"... In PAGE 6: ... Table1 compares the fault-tolerant memory ele- ments in terms of control lines, data lines, and degree of fault tolerance. All designs retain the minimal static power consumption and rapid recovery time of the original DICE.... ..."

"... In PAGE 8: ... Task flt oleranc -he rsurce tilsatin echprcs iswlbalndwhc elps increase the likelihood that tasks can be executed for a second time in case of a failure being detected. For instance, Table6 presents the schedulability analysis result for the situaion wher task 1is re-executed due to an ero. In this cae l task re stil schlb.... ..."

"... In PAGE 10: ...able 1: Fault classification.......................................................................................................................... 29 Table2... In PAGE 40: ... The design of systems with fault tolerance capabilities to satisfy particular application requirements is a complex process loaded with theoretical and experimental analysis in order to find the most appropriate tradeoffs within the design space. [Suri 95] offers a high-level design paradigm (see Table2 ) extracted from the more detailed description presented in [Avizienis 87]. System properties to be considered include dependability (i.... ..."

"... In PAGE 8: ...2 Fault-tolerance experiments The fault-tolerance experiment results are shown in Fig. 11a, b and Table2 a,b. The fault-tolerance experiments have shown that, by using mobile agents, it is possible to retain the state and data of a failing robot.... ..."

"... In PAGE 2: ... Note that NVP is commonly used to refer to voting-based fault tolerance software. Table1 lists several examples of voter-based algorithms. There are many other voting techniques that may be of practical interest to the reader [Lyu95, Lyu96].... In PAGE 2: ... Hybrid There are also ways of combining the preceding simple techniques to create hybrid fault tolerance approaches. Examples of hybrid strategies, which combine both voter-based and acceptance-based techniques are listed in Table1 . There are many other hybrid techniques... ..."

"... In PAGE 11: ...invalidating any results dependent on results from the o ending worker. Table1 shows results from using these two fault-tolerance mechanisms. In this experiment, we created a saboteur work engine that always corrupts results in a xed way.... ..."

"... In PAGE 10: ... If the results do not match, the o ending worker is blacklisted as untrustable, and the work manager backtracks through all the current results, invalidating any results depending on results from the o ending worker. Table1 shows results from using these two fault-tolerance mechanisms. In this experiment, we created a saboteur work engine that always corrupts results in a xed way.... ..."

"... In PAGE 1: ... The benchmark results are given in Ta- ble 2 for a mixed (CPU, memory, and I/O activity) workload and a stress-based injection strategy. Table2 shows benchmark results based on 50 runs of the benchmark program. The table shows that the same work- load and fault injection method produces four catastrophic incidents in Prototype A, compared with none for Prototype B.... In PAGE 2: ... The performance degrada- tion underfaults may be an important considerationfor time- critical applications. The third row of Table2 shows the av- erage number of faults needed to cause a catastrophic inci- dent. These injections were carried out with a series of faults that the machine was not specifically designed to tolerate, and thus a catastrophic incident ultimately resulted.... In PAGE 2: ... A detailed examination of these measurements is presented later in this paper. However,the numbersin Table2 demonstratethe util- ity of the benchmark in comparing different fault-tolerant machines. Table 2.... ..."

"... In PAGE 1: ... The benchmark results are given in Ta- ble 2 for a mixed (CPU, memory, and I/O activity) workload and a stress-based injection strategy. Table2 shows benchmark results based on 50 runs of the benchmark program. The table shows that the same work- load and fault injection method produces four catastrophic incidents in Prototype A, compared with none for Prototype B.... In PAGE 2: ... The performance degrada- tion under faults may be an important considerationfor time- critical applications. The third row of Table2 shows the av- erage number of faults needed to cause a catastrophic inci- dent. These injections were carried out with a series of faults that the machine was not specifically designed to tolerate, and thus a catastrophic incident ultimately resulted.... In PAGE 2: ... A detailed examination of these measurements is presented later in this paper. However,the numbersin Table2 demonstratethe util- ity of the benchmark in comparing different fault-tolerant machines. Table 2.... ..."

"... In PAGE 4: ... The reliability and availability of an object group (from the point of view of a client) is determined by a set of properties associated with the group. These properties are summarized in Table1 . They can be set when the object group is created and they can be modified later at run-time.... In PAGE 7: ... Instead of relying on a particular implementation, we assign the properties to the packages of the domain and/or group as UML tagged values. The tag name is the same as the property name, and the value is one of the values defined in the FT-CORBA specification (see Table1 ). The values assigned to an object group override the ones assigned to the FT domain.... ..."