K. G. Shin. HARTS: A Distributed Real-Time Architecture". IEEE Computer, 24:25 -- 35, May 1991.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....is inherently target specific, and hard to adapt to other systems. The system is designed for single processor systems and has no support for distributed real time systems. The software based approach HMON [22] is designed for the HARTS distributed (real time) system multiprocessor architecture [100]. A general purpose processor is dedicated to monitoring on each multiprocessor. The monitor can observe the target processors via shared memory. The target systems software is instrumented with monitoring routines, by means of modifying system service calls, interrupt service routines, and making ....

Shin K. G. HARTS: A distributed real-time architecture. IEEE Computer, 24(5), pp. 25-35, May, 1991.

....to the case of mixed codes with codewords where (For Theorems 6 and 7, is even and for all ) C. Other Topologies The next topology that we consider is a balanced ary tree. A number of hierarchical computing systems such as dictionaries and search machines can be modeled as a tree [2] [24]. Many parallel algorithms can be mapped on to ary tree, and the architecture of a general purpose multiprocessor can often be modeled by a tree structure [19] Another application of a tree structure is the data network of the Thinking Machine CM 5 [11] 16] We can uniquely identify vertices ....

K. G. Shin, "HARTS: A distributed real-time architecture," IEEE Computer, pp. 25--35, May 1991.

.... management have focused on realtime systems that are static (in both the above senses) and perform device level, sampled data monitoring and regulatory control that is usually centralized [Bak91, LRT92, LSS87, RCF97, RTL93, SB96, SKG91, SLS88, SSL89, TLS96, XP90] but occasionally distributed [CSR86, HS92, Kao95, KDK89, RSZ89, Shi91, SR91, SRC85, Ver95, WSM95]. These techniques cannot be practically employed or adapted for systems that are dynamic [Jen99, Koob96, Sta96, SK97] Dynamic real time computer systems and their applications have workload characteristics that are inherently posteriori. Thus, such systems require adaptive resource management ....

K.G. Shin, "Harts: A Distributed Real-Time Architecture," IEEE Computer, 24(5):25-- 35, May 1991.

.... real time systems that has resorted to hardware assisted solutions includes the VLSI clock synchronization unit that is part of MARS, a fault tolerant real time system implementation [3] the fault tolerant message routing unit that is part of HARTS (Hexagonal architecture for Real Time Systems) [15], and network interface chips for the FDDI protocol [14] More familiar examples of hardware co processors include floating point, graphics, DMA, cache controllers and many others. All involve tasks that could be performed in software, but which are critical enough to system performance to warrant ....

K. Shin. HARTS: A Distributed Real-Time Architecture. IEEE Computer 24(5), May, 1991.

....results can be found in [7] 3. 1 Simulation model and parameters The simulator is based on an experimental distributed real time system called Hexagonal Architecture for Real Time Systems (HARTS) being developed at the Real Time Computing Laboratory, The University of Michigan, Ann Arbor [1,8]. HARTS is a distributed system with a point to point interconnection called the C wrapped hexagonal mesh topology. This topology is also being used in Mayfly systems at Hewlett Packard Research Laboratories [2] It is regular, homogeneous topology in which each node has six neighbors. More ....

....a point to point interconnection called the C wrapped hexagonal mesh topology. This topology is also being used in Mayfly systems at Hewlett Packard Research Laboratories [2] It is regular, homogeneous topology in which each node has six neighbors. More details on this topology can be found in [8]. The messages generated in the simulator are assumed to have the following characteristics. 1. Message generation at each node is assumed to be a Poisson process. The rate of the Poisson process is chosen to correspond to a particular message traffic intensity in the network. The intensity is ....

