 본 연구에서는 임베디드 시스템에 리눅스를 사용하여, TOE와 Remote Direct Memory Access(RDMA) 메커니즘을 구현하였고 그 동작 방식을 보인다. 실험을 통해 구현된 메커니즘들을 검증하였으며, 각 구간별로 소요시간을 측정하였다. 본 논문에서는 이러한 실험 결과를 바탕으로, 추후 기가비트 환경에 적합한 TOE 및 RDMA 메커니즘의 구한 방안을 제안한다.

• 한국정보통신설비학회:학술대회논문집
• /
• 2006.08a
• /
• pp.77-81
• /
• 2006

Because of the rapid increasing of network user, there are some problems to tolerate the network overhead. Recently, the research and technology of the user-level for high performance and low latency than TCP/IP which relied upon the kernel for processing the messages. For example, there is an Infiniband technology. The Infiniband Trade Association (IBTA) has been proposed as an industry standard for both communication between processing node and I/O devices and for inter-processor communication. It replaces the traditional bus-based interconnect with a switch-based network for connecting processing node and I/O devices. Also Infiniband uses RDMA (Remote DMA) for low latency of CPU and OS to communicate between Remote nodes. In this paper, we develop the SRP (SCSI RDMA Protocol) which is Storage Access Protocol on Infiniband network. And will compare to FC (Fibre Channle) based I-SCSI (Internet SCSI) that it is used to access storage on Etherent Fabric.

• Journal of the Korea Institute of Information Security & Cryptology
• /
• v.32 no.6
• /
• pp.1127-1137
• /
• 2022

Cryptographic analysis and decryption technology utilizing the parallel operation of GPU has been studied in the direction of shortening the computation time of the password analysis system. These studies focus on optimizing the code to improve the speed of cryptographic analysis operations on a single GPU or simply increasing the number of GPUs to enhance parallel operations. However, using a large number of GPUs without optimization for data transmission causes longer data transmission latency than using a single GPU and increases the overall computation time of the cryptographic analysis system. In this paper, we investigate GPUDirect RDMA and related technologies for high-performance data processing in deep learning or HPC research fields in GPU clustering environments. In addition, we present a method of designing a high-performance cryptanalysis system using the relevant technologies. Furthermore, based on the suggested system topology, we present a method of implementing a cryptanalysis system using password cracking and GPU reduction. Finally, the performance evaluation results are presented according to demonstration of high-performance technology is applied to the implemented cryptanalysis system, and the expected effects of the proposed system design are shown.

• Archives of Plastic Surgery
• /
• v.49 no.5
• /
• pp.663-667
• /
• 2022

Finger dorsum defects are a challenging situation. Many reconstruction methods are used in these defects. Extended reverse dorsal metacarpal artery (RDMA) flap is used in dorsal finger reconstruction. Venous congestion in this flap is most important cause of flap failure. In this case, we presented a case in which we used heparin cream due to development of venous congestion in our patient who underwent an extended RDMA flap. A 24-year-old female patient presented to the emergency department with a defect of dorsal of left-hand fourth finger. Defect was covered with an extended RDMA flap. On postoperative first day, venous congestion was observed, and heparin cream was applied three times a day on flap. The signs of venous congestion were regressed. Tissue healed as a result of superficial epidermolysis and skin grafting. No functional limitation was observed in sixth-month postoperative control. Venous congestion is the most important cause of flap failure of extended RDMA flaps. Generally, subcutaneous heparin administration and leech therapy are used. In our case, heparin was applied as a cream instead of subcutaneously, and flap healing was observed as a result of superficial epidermolysis. Heparin cream application can also be used as a treatment option in flaps with venous congestion.

