Anatidae: A distributed system of highly available objects

This project is a join effort between CMU and Bellcore. The following people are the current members of the research team:

Lisa Bahler
Mootaz Elnozahy
Vivek Ratan
Mark Segal

Publications

"Experiences Using DCE and CORBA to Build Tools for Creating Highly-Available Distributed Systems," E.N. Elnozahy, V. Ratan, and M.E. Segal. In Proceedings of the IFIP International Conference on Open Distributed Processing (ICODP'95), February, 1995. Also available as technical report CMU-CS-95-117.
"Highly Available Directory Services in DCE," B. Acevedo, L. Bahler, E.N. Elnozahy, V. Ratan, and M.E. Segal. In Proceedings of the Twenty Sixth International Fault Tolerant computing Symposium (FTCS-26), June, 1996. Also available as technical report CMU-CS-96-???.

Status

Prototype of a highly available, and correct, DCE CDS services is being developed.
Implementation of a prototype of a highly available CORBA-compliant name server is underway.

Summary

High availability is a major concern for many distributed applications. A failure in a computer-controlled system could result in problems ranging from minor inconveniences to its users to a major disaster that leads to catastrophic loss of human life. Traditional techniques for adding high availability to these applications typically rely on nonstandard systems built out of custom software and hardware. These systems are expensive and difficult to maintain, and applications written on one system cannot be easily ported to other platforms. We are building a software layer that supports adding fault tolerance and high availability to distributed applications written on standard workstations and operating systems. The proposed software layer allows the creation of highly-available objects that are portable across a variety of heterogeneous hardware platforms. Our techniques are built on the CORBA and DCE standards of distributed computing, ensuring interoperability with other systems that conform to the same standards, and enhancing the commercial viability of our approach. Along with these goals, the proposed system allows legacy applications that were written for custom hardware systems to migrate to standard platforms without major software redesign. The major contributions and innovations that we claim in this work are:

application-transparent fault tolerance. Our techniques transparently add fault tolerance to distributed objects with no programmer intervention. This approach reduces the complexity of writing distributed applications by relieving the programmer from the error-prone and mundane tasks of handling failure and recovery at the application level. The proposed techniques also transparently retrofit legacy applications that were designed without consideration for fault tolerance or were written on fault-tolerant hardware platforms.
software-based TMR systems. Our approach allows a Triple Modular Redundancy (TMR) system to be built out of stock workstations connected by a standard network. Using custom TMR hardware systems has been a popular technique for many mission critical applications. The proposed techniques enable these applications to run on standard platforms with the same degree of fault tolerance and high availability but at a lower cost.
flexibility through multiple mechanisms. The proposed software layer uses seven techniques for adding high availability and fault tolerance to distributed objects, in contrast to existing systems that are built around a single technique or abstraction. Our approach recognizes that different applications have different availability requirements, and allows the application developer to select the level of availability and associated performance penalty that are most appropriate for the application.
on-line software installation. Installation of new versions software to add functionality or repair bugs can lead to downtime. The proposed software layer includes mechanisms that install new software releases without causing application downtime or significant performance degradation (dynamic program updating).

The Name

Anatidae is a latin word for the family of birds under which geese are classified. The selection of the name was inspired by the resilience of the Canadian geese that occupy Bellcore's parking lot. These geese are able to continue operation not withstanding adverse working conditions ranging from fatal car accidents to the subfreezing temperature during the winter and the scorching sun during the summer. The currently contrived acronym should read as "AN Availability Toolkit for Implementing Distributed Applications Everywhere".

Back to Mootaz's home page