BARTLETT, J. F. A `nonstop' operating system. In 11 Conference on System Sciences. University of Hawaii, 1978.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....the failure of one component may induce failures in another, e.g. a database failure may cause a program failure. We do not consider coarsegrain replication of hardware, software, data and computation, as for example in Tandem NonStop technology where processes are duplicated on similar machines [Bar78] Nor do we consider programming languages with semantics that reduce the faults that can occur, for example ESFP [TT99] which guarantees the termination of computations. 2.2 Fault Tolerance Layers Fault tolerance is provided at many levels in a system, as depicted in Figure 1. Each layer ....

J.F. Bartlett. A 'NonStop' Operating System. In The 11th Hawaii International Conference on System Sciences (HICSS'78), pages 103-117, Honolulu, USA, January 1978.

....recoverable if s A , s B and s TTP are all recoverable. 3.1. 1 Pessimistic Message Logging A naive approach to achieve recoverability of a fair exchange protocol is to use pessimistic message logging, which has been applied in many systems to support transparent, application independent recovery [5, 9]. The mechanism is pessimistic because it never rolls back process computations. The application of pessimistic message logging to fair exchange is straightforward. In a data exchange system modeled by Figure 3, a buffered message m is always first logged before being delivered to another player. ....

J. F. Bartlett. A `nonstop' operating system. In Proceedings 11th Hawaii International Conference on System Sciences, Hawaii, 1978. 9

....and nancial on line transactional applications. These applications require computer systems to continue functioning in the presence of hardware and software failures. In the past, custom designed, fault tolerant hardware has been used to provide highly reliable and available systems. Tandem [2, 1, 32, 18] and Stratus [33] are examples of such systems. However, custom systems tend to be expensive and do not track technology trends as quickly as commodity hardware. Recent e orts have focused on building fault tolerant systems using clusters of commodity components. The Microsoft cluster service ....

....To make this method work well, it requires an ecient mechanism to ush dirty cache lines of a virtual address space at a checkpoint. 6 RELATED WORK In the past, custom designed fault tolerant hardware has been used to provide highly reliable and available systems. The state of art Tandem system [2, 1, 32, 18] is the rst commercially available system designed speci cally for online transaction processing applications. The Stratus system [33] presents the same inputs to two processor boards and uses a comparison logic on each board for failure detection. All these systems are custom designed, therefore ....

Joel F. Bartlett. A NonStop Operating System. Tandem Computers, Inc., 1977.

....global exchange system state S is recoverable if s A , s B and s TTP are all recoverable. A naive approach to achieve recoverability of a fair exchange protocol is to use pessimistic message logging, which has been applied in many systems to support transparent, application independent recovery [5, 9]. The mechanism is pessimistic because it never rolls back process computations. The application of pessimistic message logging to fair exchange is straightforward. In a data exchange system modeled by Figure 3, a message m is always rst logged before being delivered to another agent. It is easy ....

J. F. Bartlett, A `nonstop' operating system, in: Proceedings 11th Hawaii International Conference on System Sciences, Hawaii, 1978.

....The discs connected to the SCSI busses are conventional mini or micro computer discs. Replication of data is handled by software on higher levels which gives the added capability to dynamically reconfigure the database. This as opposed to disc controller based mirroring or striping ( 29] [4], 33] The number of discs on each node depends on the data access pattern and 1 i486 is a trademark of Intel Corporation storage requirements. This is balanced with the capacity of the SCSI busses and the computational power of the node processor. The individual disc is a customer replaceable ....

Joel F. Bartlett. A NonStop Operating System. In Proc. 11'th Hawaii International Conference on System Sciences, volume 3, pages 103--117, 1978.

....processing nodes. Another means to detect faulty nodes is given by mutual tests between nodes. The most widely used technique is based on the system wide exchange of I m alive messages (or heartbeat) This technique had first been implemented into the fault tolerant multiprocessors of Tandem [6]. 3. System level Diagnosis The step after error detection is fault diagnosis. Its resolution should allow, in case of permanent faults, the reconfiguration of the system in order to isolate faulty nodes, thus limiting the effect of the fault on performance. For the user, the subsequent recovery ....

J.F. Bartlett, A `Non-Stop' Operating System, in: Proc. Hawaii Intl. Conf. of System Sciences, 1978.

.... was typically included in the mirror pair to provide what is known as triple modular redundancy (TMR) Some commercial vendors extended the hardware mirroring concept throughout the entire computer system to provide very high degree of availability in face of hardware component failures [1, 22]. Synchronizing processor pairs at the instruction level provides a transparent means to replicate program instances; the program need not be rewritten in order to run such a processor pair. However, this technique not only requires expensive engineering efforts, but also faces a tremendous ....

J. F. Bartlett. A nonstop operating system. In Proceedings of the 11th Hawaii International Conference on System Sciences, volume 3, 1978.

....criteria is expected to be greater than or equal to the probability of successfully accessing a particular copy. A number of data replication methods have been incorporated into local area network file system services: LOCUS [PW85] ROE [EF83] Isis [Bir85] PULSE [TKW85] Eden [PNP86] Guardian [Bar78], Saguaro [PSA87] Gemini [BMP87] GAFFES [GGK87] and a Unix United variant [Bre86] Most assume that the number of copies is small, and that they are stored among a small number of hosts. A further assumption is that failures are rare, or of suf 3 Three notable exceptions: the 1987 ....

Joel F. Bartlett. "A NonStop operating system." In Proceedings of the Eleventh International Conference on Systems Sciences, pp. 103--117. Society for Computer Simulation, January 1978.

No context found.

BARTLETT, J. F. A `nonstop' operating system. In 11 Conference on System Sciences. University of Hawaii, 1978.

No context found.

Joel F. Bartlett. A nonstop operating system. In Proceedings 11'th Hawaii International Conference on System Sciences, volume 3, pages 103--117, 1978.

No context found.

Joel F. Bartlett. A NonStop Operating System. Tandem Computers, Inc., 1977.

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