• Annual Conference of KIPS
• /
• 2017.11a
• /
• pp.679-681
• /
• 2017

Apache Storm은 대용량 데이터 스트림을 처리하기 위한 실시간 분산 병렬 처리 프레임워크이며, 이를 사용해 다수의 프로세스 및 스레드를 동시에 동작시킬 수 있다. 하지만, 이러한 멀티 프로세스 및 스레드 환경을 제공하는 Storm은 많은 네트워크 시스템 호출을 수행하고, 이는 잦은 문맥 전환(context switch), 운영체제로의 버퍼 복사, 운영체제 내의 버퍼 복사 등으로 인해 CPU 과부하 문제를 발생시킬 수 있다. 이러한 문제는 고성능 네트워크 장비인 InfiniBand의 IPoIB(IP over InfiniBand) 통신을 사용할 때, InfiniBand가 지원하는 대역폭(bandwidth) 대비 저용량 데이터의 송수신으로 인해 더 잦은 문맥 전환과 버퍼 복사가 발생하여 CPU 과부하 문제가 더욱 심각해진다. 따라서, 본 논문에서는 InfiniBand의 RDMA(Remote Direct Memory Access)를 Storm에 적용하는 설계안을 제시함으로써 CPU 과부하 문제를 해결한다.

• Annual Conference of KIPS
• /
• 2013.05a
• /
• pp.148-150
• /
• 2013

최근 TCP/IP에서 세션을 통하여 노드들 간의 통신을 연결하는 방식에서 현재는 하나의 채널을 통해 고속의 I/O가 가능하도록 하는 인피니밴드 같은 기술이 많이 연구되고 있다. 인피니밴드는 프로세싱 노드와 입출력 장치 사이의 통신, 프로세스간 통신에 대한 산업 표준이 되고 있고 프로세싱 노드와 입출력 장치를 연결하기 위해 스위치 기반의 상호 연결은 전통적인 버스 입출력을 대체하는 새로운 입출력 방식이다. 또한 인피니밴드에서는 현재 이슈가 되고 있는 RDMA 방식을 이용해 원격지 서버들간에 직접 메모리 접근 방식을 통해 CPU와 OS의 로드를 최소화하고 있다. 본 논문에서는 인피니밴드 네트워크를 이용하는 저장장치 접근 프로토콜인 iSER(iSCSI Extension RDMA Protocol)와 기존 이더넷망에서 사용되는 iSCSI(Internet SCSI) 프로토콜을 이용하여 서버와 저장장치 간의 IOPS 와 초당 데이터 전송량에 대한 성능을 평가한다. 우리는 성능평가를 위해 Intel에서 제공하는 저장장치 I/O 성능평가 도구인 IO meter를 이용했다.

• International journal of advanced smart convergence
• /
• v.9 no.3
• /
• pp.253-259
• /
• 2020

Most of today's large-scale cloud systems and enterprise data centers are distributing resources to improve scalability and resource utilization. NVMe-over-Fabric protocol allows submitting NVMe commands to a remote NVMe SSD through RDMA (Remote Direct Memory Access) network. It is attracting attention recently because it is possible to construct a disaggregation storage system with low latency through the protocol. However, the current I/O stack of NVMe-over-Fabric has an inefficient structure for maintaining compatibility with the traditional I/O stack. Therefore, in this paper, we propose a new mechanism to reduce I/O latency and CPU overhead by modifying I/O path of NVMe-over-Fabric to pass through legacy block layer. According to the performance evaluation results, the proposed mechanism is able to reduce the I/O latency and CPU overhead by up to 22% and 24% compared to the existing NVMe-over-Fabrics protocol, respectively.

• Annual Conference of KIPS
• /
• 2016.10a
• /
• pp.58-61
• /
• 2016

.PCI Express는 고속, 저전력 등의 특성으로 업계 표준으로서 많이 쓰이고 있는 프로세서와 주변 I/O 장치들을 연결하는 버스 기술이다. 또한, PCI Express는 인피니밴드와 이더넷과 더불어 고성능 컴퓨터나 컴퓨터 클러스터를 위한 시스템 인터커넥트 기술로 널리 쓰이고 있다. PGAS(partitioned global address space) 프로그래밍 모델은 컴퓨터 클러스터와 같은 다중 호스트 시스템에서 단측 RDMA(remote direct memory access)를 구현하는데 많이 이용된다. 본 논문에서는 PCI Express 기반 RDMA를 구현하기 위해 PGAS 프로그래밍 모델인 OpenSHMEM의 기존의 특징을 유지하여 PCI Express 기반 OpenSHMEM API를 설계 및 구현하였다. 구현한 OpenSHMEM API는 PCI Express의 NTB(non-transparent bridge) 기술로 2대의 PC를 연결한 시스템에서 매트릭스 곱셈 예제를 통하여 실험하였다.