• IEMEK Journal of Embedded Systems and Applications
• /
• v.14 no.2
• /
• pp.97-102
• /
• 2019

Multi-queue I/O block layer has been recently employed in Linux kernel to support fast storage devices such as NVMe SSDs, but it lacks differentiated I/O services yet. In this paper, we propose an I/O scheduling scheme that can improve the user responsiveness of foreground processes, which are closely related to user satisfaction. To this end, we redesign the existing multi-queue block layer to classify the I/O requests from foreground processes and schedule them by exploiting the feature of NVMe interface. Experimental results show that latency and launch time of the foreground processes have been significantly improved compared to original Linux kernel.

• Journal of KIISE
• /
• v.44 no.5
• /
• pp.469-475
• /
• 2017

The emerging NVMe-based multi-queue SSDs provides a high bandwidth by parallel I/O, i.e., each core performs I/O through its dedicated queue in parallel with other cores. To provide a bandwidth share for each application with I/O, a fair-share scheduler that provides a bandwidth share to each core is required. In this study, we proposed a multi-core scalable fair-queuing algorithm for multi-queue SSDs. The algorithm adopts randomization to minimize the inter-core synchronization overheads and provides a weight-proportional bandwidth share to each core. The results of our experiments indicated that the proposed algorithm gives accurate bandwidth partitioning and outperforms the existing FlashFQ scheduler, regardless of the number of cores for a Linux kernel with block-mq.

• Proceedings of the Korean Information Science Society Conference
• /
• 2012.06c
• /
• pp.151-153
• /
• 2012

최근 고속 네트워크와 저장 기술의 발전으로 인하여 대용량 데이터 분산 처리 시스템에 관한 연구가 활발히 진행되고 있다. 또한 서버의 통합을 통해 시스템 자원의 효율적인 활용을 제공할 수 있는 시스템 가상화가 많은 주목을 받고 있다. 그러나 가상 머신 환경에서 대용량 데이터 분산처리 시스템을 구성할 경우 많은 문제가 발생하게 된다. 본 논문에서는 가상 머신 환경에서 Hadoop 클러스터를 활용할 때 가상 데이터 노드의 개수에 따른 I/O 대역폭 최적화에 대한 실험을 하고 평가를 한다. 본 논문에서 수행한 실험 결과는 가상 머신 환경에서 I/O 대역폭 밸런싱(balancing)을 지원하는 Hadoop Scheduler의 개발 연구에 사용될 것이다.

• Journal of KIISE:Computing Practices and Letters
• /
• v.15 no.5
• /
• pp.400-404
• /
• 2009

Native Command Queueing is a technology to maximize throughput of disk system by reordering requests in its command queue. Recent S-ATA 2 standard specifies a protocol for the purpose of dealing with NCQ feature, making most vendors implementing it in their disk devices. However, the new feature, NCQ, may lead to information gap between OS and disk drive. 05 considers that disk will service I/O requests in the order as OS has dispatched. Unfortunately, it isn't true any more since NCQ would simply ignore the policy of OS and reorder the sequence OS has expected. Let us define the term as expectation-discord. Due to the discord, serious performance anomalies or starvation of requests may occur in I/O subsystem. In this paper, we confirm that the expectation-discord actually brings about starvation of requests. We propose a solution to settle it. Our solution is very simple, effective, cheap(not requiring any hardware modification) and portable across various OS. Experimental results show that our solution can balance throughput and response time very well.

• Journal of the Institute of Electronics and Information Engineers
• /
• v.51 no.9
• /
• pp.103-108
• /
• 2014

In the virtualized server systems, a scheduler in a hypervisor is responsible to assign physical resources for virtual machines. However, the traditional scheduler is hard to provide optimized resource allocation considering the amount of I/O requests. Especially, the drawback hinders performance of software load balancer which runs on virtual machines to distribute I/O requests from the clients. In this paper, we propose a new architecture to improve the performance of software load balancer using multiprocess. Our architecture aims to improve hardware resource utilization and overall performance of the server systems which utilize virtualization technology. Experimental results show the effectiveness of the proposed architecture for the various cases.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2019.05a
• /
• pp.68-71
• /
• 2019

하이퍼바이저 Xen 은 처음 개발되었을 때부터 현재까지 크레딧-스케줄러를 이용하고 있다. 크레딧-스케줄러는 CPU 이용률을 높이기 위해 디자인되어 있으며 효율적인 I/O 처리를 위해 부스트라는 개념을 도입하였지만, 부스트로 인하여 공정성이 깨질 수 있는 문제를 가지고 있다. 이 논문에서는 I/O 중심 처리 환경에서 크레딧-스케줄러의 불공정성을 증명하였으며, 실험을 통해 네트워크 성능에서 최대 약 60%의 차이를 보임을 확인할 수 있었다.

