Optimization and functional improvement of disk arrays
- Autores: José Luis González Compeán
- Directores de la Tesis: Antonio Cortés Roselló (dir. tes.)
- Lectura: En la Universitat Politècnica de Catalunya (UPC) ( España ) en 2009
- Idioma: español
- Tribunal Calificador de la Tesis: Jesús Carretero Pérez (presid.), Xavier Martorell Bofill (secret.), André Brinkmann (voc.), María Pérez Hernández (voc.), Stergios Anastasiadis (voc.)
- Materias:
- Texto completo no disponible (Saber más ...)
- Resumen
Redundant Arrays of inexpensive disks or RAID systems have been a traditional solution for achieving improved I/O performance, increased capacity and strong data availability.
RAID systems were proposed on 80's decade as a static storage solution using a fixed data layout, which allows using deterministic methods for allocating and locating data within disks of the array. A deterministic data placement method produces balanced load per disk, low compute complexity and does not degrade with large volumes of information.
However, the capacity and I/O performance demands of current applications are growing at exponential rates, which means performing periodical upgrade processes adding new disks to RAID systems in order to satisfy the applications demands.
As the new disks are not included in the data layout, an upgrade process that completely rearranges the data layout is required for taking advantage of capacity and bandwidth of new disks while preserving the benefits of a fixed data layout.
This kind of upgrade is not quite suitable for applications working in 24x7 manner because induces overhead to the normal operation of disk arrays, which can be suffered even during weeks when upgrading environments managing large volumes of information.
The global idea of our thesis is proposing adaptive disk arrays using online upgrade strategy in order to keep the benefits of the deterministic data placement while reducing the effects of upgrade operations on the normal operation of disk arrays.
In the first part of this job we propose an online re-striping strategies to upgrade RAID systems to increase the capacity and bandwidth of the system with special focus on reducing the effects of the online upgrade process.
In the second part of this job we extend our upgrade strategies to a heterogeneous storage systems in order to avoid wasting capacity and performance from disks with higher capabilities.
In the third part of this job we propose a new adaptive data placement method based on deterministic zones. In order to reduce the time employed to upgrade the RAID system and using our upgrade strategies in mass storage RAID systems, this proposal only moves a fraction of the data layout by performing a partial re-striping with special focus on increasing or keeping the parallelism in the final configuration delivered by the upgrade process.
The last part of this job proposes an adaptive data placement method able to preserve the effectiveness of the striping redundancy schemes even after several upgrade processes.
This proposal is focused on delivering strong availability for heterogeneous disk arrays able to adapt its data layout to the capacity/performance demands.