K. G. Shin, "HARTS: A distributed real-time architecture," IEEE Computer, vol. 24, no. 5, pp. 25--35, May 1991. 18

....[Sta88] Architectural support for RTSs in general has focussed on embedded approaches [Sta88] which do not have the flexibility required of an RTDB application. Other designs have been geared 2 toward general purpose platforms that use traditional methods. Examples include the design of HARTS [Shi91] based on interconnecting off the shelf components) the MAFT [McE88] design (where the concern is failure resilience and scheduling) MAX [R 87] which is a loosely coupled multicomputer for supporting real time dataflow programs) and SMART [Kir89] that is concerned with real time ....

K. G. Shin. HARTS: A distributed real-time architecture. IEEE Computer, pages 25--35, May 1991.

....target specific, and hard to adapt to other 25 systems. The system is designed for single processor systems and has no support for distributed real time systems. The software based approach HMON [12] is designed for the HARTS distributed (real time) system multiprocessor architecture [59]. A generalpurpose processor is dedicated to monitoring on each multiprocessor. The monitor can observe the target processors via shared memory. The target systems software is instrumented with monitoring routines, by means of modifying system service calls, interrupt service routines, and making ....

Shin K. G. HARTS: A distributed real-time architecture. IEEE Computer, 24(5), pp. 25-35, May, 1991.

....29, 36] distributed HRT systems can be classified as responsive systems, i.e. distributed, fault tolerant, real time systems. In this tutorial paper, we shall focus on issues of design andimplementation of distributed RT systems, and describe five operational examples of those systems, namely [52, 17, 33, 56, 47]. In particular, we shall discuss the key paradigms for the design of timely and available RT system services, and examine techniques for process scheduling, time management, and interprocess communications over local and wide area networks. This paper is structured as follows. In the next ....

....fails, SPRING allocates a different node for that request. Finally, it is worth mentioning that fault tolerance issues have received little attention in the design of the SPRING Kernel. 4. 2 HARTS The distributed RT architecture of the HARTS project (Hexagonal Architecture for Real Time Systems [56]) is based on shared memory multiprocessor nodes interconnected by a wrapped hexagonal network (Fig.12 shows an example of the HARTS hexagonal network) This architecture aims to providing RT applications with high performance, reliability, and predictability. The HARTS network can be conveniently ....

Shin K. G.: Harts: A Distributed Real-Time Architecture. IEEE Computer: May 1991, 25--35.

....Sense Multiple Access with Collision Detection) In [69] Strosnider et al. proposed the use of rate monotonic scheduling (RMS) on an IEEE 802.5 token ring. The use of timed token protocol for real time traffic has been studied in [3, 18] Point to point networks have also received attention [8, 19, 47, 53, 54, 62]. The key advantage of point to point networks is the existence of multiple paths between nodes. However, messages may have to travel multiple hops to reach their destinations. A study of scheduling mixed traffic of real time and non real time traffic is presented in [77] Three schemes are ....

K. G. Shin, "HARTS: A distributed real-time architecture," IEEE Computer, vol. 24, no. 5, pp. 25--35, May 1991.

....environment. The basic idea here can also be used to enhance the scheme presented in [13,14] The effectiveness of the proposed approach is evaluated using a simulator of a C wrapped hexagonal mesh topology [1] This topology is currently being used in two experimental distributed systems [2,15]. The results indicate that reductions in the expected cost incurred by the system due to missed deadlines are substantial. Except for very high loads, the reductions are over 66 . This paper is organized as follows. Section 2 describes the target distributed system. A detailed description of the ....

....scheme proposed in the previous section. 6 Evaluation of the Proposed Approach In this section, the performance of the proposed approach is evaluated through simulation of a C wrapped hexagonal mesh topology. This topology is a regular and homogeneous graph in which each node has six neighbors [15]. The graph can be visualized as a simple hexagonal mesh with wrap links added to the nodes on the periphery. A simple hexagonal mesh is composed of a set of concentric hexagons with a central node, where each hexagon has one more node on each edge than the one immediately inside of it. Fig. 3 ....

