High availability for parallel computers

1996

For massively parallel systems, the probability of cr s~Yslenc failure clue to u random hardware fault becomes statistically very significant because of the huge number of components. Besides, filult injection experiments show that multiple failures go undetected, leading to incorrect results. Hence, massively parallel systems reguirc abilities to tolerate: these faults that will occur. The FTMPS project presents a scalable implementation to integrate the different steps to,laull tolerance into existing HPC systems . On the initial parallel .system only 4017v of (randomly injected),faulls do not cause the application to crash or produce wrong results . 1n. the resulting FTMPS prototype more than. 80%, of these ftiults are correctly detected and recovered. Resulting overhead for the application is only between 10 and 20%. Evaluation. of the different, co-operating fault tolerance modules shows the,llexibility and the ,.scalability of the approach.

Euro-Par 2008–Parallel …, 2008

The current supercomputers are almost achieving the petaflop level. These machines present a high number of interruptions in a relatively short time interval. Fault tolerance and preventive maintenance are key issues in order to enlarge the MTTI (Mean Time To ...

2012

The development of scientific software, reliable and efficient, in distributed computing environments, requires the identification and the analysis of issues related to the design and the deployment of algorithms for high-performance computing architectures and their integration in distributed contexts. In these environments, indeed, resources efficiency and availability can change unexpectedly because of overloading or failure i.e. of both computing nodes and interconnection network. The scenario described above, requires the design of mechanisms enabling the software to "survive" to such unexpected events by ensuring, at the same time, an effective use of the computing resources. Although many researchers are working on these problems for years, fault tolerance, for some classes of applications is an open matter still today. Here we focus on the design and the deployment of a checkpointing/migration system to enable fault tolerance in parallel applications running in distributed environments. In particular we describe details about HADAB, a new hybrid checkpointing strategy, and its deployment in a meaningful case study: the PETSc Conjugate Gradient algortithm implementation. The related testing phase has been performed on the University of Naples distributed infrastructure (S.Co.P.E. infrastructure).

International Journal of Engineering Sciences & Research Technology, 2013

Software almost inevitably contains defects. Do everything possible to reduce the fault rate; Use faulttolerance techniques to deal with software faults. Fault tolerance is the ability of a system to perform its function correctly even in the presence of internal faults. Most of the ordinary systems lack fault tolerant software fix. This paper surveys various software Fault Tolerance techniques and methodologies. The conventional fault tolerant approaches viz., Recovery Block (RB), N Version Programming (NVP) etc., are too costly to fix in an ordinary lowcost application system because, both the RB and NVP rely on multiple (at least three) versions of both software and computing machines.

PUBLISHED BY IMPERIAL COLLEGE PRESS AND DISTRIBUTED BY WORLD SCIENTIFIC PUBLISHING CO. eBooks, 2006

Today's highly available systems deliver four years of uninterrupted service. The challenge is to build systems with 100-year mean time to failure and one-minute repair times. September 1991 aradoxically, the larger a system is, the more criticalbut less likelyit is to be highly available. We can build small ultra-available modules, but building large systems involving thousands of modules and millions of lines of code is a poorly understood art, even though such large systems are a core technology of modern society. Three decades ago, hardware components were the major source of faults and outages. Today, hardware faults are a relatively minor cause of system outages when compared with operations, environment, and software faults. Techniques and designs that tolerate these broader classes of faults are still in their infancy. This article sketches the techniques used to build highly available computer systems. Historical perspective Computers built in the late 1950s offered a 12-hour mean time to failure. A maintenance staff of a dozen full-time computer engineers could repair the machine in about eight hours. This failure-repair cycle provided 60 percent availability. The vacuum tube and relay components of these computers were the major sources of failures; they had lifetimes of a few months. So the machines rarely operated for more than a day without interruption.' Many fault-detection and fault-masking techniques used today were first used on these early computers. Diagnostics tested the machine. Self-checking computational techniques detected faults while the computation progressed. The program occasionally saved (checkpointed) its state on stable media. After a failure and repair, the program read the most recent checkpoint and continued the computation from that point. This checkpoint-restart technique let computers that failed every few hours perform long-running computations. Device improvements have increased computer system availability. By 1980, typical well-run computer systems offered 99 percent availability.2 This sounds good, but 99 percent availability is 100 minutes of downtime per week. Such I

1994

In massively parallel systems (MPS), fault tolerance is indispensable to obtain proper completion of long-running computation-intensive applications . To achieve this at reasonable low cost, we present a global approach . A flexible and powerful backbone is provided through the combination of hardware and software error detection techniques, fault diagnosis and operator-site software together with reconfiguration of the system . Application recovery is based on checkpointing and rollback . The red line (i.e. applicability for a massively parallel system) comprises scalability as well as simplicity. A unifying system model is introduced that allows the mapping of a global concept for fault tolerance to a wide variety of MPS . The framework for implementation in an existing MPS is discussed .' KEY WORDS fault tolerance ; massively parallel system ; error detection; fault diagnosis ; backward error recovery : reconfiguration

IEEE Micro, 1998

Journal of Global Research in Computer Science, 2011

In this paper the author has concentrated on architecture of a cluster computer and the working of them in context with parallel paradigms. Author has a keen interest on guaranteeing the working of a node efficiently and the data on it should be available at any time to run the task in parallel. The applications while running may face resource faults during execution. The application must dynamically do something to prepare for, and recover from, the expected failure. Typically, checkpointing is used to minimize the loss of computation. Checkpointing is a strategy purely local, but can be very costly. Most checkpointing techniques, however, require central storage for storing checkpoints. This results in a bottleneck and severely limits the scalability of checkpointing, while also proving to be too expensive for dedicated checkpointing networks and storage systems. The author has suggested the technique of replication implemented on it. Replication has been studied for parallel databases in general. Author has worked on parallel execution of task on a node; if it fails then self protecting feature should be turned on. Selfprotecting in this context means that computer clusters should detect and handle failures automatically with the help of replication.

2017

Checkpoint and recovery cost imposed by checkpoint/restart (CP/R) is a crucial performance issue for high-performance computing (HPC) applications. In comparison, Algorithm-Based Fault Tolerance (ABFT) is a promising fault tolerance method with low recovery overhead, but it suffers from the inadequacy of universal applicability, i.e., tied to a specific application or algorithm. Till date, providing fault tolerance for matrix-based algorithms for linear systems has been the research focus of ABFT schemes. As a consequence, it necessitates a comprehensive exploration of ABFT research to widen its scope to other types of parallel algorithms and applications. In this thesis, we go beyond traditional ABFT and focus on other types of parallel applications not covered by traditional ABFT. In that regard, rather than an emphasis on a single application at a time, we consider the algorithmic and communication characteristics of a class of parallel applications to design efficient fault tole...

Journal of Systems and Software, 1997

Providing reliable services is one of the primary goals in designing a distributed operating system. Nowadays, we h a ve seen a trend in distributed operating system design to shift from large kernel architectures or even monolithic architectures to microkernel architectures supported by the client server model. This means that a lot of services of an operating system originally provided by the monolithic kernel are moved out of the kernel, forming individual servers. It is then crucial to guarantee that these servers will provide reliable services. This paper describes a design, based on a twin-servers model, of fault-tolerant servers for the microkernel-based RHODOS distributed operating system. A model that supports faulttolerant services is designed. The performance of the model is simulated and analysed. A design that implements the model in the RHODOS environment is also outlined.