Description

A system's ability to function normally in the event of one or more faults in one or more of its components is known as fault tolerance. In contrast to a naively designed system, where even a minor failure can result in a complete breakdown, a properly designed system's operating quality decreases proportionally to the severity of the failure. In high-availability, mission-critical, or even life-critical systems, fault tolerance is especially important. Grateful degradation is the capacity of a system to continue functioning even when parts of it fail. When one component of the system fails, a fault-tolerant design allows the system to continue its intended operation, albeit at a reduced level, rather than completely failing. The majority of the time, the term is used to describe computer systems that are built to operate almost completely even in the event of a partial failure, with the possibility of reducing throughput or increasing response time. That is, there are no hardware or software issues that stop the system as a whole. A vehicle designed to continue driving even if one of its tires is punctured or a structure that can maintain its integrity in the face of damage from fatigue, corrosion, manufacturing flaws, or impact is examples from another field.

Fault-Tolerant Design

Fault tolerance can be achieved within the scope of a single system by anticipating unusual conditions and designing the system to deal with them. In general, the goal should be self-stabilization so that the system converges on an error-free state. However, utilizing some form of duplication may be a better option if the consequences of a system failure are catastrophic or the cost of making it sufficiently reliable is extremely high. In any case, the system must be able to use reversion to return to a safe mode in the event of a catastrophic system failure. This can be a human action if humans are present in the loop, like roll-back recovery. Antonin Svoboda built SAPO, the first known fault-tolerant computer, in 1951 in Czechoslovakia. Its basic design consisted of magnetic drums connected by relays and a voting method for memory error detection (triple modular redundancy). This led to the creation of several additional machines, the majority of which were intended for use in the military. They eventually fell into three distinct groups: machines that wouldn't need to be maintained for a long time, like the ones on NASA space probes and satellites; computers that were very reliable but needed to be constantly watched, like those used to control and monitor nuclear power plants or experiments on supercolliders; Lastly, computers that have a lot of runtime and would be used a lot, like many of the supercomputers that insurance companies use to monitor their probabilities. NASA carried out the majority of the development of the so-called LLNM computing in the 1960s in preparation for Project Apollo and other aspects of research. The JSTAR computer, the second attempt by NASA, was utilized in Voyager. The first machine was used in a space observatory. The computer was known as the JPL Self-Testing-And-Repairing computer because it had a backup of memory arrays so that it could use memory recovery techniques. It could fix its own mistakes or bring up redundant modules when necessary. As of early 2022, the computer is still operational.

Triple Modular Redundancy

In the United States, aircraft manufacturers, 210 nuclear power companies, and the railroad industry were the primary innovators of hyper-dependable computers. These computers needed to have a lot of uptime and be able to gracefully fail if something went wrong. They also needed to rely on the fact that the output of the computers would be constantly monitored by humans to find problems. Again, IBM created the first computer of this kind for NASA's Saturn V rocket guidance, but later BNSF, Unisys, and General Electric built their own versions. There has been a lot of work done in the field since the 1970s, such as; Redundancy and self-testing were built into F14 CADC. As a rule, the early endeavors to blame open minded plans were centred primarily around interior determination, where an issue would show something was fizzling and a labourer could supplant it. For example, SAPO had a way to tell when memory drums were failing by making a noise. Later efforts demonstrated that the system needed to be self-repairing, diagnosing and isolating a fault, implementing a redundant backup, and alerting about the need for repair in order to be fully functional. Faults cause automatic fail-safes and a warning to the operator in this type of N-model redundancy, which is still the most prevalent level one fault-tolerant design in use today. Voting was another initial method, as previously mentioned. Multiple redundant backups were constantly running and checking each other's results, so if four components gave a 5 and one gave a 6, the other four would "vote" that the fifth component was defective and have it taken out of service. The term for this is "M out of N majority voting." Because of the complexity of systems and the difficulty of ensuring that the transitive state from fault-negative to fault-positive did not disrupt operations, the historical trend has always been to move further from the N-model and more toward the M model instead of the N model. Tandem and Stratus were among the first businesses to focus on developing fault-tolerant computer systems for processing online transactions. In some cases, hardware fault tolerance necessitates that broken components be removed from the system and replaced with new ones while it is still operational a process known as "hot swapping" in computing. The vast majority of fault-tolerant systems are single point tolerant, or a system that works with only one backup. These kinds of systems should have a mean time between failures that is long enough for the operators to fix the broken devices mean time to repair before the backup fails as well. Although a fault-tolerant system does not specifically require this, it is helpful if there is as much time as possible between failures.