[Article contains additional citation context not shown here]

K. G. Shin, "HARTS: A distributed real-time architecture," IEEE Computer, vol. 24, no. 5, pp. 25--35, May 1991.

....arbitrary waits for resources or events, or treating the operation of a task as a random process will cause great uncertainty in accomplishing the application level end to end requirements. As an example, the Mars project [3] the Spring project [17] and a project at the University of Michigan [11] are all attempting to solve this problem. The Mars project uses an a priori analysis and then statically schedules and reserves resources so that distributed execution can be guaranteed to make its deadline. The Spring approach support dynamic requests for real time virtual circuits (guaranteed ....

....systems. 5 Architecture and Fault Tolerance Hard real time systems are usually quite special purpose. Architectures to support such applications tend to be special purpose too. The current trend is one in which more off the shelf components are being used to produce more generic architectures [11, 18], rather than the previously developed highly special purpose architecture. As an example, consider the SpringNet architecture. SpringNet [18] is a physically distributed system composed of a network of three multiprocessors each running the Spring Kernel. Each multiprocessor currently contains ....

K. Shin, "HARTS: A Distributed Real-Time Architecture," IEEE Computer, Vol. 24, No. 5, May 1991.

....19] and IBM s [7, 8] and Tandem s [11] failure analysis are some studies done in this area. These studies attempt to propose a small set of failures for fault injection testing. A number of tools for performing fault injection testing exist [13] MESSALINE [2, 3] FERRARI [14] FIAT [4, 10] SFI [20, 21] and FINE [15] are a few tools which were used for hardware and software fault injection testing. These tools suffered from a lack of (i) a notion of test adequacy and (ii) a quantitative measure for the process of validation of the system properties. Faults were injected at low level and the ....

K. Shin. "HARTS: A Distributed Real-time Architecture". IEEE Computer, 24:25--35, May 1991.

....A processor and its memory are connected by a conventional high bandwidth connection. The I O processors connect to input and output units of the system. The interconnection topology is a set of buses, each one with one bus controller. A more advanced solution was presented in the HARTS project [Shi91]. HARTS uses a hexagonal mesh of clusters, each one consisting of application processors, a system controller, a shared memory, an Ethernet processor, and a Network Processor (designed as interface to the interconnection network) A hexagonal mesh was chosen as it meets the requirements of fixed ....

....mode. I O devices are clustered, and a controller manages access to the devices of each cluster. Each I O controller is placed as the center of one triangle derived from the hexagonal mesh, giving three possible ways of access to each controller. Figure 3. Placement of I O controllers in HARTS [Shi91]. Simple store and forward switching schemes are not suitable for real time routing because real time applications normally require short response times. Hence, HARTS supports fast switching methods, as virtual cut through and wormhole routing. The algorithms presented in [Shi91] does not attack ....

[Article contains additional citation context not shown here]

SHIN, K. "HARTS: a Distributed Real-Time Architecture". IEEE Computer, May 1991. pp. 25-35.

....more difficult. What is required is a strong underlying theory, proper system level support (such as real time languages [7] on line scheduling algorithms, synchronization mechanisms, agreement protocols, recovery algorithms, mechanisms to support adaptive fault tolerance, architecture support [15,20]) and useful design and analysis tools all developed by experts to insulate users from the complexity and to avoid mistakes. Each of these areas must focus on the integration of scheduling and fault tolerance and not treat them independently. While new approaches that arise from considering ....

K. Shin. HARTS: A distributed real-time architecture. IEEE Computer, Vol. 24, No. 5, May 1991.

....are guaranteed to be tolerated. To manage the available resources (commonly the processors) in order to achieve timing and fault tolerance guarantees, specialized scheduling algorithms are required. Many real time systems focus on the use of hardware redundancy to provide fault tolerance, e.g. [32, 35, 78]) The main advantage of using hardware redundancy is that permanent hardware faults can be tolerated. However, hardware redundancy also has some drawbacks. First of all, hardware redundancy mainly targets permanent faults in hardware. Since it is well known that transient faults are significantly ....

....alternatives or replicas. The user can define the level of fault tolerance, where redundancy can be established either temporally (re execute task) or physically (alternatives executed in parallel) Maruti 29 also supports node and link failures, using roll forward recovery techniques. The HARTS [78] architecture consists of a hexagonal mesh of nodes. Each node is connected to six of its neighbors. Due to the redundancy in the interconnection network, the system can tolerate multiple node and link failures. Another mechanism that provides fault tolerance is the replication of processes. The ....

K. G. Shin, HARTS: A Distributed Real-Time Architecture, IEEE Computer, 24(5): 25--35, May 1991.

....mentioned) is based on a heuristic greedy algorithm. At the processor level, scheduling is carried out by using an Earliest Deadline First dispatcher. Other significant research issues have been covered as part of the Arts (Tokuda and Mercer, 1989) Harts (Kandlur, Kiskis and Shin, 1989) (Shin, 1991), Maruti (Levi and Agrawala, 1990) and RT Mach (Tokuda and Nakajima, 1990) Projects. 5. CONCLUSION Our approach tends to address a wide variety of task timing characteristics and makes use of DF programming to increase the flexibility in deadline assignments. Thus, trying to compare our project ....

Shin, K.G. (1991). HARTS: A Distributed RealTime Architecture. IEEE Computer 24(5), 25 -- 35.

....the ability to interact with each of the processor boards individually, and to collect streams of debugging output from each processor while minimizing the effect on system behavior. The HARTS project features sophisticated hardware, but an essentially conventional approach to the operating system [34, 78]. This project primarily focuses on the communication aspects of distributed real time systems, and by extension on fault tolerance. The hardware design uses a hexagonal mesh layout for the processor interconnections, and a special network processor takes care of much of the networking overhead. ....

Shin, K. HARTS: A Distributed Real-Time Architecture. In Proceedings of the 3rd ONR Workshop on the Foundations of Real-Time Computing, pages 225--254. ONR, October 1990.

....target locations through the help of the compiler and the loader. It was used to get a profile of error detection coverage and fault latency during the fault injection testing process. 3.2. 4 SFI SFI, Software Fault Injector, is targeted specifically to a distributed real time system called HARTS [19, 20]. Various faults like memory faults, CPU errors and message related errors can be injected with three levels of persistence transient, intermittent and permanent. The fault injector was a high priority process and its implementation required modification to the underlying operating system. ....

K. Shin. "HARTS: A Distributed Real-time Architecture". IEEE Computer, 24:25--35, May 1991.

....injection testing. 3.2. Survey of Fault Injection Techniques Various fault injection techniques have been used on systems. These techniques range from hardware fault injection at pin level in chips to software fault injection in computer programs. MESSALINE [2, 3] FERRARI [10] FIAT [4, 8] SFI [17, 18], FINE [11] are some fault injection tools which have been reported in the literature. All these tools relied on low level fault injection hardware and memory images. Fault injection at the hardware level requires a large number of fault injections. Often, special purpose hardware support is ....

K. Shin. "HARTS: A Distributed Real-time Architecture". IEEE Computer, 24:25--35, May 1991.

....sets, multi port communication. 1 Introduction The availability of inexpensive, high performance microprocessors has made it attractive to link together many powerful and autonomous computers to build a distributed computing system for better availability and cost performance [7] 17] 19] [21]. In such a system, instead of using a shared memory and a global clock, all the synchronization and communication between the processing nodes is done via message passing [1] 22] Since data are distributed, not shared, special schemes are generally required to perform various distributed ....

K. G. Shin. HARTS: A Distributed Real-Time Architecture. IEEE Computer, pages 25--35, May 1991.

....across many areas. In this section we discuss real time kernels, real time scheduling, and realtime architectures and fault tolerance. Due to space limitations many other areas of real time systems are not discussed such as programming languages, communication protocols [1] distributed systems [3, 8, 9, 11, 21] and design methodologies. 3.1 Real Time Kernels One focal point for next generation real time systems is the operating system. The operating system must provide basic support for predictably satisfying real time constraints, for fault tolerance and distribution, and for integrating ....

