Current projects on probe-based storage include those by the Carnegie
Mellon Center for High Integrated Information Processing and Storage
Systems (CHI
PS [3]), Despont et
al. [25], Mamin et al. [11], and the
Atomic Resolution Storage (ARS) project at Hewlett-Packard
Laboratories [23]. While these projects use different
recording technologies, they are all based on tip arrays and media
sleds. