A4 - Adaptive non-volatile storage systems for adaptive applications (TES.6369)

Goals

"The A4 project will introduce the concept of adaptivity into mass storage systems. The central idea is to vary the quality of service (QoS) provided to the user, depending on the available power budget. You could for instance reduce the quality of the video you are watching if you wish to conserve battery power. Current mass storage devices, like the Flash memory and hard-discdisk, will be considered. They are however not very adaptive. Therefore we will also look at new systems based on probe storage." --- http://dies.cs.utwente.nl/research/#a4

Research environment

The A4 project is conducted by the Systems and Materials for Information storage (SMI) group (http://www.el.utwente.nl/smi/) (Electrical Engineering) and the Embedded Systems cluster (http://wwwes.cs.utwente.nl/) (Computer Science) of the Faculty of Electrical Engineering, Mathematics and Computer Science (EEMCS) (http://www.ewieemcs.utwente.nl/en/index.html) at the University of Twente (http://www.utwente.nl/en/).
The Embedded Systems cluster exists of two research groups, namely Distributed and Embedded Systems (DIES) (http://dies.cs.utwente.nl/) and Computer Architecture, Design and Testing of Embedded Systems (CADTES) (http://wwwes.cs.utwente.nl/cadtes/). The A4 project is part of both groups.

Probe storage vs. rotating disk storage

In the existing literature, three basic differences between probe storage systems and rotating disk storage systems are identified:

• Heads (or actually: the medium) can be moved in two dimensions instead of just one. This could decrease seek time dramatically. Also, bits could be aligned in more than one direction, opening possibilities for designing very efficient application-specific data layouts. However, researchers at CMU assume that these benefits will be limited by the fact that stabilisation of the medium after movement in the direction orthogonal to the bitstream direction will take considerable amounts of time.
• As probe storage devices are expected to have thousands of heads, interesting techniques may be developed for their on/off switching patterns. This is especially challenging since many researchers expect that not all heads may be active simultaneously due to excessive heat dissipation and limited peak power requirements.
• There is no rotational delay. This is an especially important difference when the same or neighbouring (in the head-movement direction of a harddisk) pieces of the medium need to be accessed sequentially, such as in the case where information needs to be read, quickly processed and then written back.

• Each of the clusters of heads (there may for example be 64 clusters of 81 heads) may be moved (or actually, the corresponding medium tile is moved) independently, essentially creating what could be compared to a very large disk array. Research has been done concerning how disk arrays can be more effectively (than through by pretending to be a single IDE/SCSI disk) accessed for certain workloads. The difference here is that the clusters are larger, possibly not every part of the cluster can be accessed simultaneously, and that the features mentioned in the list above should be taken into account.