HARNESS: Hardware- and Network-Enhanced Software Systems for Cloud Computing

Alexander Wolf
Imperial College London
Topics recommended for the 2016-2017 Work Programme: 

Support for multi-data centre cloud computing - Commercial or governmental entities will increasingly operate multiple data centres. They will do this for various reasons, including to:

  • Lower network access latency.
  • Improve fault resilience.
  • Increase available resources.
  • Adhere to jurisdictional regulations.

While multiple data centres are operated today by many vendors of cloud services, they are treated as independent enterprises for which there is little or no support for flexibly and dynamically deploying or executing applications across them. There is a need for fundamental research into how this can be accomplished, along the following lines:

  • Application deployment and management: physical placement and run‐time adaptation.
  • Data management and dependability: consistency models that can trade resiliency against scalability.
  • Policy‐based regulatory compliance: multiple domains, flexible security, guaranteed disposal.
  • Programming models and software engineering: annotations, architectures, tools, and environments.
Projects major results: 

We expect the following major project outcomes:

  1. The architectural principles that allow heterogeneous hardware and network resources, such as GPUs, FPGAs, SSDs, software routers, and the like, to be seamlessly incorporated into a cloud platform.
  2. The development of a data-centre infrastructure that manages the execution of applications on heterogeneous resources.
  3. The design of low-level management algorithms that optimise the allocation of different types of resources.
  4. The implementation of virtualisation technologies that allow traditionally single-tenant resources to be efficiently shared among multiple tenants.
  5. The design of programming models that allow applications to make effective use of HARNESS-managed clouds.

The following results have been achieved:

  1. A generic IaaS/PaaS architecture that integrates various prototypical heterogeneous resources.
  2. An extended version of an open-source PaaS implementation that mediates between applications and heterogeneous resources.
  3. The development of computation, network and storage resource management systems.
  4. The virtualisation of data-flow computation engines.
  5. The design of aspect-oriented and dataflow programming models for heterogeneous platforms.
Potential exploitation strategy: 

HARNESS is exploring largely new territory, and developing a wide range of novel technologies. Thus, the project does not plan to exploit its outcomes as a single entity (e.g. a start-up). Instead, individual technology components have their own exploitation plan, falling into 3 main categories:

  • Commercialisation as part of existing product offerings: Maxeler and SAP (the two HARNESS industrial partners) plan to evaluate commercialisation of HARNESS results within their products. In Maxeler's case, innovations developed in HARNESS are already part of 2014 product releases.
  • Release as part of open source platforms: HARNESS is contributing to several open source projects, most especially ConPaaS and XtreemFS. Developed technology will add to the capabilities of these platforms.
  • Publication and use for future teaching and research: Outcomes not immediately suitable for commercialisation will be disseminated as the basis for future research and teaching by the academic partners.
An update since the last Concertation meeting (March 2014): 

The consortium has selected a collaboration strategy whereby collaboration with other projects is effected by those partners participating in both projects. This provides significant benefits from collaboration under minimal overhead, which is important given the wide range of HARNESS technology areas. Results from such collaborations include:

  • Integration of HARNESS code within ConPaas and XtreemFS in conjunction with FP7 Contrail.
  • Three patents filed by Maxeler as a result of joint HARNESS and SAVE project development.
  • Collaboration with FP7 FASTER on one of the use case applications, where both Maxeler and Imperial are partners in both projects, providing enhanced use cases for study.

Moving into the third year, the consortium is also considering developing collaborations with earlier-stage projects that may be interested in building upon some of the HARNESS outcomes. One example is with the FP7 POLCA project (where Maxeler is a partner), which may wish to build upon programming tools developed by Imperial during HARNESS.