Virtualization techniques for memory resource exploitation

• Autores: Luis Angel Garrido Platero
• Directores de la Tesis: Paul Matthew Carpenter (dir. tes.), Rosa M Badia (codir. tes.)
• Lectura: En la Universitat Politècnica de Catalunya (UPC) ( España ) en 2019
• Idioma: español
• Tribunal Calificador de la Tesis: Anastassios Nanos (presid.), Petar Radojkovic (secret.), Vinicius Petrucci (voc.)
• Programa de doctorado: Programa de Doctorado en Arquitectura de Computadores por la Universidad Politécnica de Catalunya
• Materias:
  • Ciencias tecnológicas
    • Tecnología de los ordenadores
      • Arquitectura de ordenadores
• Texto completo no disponible (Saber más ...)
• Resumen

  • Cloud infrastructures have become indispensable in our daily lives with the rise of cloud-based services offered by companies like Facebook, Google, Amazon and many others. These cloud infrastructures use a large numbers of servers provisioned with their own computing resources. Each of these servers use a piece of software, called the Hypervisor (``HV''), that allows them to create multiple virtual instances of the server's physical computing resources and abstract them into "Virtual Machines'' (VMs).

    A VM runs an Operating System, which in turn runs the applications. The VMs within the servers generate varying memory demand behavior. When the demand increases, costly operations such as (virtual) disk accesses and/or VM migrations can occur. As a result, it is necessary to optimize the utilization of the local memory resources within a single computing server.

    However, pressure on the memory resources can still increase, making it necessary to migrate the VM to a different server with larger memory or add more memory to the same server. At this point, it is important to consider that some of the servers in the cloud infrastructure might have memory resources that they are not using. Considering the possibility to make memory available to the server, new architectures have been introduced that provide hardware support to enable servers to share their memory capacity.

    This thesis presents multiple contributions to the memory management problem. First, it addresses the problem of optimizing memory resources in a virtualized server through different types of memory abstractions. Two full contributions are presented for managing memory within a single server called SmarTmem and CARLEMM. In this respect, a third contribution is also presented, called CAVMem, that works as the foundation for CARLEMM.

    Second, this thesis presents two contributions for memory capacity aggregation across multiple servers, offering two mechanisms called GV-Tmem and vMCA, this latter being based on GV-Tmem but with significant enhancements. These mechanisms distribute the server's total memory within a single-server and globally across computing servers using a user-space process with high-level memory management policies.