Jaemyoun Lee

Guaranteeing the End-to-End Latency of an IMA System with an Increasing Workload

New features are often added incrementally to avionics systems to minimize the need for redesign and recertification. However, it then becomes necessary to check that the timing constraints of existing as well as new applications are met. We facilitate these checks by introducing a new data switch that bounds the latency of end-to-end communications across a network. This switch runs a clock-driven switching algorithm that is throughput-optimal with a bounded worst-case delay for all feasible traffic. We propose associated heuristics that determine whether the timing constraints of an integrated modular avionics (IMA) system network that uses this switch are met, even if new features have caused traffic to increase, and then search for alternative network configurations if necessary. Virtual integration is used to make a combined analysis of the worst-case delay in the network and the local buses of individual computing modules. This analysis considers the shared network topology, local hardware architectures, and specified IMA configurations. Our approach can be used by a system architect as an effective method for quickly determining which possible system architectures should be pursued to meet timing constraints, and it allows the cascading effects of changes to be tracked and managed. We demonstrate how these heuristics work through an example in which changes are made to an environmental monitoring facility within an avionics system that uses our switch.

Bounding End-to-End Delay for Real-Time Environmental Monitoring in Avionic Systems

Timing guarantees and predictive early analysis are essential considerations for the design of reliable and verified real-time avionics systems. In this paper, we consider an environmental monitoring avionic system, which allows physical circumstances to be visually monitored continuously in real-time. We analyze timing aspects on the partitions of front-end and back-end nodes, and intermediate node which interconnects those systems. On the end nodes, we use ASIIST to evaluate the worst-case delay for PCI bus analysis. And then, we propose a novel real-time switching algorithm which ensures the delay bound on the intermediate node. Finally, we derive the end-to-end delay on the whole system accurately and show how it can be bounded. A predictive analysis on the worst-case end-to-end delay of a system, before deployment, can result in more reliable and well-verified environmental monitoring systems. We also expect this to reduce the cost of designing and implementing environmental monitoring avionic systems.

Analyzing I/O patterns for the design of energy-efficient image servers

Hard disks are one of the largest sources of power consumption in large-scale storage systems. The disk spin up/down technique has been shown to be an effective solution to this problem. Accordingly, the Open Compute Project (OCP) proposed a storage server for cold data, known as Cold Storage, to reduce power consumption using the spin-down technique in large-scale storage systems. With the aim of making effective use of Cold Storage, we characterize the power consumed by a hard disk in its various modes of operation. We then analyze the workload of an instant messaging service. These our contributions should provide guidelines for the implementation of a stable and energy-efficient distributed file system on a Cold Storage server, and to establish a spin-down policy that is power-proportional and promotes reliability.

File-system-level flash caching for improving application launch time on logical hybrid disks

Application launch time is an important performance metric to user experience in desktop environment. The launch time mostly depends on the performance of secondary storage. There is a cost-performance trade-off in using hard disk drive (HDD) or solid-state drive (SSD). Thus, application launch times can be reduced by utilizing SSDs as caches for slow HDDs. We propose a new SSD caching scheme which migrates data blocks from HDDs to SSDs. Since our scheme operates entirely in the file system level and does not require an extra layer for mapping SSD-cached data, which is essential in most other schemes, our scheme does not incur mapping overheads that cause significant burdens on main memory, CPU, and SSD cache itself. Experimental results demonstrate our scheme yields 56% of performance gain in application launch.

Energy-Efficient Storage Policy for Instant Messenger Services

An instant messenger service is one of the most useful mobile apps for exchanging messages and photos with friends. The infrastructural needs of such services are rapidly growing with user demand for high-quality service, and highend computing with sufficient resources are required to meet these demands. However, the amount of power required to maintain such a massive infrastructure is impractically high. We present a novel policy to exploit energy-efficient storage servers for instant messenger services. Because instant messages often include personal information, the storage servers must be protected from unauthorized access. This requirement can be expanded to develop a creative policy for reducing power consumption. We validate our proposal by conducting four experiments, the results of which are expected to play an important role in designing energy-efficient servers for instant messenger services.