....arbitrary waits for resources or events, or treating the operation of a task as a random process will cause great uncertainty in accomplishing the application level end to end requirements. As an example, the Mars project [3] the Spring project [17] and a project at the University of Michigan [11] are all attempting to solve this problem. The Mars project uses an a priori analysis and then statically schedules and reserves resources so that distributed execution can be guaranteed to make its deadline. The Spring approach support dynamic requests for real time virtual circuits (guaranteed ....

[Article contains additional citation context not shown here]

K. Shin, "HARTS: A Distributed Real-Time Architecture," IEEE Computer, Vol. 24, No. 5, May 1991.

....[1, 21, 48] 21 Hardware Issues, Hardware Software Interfaces, Systems Integration, and Test [4, 29: Chapters 2 13] 22 Systems Integration Continued 23 Distributed Real Time Systems, Reliability and Fault Tolerance [28; 29: Chapters 11 12] 24 Reliability and Fault Tolerance Continued Add. Ref. [8, 9, 21, 22, 24, 25, 30, 34, 35, 41, 44, 45, 55, 56, 57, 66] 25 Case Study: Real Time Communications 26 Case Study: Radar Applications 26 Project Presentations Lecture Topics Readings 7 . List and define several software life cycle models. Suggest disadvantages and advantages of using a particular life cycle model in the development of real time ....

Shin, K. (May 1991). HARTS: A Distributed Real-Time Architecture. Computer, Vol. 24, No. 5, pp. 25-35.

....workloads, the performance of routing algorithms varies significantly. In particular, routing performance is fairly sensitive to the selection function, adaptivity, and the traffic pattern. Applying these results to improve routing performance can easily be done for systems such as HARTS [27], the nCube 3 [25] and Hnet [24] which support multiple routing policies. For example, in HARTS, the Programmable Routing Controller (PRC) 19, 26] can be reprogrammed with different routing policies as the application communication workloads change. By having multiple algorithms programmed into ....

K. G. Shin, "HARTS: A distributed real-time architecture," IEEE Computer, vol. 24, pp. 25--35, May 1991. Appendix A Simulation Results

....be permanent, transient or intermittent. A fault injection plan can be formulated probabilistically, or based on the past event history. The modular organization of tools is particularly designed for distributed architectures. DOCTOR is implemented on a distributed real time system called HARTS [1], and its capability has been tested through numerous experiments. 1 Introduction In real time systems the correctness of a computation depends not only on the logical correctness of the result but also on the time at which the result is produced [2] There are a wide range of real time ....

K. Shin, "HARTS: A distributed real-time architecture, " IEEE Computer, vol. 24, no. 5, pp. 25--35, May 1991.

....in a hexagonal mesh. In this paper, our aim is to integrate and implement these solutions to provide communication services for applications. Our approach is to develop a communication subsystem which supports real time communication and provides a global time base for the system. We use HARTS [37], which is a distributed real time system being developed at the Real time Computing Laboratory, The University of Michigan, as the vehicle for our development. The system is comprised of several multiprocessor nodes connected by a point to point interconnection network with a hexagonal mesh ....

K. G. Shin, "HARTS: A distributed real-time architecture," IEEE Computer, vol. 24, no. 5, pp. 25--35, May 1991.

....fault injection tool supports three types of faults: processor faults, memory faults, and communication faults. It also allows for injecting permanent, transient or intermittent faults. The proposed design methodology for DOCTOR has been implemented on a distributed real time system called HARTS [1], and its capability is demonstrated through numerous experiments. Dependability measures, such as detection coverage latency and the associated performance overhead, are evaluated through extensive experiments. Communication fault injection is used to evaluate a probabilistic distributed ....