• Journal of KIISE
• /
• v.42 no.2
• /
• pp.169-176
• /
• 2015

The atomic write technique is a good solution to solve the problem of the double write buffer. The atomic write technique needs modified I/O subsystems (i.e., file system and I/O schedulers) and a special SSD that guarantees the atomicity of the write request. In this paper, we propose the writing unit aligned block allocation technique (for EXT4 file system) and the merge prevention of requests technique for the CFQ scheduler. We also propose an atomic write-supporting SSD which stores the atomicity information in the spare area of the flash memory page. We evaluate the performance of the proposed atomic write scheme in MariaDB using the tpcc-mysql and SysBench benchmarks. The experimental results show that the proposed atomic write technique shows a performance improvement of 1.4~1.5 times compared to the double write buffer technique.

• International journal of advanced smart convergence
• /
• v.8 no.1
• /
• pp.9-17
• /
• 2019

In this article, we present a performance enhancement scheme for mobile applications by adopting persistent memory. The proposed scheme supports the deadline guarantee of real-time applications like a video player, and also provides reasonable performances for non-real-time applications. To do so, we analyze the program execution path of mobile software platforms and find two sources of unpredictable time delays that make the deadline-guarantee of real-time applications difficult. The first is the irregular activation of garbage collection in flash storage and the second is the blocking and time-slice based scheduling used in mobile platforms. We resolve these two issues by adopting high performance persistent memory as the storage of real-time applications. By maintaining real-time applications and their data in persistent memory, I/O latency can become predictable because persistent memory does not need garbage collection. Also, we present a new scheduler that exclusively allocates a processor core to a real-time application. Although processor cycles can be wasted while a real-time application performs I/O, we depict that the processor utilization is not degraded significantly due to the acceleration of I/O by adopting persistent memory. Simulation experiments show that the proposed scheme improves the deadline misses of real-time applications by 90% in comparison with the legacy I/O scheme used in mobile systems.

• Journal of KIISE
• /
• v.42 no.11
• /
• pp.1322-1331
• /
• 2015

According to the hypervisor scheduler, the vCPU (virtual CPU) operates under two states: the running state and the stop state. When the vCPU is in the stop state, incoming events are delayed until that vCPU's state changes to the running state. The latency in handling such events that are sent to the vCPU is regarded as the I/O latency. Since a SMP (symmetric multiprocessing) VM (virtual machine) incorporates multiple vCPUs, the event latency on a SMP VM can vary according to specific vCPU that receives the event. In this paper, we propose a new scheme named event routing that sends events according to the operation state of each vCPU to reduce the event latency on an SMP VM. We implemented the proposed event routing scheme in Xen ARM hypervisor and confirmed the reduction of I/O latency from measuring the network RTT (round trip time) and the TCP bandwidth under a variety of testing conditions. The network RTT decreases by up to 94% and the TCP bandwidth increases up to 35% when compare to native Xen ARM.

• Journal of Korea Multimedia Society
• /
• v.15 no.12
• /
• pp.1499-1505
• /
• 2012

Recently, the distributed processing system for big data has been actively investigated owing to the development of high speed network and storage technologies. In addition, virtual system that can provide efficient use of system resources through the consolidation of servers has been increasingly recognized. But, when we configure distributed processing system for big data in virtual machine environments, many problems occur. In this paper, we did an experiment on the optimization of I/O bandwidth according to the creation and placement of VMs and tasks with composing Hadoop cluster in virtual environments and evaluated the results of an experiment. These results conducted by this paper will be used in the study on the development of Hadoop Scheduler supporting I/O bandwidth balancing in virtual environments.