A test-bed for the assessment of power management strategies in tiered storage systems

Cloud computing systems require a huge number of storage servers, with growing implications for power bills, carbon emissions and the logistics of data centers. These considerations have motivated researchers to improve the energy efficiency of storage servers. Most servers use a lot of power irrespective of the amount of computing that they are doing, and one important goal is to redesign servers to be power-proportional, meaning that the power consumption is in step with the workload. Research on large-scale storage systems is hampered by their cost (Wang et al. 2014). It is therefore desirable to develop a scalable test-bed for evaluating the power consumption of large-scale storage systems. We are building on two open-source projects to construct a test-bed which will contribute to the assessment of power consumption in tiered storage systems.

SVC-aware selective repetition for robust streaming of scalable video

Abstract3G video broadcast services employ forward error correction, together with Reed-Solomon erasure coding with appropriate interleaving, in the MAC layer to deal with bursts of errors. However, this method of error recovery is less effective at the margins of coverage where channel conditions are bad, so that mobiles experience higher error rates and longer error bursts: this effectively limits the service area for video applications. To overcome this problem, we augment scalable video coding (SVC) with a selective repetition scheme, which retransmits those packets within the layered structure of SVC that are most important for video. Simulations using an the example MPEG-4 fine granularity scalability video demonstrate that our scheme safeguards the important packets and improves the service quality. Furthermore, energy measurements show that it also extends battery life.

Strategy to Reduce I/O Access Time of Applications in Virtual Machines

An input/output (I/O) device can incorporate mechanical devices that require physical movements, such as a hard drive seeking a track for a data read or write. Hence I/O operations on a computer can be extremely slow compared to data processing. When a program conducts many I/O operations, the processor often spends most idling as it waits for the I/O operations to complete. The same holds true for a virtual machine (VM), and to improve the overall performance of a VM, this problem needs to be addressed. In this paper, we propose a pinning technique, which migrates frequently accessed disk blocks from hard disks to flash memories, to reduce the I/O waiting time. Experimental results show that the start-up times dramatically decrease by up to 38% on average when using this method.

Design of an Optical Packet Switch for Real-Time Applications

Network switches are typically designed for best- effort Internet traffic. Most of existing studies have been focused on improving throughput and delay performance in an average sense rather than providing guaranteed delay bound that is critical for real-time applications. It has not been fully investigated how to design an efficient packet switching algorithm for real-time applications. In this paper, we propose a design framework for a real-time optical switch that is intended for use as an optical switch fabric. Our contributions are two folds: First, by introducing a clearance-time optimal switching together with clock-based scheduling, our switching design guarantees any feasible real-time traffic to be switched in two- clock periods. Second, we investigate key implementation issues of an optical packet switch such as packet size and buffering for real-time applications, and take account of these issues in design and performance evaluation of a switching algorithm. Our numerical study shows that the proposed switching algorithm provides a larger schedulability region with significantly reduced delay compared to the well-known iSLIP scheme.

Advanced Primary–Backup Platform with Container-Based Automatic Deployment for Fault-Tolerant Systems

Within mission-critical systems, the primary–backup scheme is a desirable approach for improving reliability and fault tolerance. It can be used to ensure a high mission success rate despite unexpected errors. However, it must cope with the need to maintain consistency between a primary and a backup whenever the primary encounters unexpected errors. We overcome this issue by introducing a platform that uses container-based light virtualization and an automatic build system to isolate an application so that it may then be deployed on different devices without manual intervention. We believe an advanced deployment procedure can retain the consistency of the primary–backup systems with low implementation complexity. Integrated with a cloud application, it can also manage mission-critical systems effectively, communicate with the redundant systems, and detect unexpected errors by using sophisticated fault-detection technologies. We demonstrate that the platform can improve the reliability of mission-critical systems through realistic experiment using a model electronic vehicle and can reduce hardware dependencies.