....an automated and user transparent multi run facility allows experiments to be repeated any number of times with no additional user intervention. The proposed SFI environment, called an integrateD sO ftware fault injeC T iO n enviR onmenti or DOCTOR for short, is implemented on HARTS [1], and several error detection and recovery mechanisms are evaluated using these tools. Section 2 presents the fault model of our SFI, as well as the architecture of DOCTOR. Section 3 gives a brief description of the initial target system, HARTS, and discusses the underlying implementation issues. ....

[Article contains additional citation context not shown here]

K. G. Shin, "HARTS: A distributed real-time architecture," IEEE Computer, vol. 24, no. 5, pp. 25--35, May 1991.

....SW design was later studied extensively by Ferrari [9, 10, 11] However, his work also concentrated on general purpose uniprocessor computers. At the Real Time Computing Laboratory of The University of Michigan, we are designing and building the Hexagonal Architecture for Real Time Systems (HARTS) [31]. HARTS is a distributed real time system consisting of a number of multiprocessor nodes connected by a custom hexagonal mesh interconnection network. HARTS is an experimental system which is to serve as a testbed for developing and evaluating real time communication, fault tolerance, and ....

K. G. Shin, "HARTS: A distributed real-time architecture," IEEE Computer, vol. 24, no. 5, pp. 25--35, May 1991.

....the output channel of node (3,3) of V IN 1 in the negative direction of dimension x. Due to the limited space, we only present an adaptive, deadlock free routing algorithm for n dimensional meshes by using the 3P routing. For other topologies, such as k ary n cubes, C wrapped hexagonal meshes [7] and so on, a similar method can be used, though the virtual channels required in V IN 1 may be different. An n dimensional mesh consists of k 0 Theta k 1 Theta Delta Delta Delta Theta kn Gamma2 Theta kn Gamma1 nodes, where k i 2 is the number of nodes along dimension i. Each node X is ....

....algorithms for several topologies. The proposed 3P routing requires only one additional virtual channel as compared to the deterministic routing. Furthermore, it is minimal and fully adaptive. It can also be used to construct an adaptive deadlock free routing algorithm for C wrapped H meshes [7]. The number of virtual channels required by the 3P routing on H meshes is the same as that of k ary n cubes. The 3P routing turns out to have a flavor somewhat similar to the one proposed in [3] although both have been developed independently. However, the method proposed in [3] dealt only ....

K. G. Shin, "HARTS: A distributed real--time architecture," IEEE Computer, vol. 24, no. 5, pp. 25--34, May 1991.

....for supporting different higherlevel performance requirements. The section compares the schemes running on a cycle level simulation model of SPIDER (Scalable Point to point Interface DrivER) a hardware adapter for real time multi hop networks [7, 8] Designed as the front end interface for HARTS [17], SPIDER supports a variety of routing and switching schemes through flexible, low level control over the network links. Section 3 presents mechanisms for utilizing such low level hardware support to regulate the mixing of guaranteed and best effort traffic. The paper concludes with Section 4. 2 ....

K. G. Shin, "HARTS: A distributed real-time architecture, " IEEE Computer, vol. 24, no. 5, pp. 25--35, May 1991.

....other related approaches. We conclude in Section 8 with an analysis of the results and suggest directions for future work. 2 The Experimentation Platform In this section we describe the hardware and software architecture of our experimentation platform, which is being developed as a part of HARTS [5]. The primary goal of HARTS is to investigate architectural and operating system issues in distributed real time computing. 2.1 Hardware Each HARTS node (also referred to as end host) is a VME bus based multiprocessor with 2 4 processors, as shown in Figure 1. This multiprocessor configuration ....

K. G. Shin, "HARTS: A distributed real-time architecture," IEEE Computer, vol. 24, no. 5, pp. 25--35, May 1991.

....component failures as long as the failures are isolated. This kind of fault tolerance cannot be achieved with other commonly known topologies like rings, rectangular meshes, and hypercubes. The proposed approach is to be implemented in an experimental distributed real time system called HARTS [2]. Submitted to IEEE Transactions on Networking. 201 Broadway Cambridge Massachusetts Publication History: 1. First printing, TR 93 10, June 1993 Copyright c fl Mitsubishi Electric Research Laboratories, 1991 201 Broadway; Cambridge Massachusetts 02139 This work may not be copied or reproduced ....

