The Laboratory for Computer Science is sharing a $12 million grant to develop a secure, decentralized Internet infrastructure that's resistant to failure and attack.
As concerns rise over the vulnerability of the Internet and other networked applications to malicious attack, securing distributed applications has become a high priority for governments, institutions and businesses worldwide. Today's traditional client-server approach to distributed systems suffers from significant security and scalability problems when hosting complex applications over wide area networks.
In response to this problem, the Iris project - Infrastructure for Resilient Internet Systems - is being launched by four organizations: MIT, the University of California at Berkeley, the International Computer Science Institute (ICSI) at New York University, and Rice University. Iris is funded by an Information Technology Research award from the National Science Foundation.
The Iris project aims to use "distributed hash table" technology to develop a robust common framework and infrastructure for distributed applications without creating central points of vulnerability. The secure networks that emerge from this project will streamline distributed application programming and offset development expenses.
The project will investigate two conjectures: that a wide variety of distributed applications can be built on a distributed hash table platform (a peer-to-peer network with no central controlling server) with application-independent, unconstrained keys and values; and that these applications can inherit basic levels of security, robustness, ease of operation and scaling from the technology.
The project is led by Frans Kaashoek, professor of electrical engineering and computer science (EECS) at MIT, and ICSI's Scott Shenker.
In the first phase of the project, researchers will develop distributed hash table algorithms and building a large-scale open testbed equipped with some enabling libraries and a few sample applications. They will then be encouraged to build - and break - more advanced applications on the network. Testing the second conjecture will involve developing techniques that enable applications to maintain their performance and security even when operating over nodes controlled by hackers.
Acting as the cornerstone of the new robust shared infrastructure, distributed hash table technology will securely orchestrate data retrieval and computation on open-ended large-scale networks such as the Internet, even when the individual nodes on the network are insecure or unreliable.
The underlying network will also be self-configuring, allowing the addition and removal of nodes without manual oversight while also automatically balancing excess loads across the network. The desired end result is a large distributed system that's reliable, even though it may be composed of inexpensive and unreliable components.
While one of the immediate goals of the research is to protect widely distributed applications from eavesdropping, tampering and malicious destruction, some long-term goals for the future of software may prove even more important. Researchers foresee the technology being used as the platform for robust, global Internet applications, vastly improved wide area storage networks and more.
"A peer-to-peer approach to distributed systems has gained momentum in recent years because it offers scalability and robustness, but a lot of critical research problems remain," said Professor Victor Zue, director of the Laboratory for Computer Science. "With some of the best minds in this community collaborating and with sustained support from NSF, significant advances will undoubtedly be made."
The Iris project includes a multidisciplinary team of researchers from fields including networking, algorithms, security, systems and databases. Others involved from MIT's EECS department include associate professors Hari Balakrishnan and David Karger; Barbara Liskov, the Ford Professor of Engineering and associate department head; and Assistant Professor Robert Morris.
In addition to directing pure research, the Iris project intends to build outreach and education programs to steer graduate and undergraduate computing programs toward distributed applications. Students at each of the participating institutions will be encouraged to join the distributed hash table testbed. The group hopes to build on student interest in peer-to-peer technologies to create the distributed applications of the future.
A version of this article appeared in MIT Tech Talk on October 2, 2002.
