Skip to main content
more options

Project Overview

Cloud image

Current benefits and challenges in Cloud Computing

Cloud computing offers many advantages including:

  • Ability to collect massive data sets and apply massive computing power on demand (as needed)
  • New machine learning tools optimize decision making
  • Potential to support huge numbers of user systems

But today’s cloud is inconsistent (often by design) and poorly secured. Imagine offering home-care patients automated insulin pumps, with blood sugars relayed to the cloud system and adjusted insulin doses relayed back. Today's clouds lack the reliability and security to trust with applications such as direct patient care. Other applications where current cloud systems do not offer high enough levels of trust include real-time image support for surgical procedures, cloud-hosted financial systems, technology to control a smart electric power grid, disaster response, and information systems used by law enforcement or the military.

A Scientific Approach to Scalable Trust

We plan to look closely at the consistency issue as it arises in large-scale systems that replicate information or execution, or that indirectly depend on replication, for example as part of a time-critical response loop, or in support of a cloud-scale security or privacy guarantee. We believe this problem can be solved, and that doing so will enable progress on the full spectrum of properties reducible to replication. Moreover, because the central question here is to study scalability in previously overlooked dimensions (for example, by looking at efficient sharing when protocols contend for scarce resources, and at robustness relative to metrics such as freedom of oscillatory behaviors, convoy phenomena, or bottlenecks), our work will also illuminate questions that might usefully be studied in other domains. Success will contribute towards a scientific foundation for scalable trust in cloud computing.

More information about the project is in this Chronicle article.

People

back to top
Ken Birman

Ken Birman

Robbert VanRenesse

Robbert VanRenesse

Hakim Weatherspoon

Hakim Weatherspoon

Project Details

back to top

Key to work under the NSF CiC project is testing, primarily through support to student experiments in partnership with Microsoft's Azure cloud platform. While we do a great deal of theoretical work and feel that doing so is of critical importance, theory cannot easily anticipate the oddities of cloud scheduling, performance and latency variability stemming from scheduling phenomena, message loss and resource contention, and even the overheads introduced by cloud platforms such as Azure. This experimental work is the sole way of gaining the needed hands on insight and of discovering the relevant performance and scale-limiting barriers.

Isis2 Cloud Platform - Birman's group will demonstrate on real cloud platforms that highly assured data replication is in fact possible at performance and scalability levels comparable to standard approaches.

Superclouds - Weatherspoon's group will demonstrate virtualization and cloud storage with: computation migration, replication, and consistency; VirtualWire, virtual network migration; and reliable cloud storage even in a contested environment.  

Van Renesse's group is building a new kind of elastic replication layer that combines Byzantine Fault Tolerant state machine replication with distributed hash tables that shard data into small groups organized along the edge of the DHT ring. He then implements a standard put/get API, but with far stronger consistency and assurance properties than are possible with any standard solution to this problem.

GridControl - Our group is working with cloud computing projects in the smart power grid community, where there is a desire to host high assurance power grid monitoring and control applications on the cloud and to explore the potential of the cloud model for real-time reactive control algorithms that might optimize the configuration of the grid as conditions evolve, or even intervene to prevent a blackout.