....the schemes of establishing isolated failure immune real time channels in HARTS. The paper concludes with Section 4. 2 HARTS and Real time Channels HARTS is an experimental distributed real time system currently being built in the Real Time Computing Laboratory of the University of Michigan [2]. As shown in Fig. 3, the interconnection 1 2 3 4 5 6 source destination Figure 2: An IFI channel and one pattern of tolerable link node failures. 17 14 15 16 6 7 8 9 18 0 1 2 10 11 12 13 3 4 5 13 2 3 3 4 4 5 17 17 10 13 5 5 16 9 10 3 14 14 6 6 17 16 16 15 15 14 2 2 9 17 X Y Z Figure 3: A ....

[Article contains additional citation context not shown here]

K. G. Shin, "HARTS: A distributed real-time architecture," IEEE Computer, vol. 24, no. 5, pp. 25--35, May 1991.

....the establishment of the route with the reservation of resources and bandwidth, and (iii) run time support for routing and scheduling a mix of real time and best effort messages through the network. We now show how these can be achieved using the x kernel [5] on the Network Processor (NP) of HARTS [6]. The x kernel is designed for distributed system support, and provides facilities for implementing protocols like a uniform protocol interface, libraries to efficiently manipulate messages, and tools to configure and test different protocol stacks. We implement real time channels as a set of ....

K. G. Shin, "HARTS: A distributed real--time architecture," IEEE Computer, vol. 24, no. 5, pp. 25--36, May 1991.

....workloads, the performance of routing algorithms varies significantly. In particular, routing performance is fairly sensitive to the selection function, adaptivity, and the traffic pattern. Applying these results to improve routing performance can easily be done for systems such as HARTS [19], the nCube 3 [16] and Hnet [15] which support multiple routing policies. For systems which only support a single routing policy, these results can be used to influence how the operating system maps the tasks onto multicomputer nodes, in effect, tailoring the communication workload to the routing ....

K. G. Shin, "HARTS: A distributed real-time architecture, " IEEE Computer, vol. 24, pp. 25--35, May 1991.

....model as long as the models adhere to the same protocol. For this study, two routing algorithms based on Duato s methodology were written, one of which was used in these simulations. pp mess sim currently includes a cycle level model of SPIDER [14, 15] a communication adapter designed for HARTS [25]. Our simulations use this model as the basis for evaluating these algorithms. Some of the more important aspects of SPIDER include a demand slotted bus interconnection between input and output channels and a fair arbitration scheme between virtual channels competing for a given physical link. The ....

K. G. Shin, "HARTS: A distributed real-time architecture," IEEE Computer, vol. 24, no. 5, pp. 25--35, May 1991.

....we have been designing, implementing, and evaluating a 19 node hexagonal mesh, called HARTS (Hexagonal Architecture for Real Time Systems) in the Real Time Computing Laboratory (RTCL) the University of Michigan. The architectural aspects of HARTS have been treated in a companion paper [25]. The main intent of this chapter is to cover the software environment of HARTS, which is comprised of operating systems and evaluation tools. Two versions of the HARTS operating system, called HARTOS, have been developed. The first version of HARTOS primarily extends the functionality of the ....

K. G. Shin, "HARTS: A distributed real-time architecture," IEEE Computer, vol. 24, no. 5, pp. 25--35, May 1991.

No context found.

K. G. Shin. HARTS: A Distributed Real-Time Architecture". IEEE Computer, 24:25 -- 35, May 1991.

No context found.

K.G. Shin, "Harts: A Distributed Real-Time Architecture," IEEE Computer, 24(5), pp. 25-35, May 1991.

No context found.

K.G. Shin, "HARTS: A distributed real-time architecture," IEEE Computer, vol.24, no.5, pp.25-35, May 1991. 202

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC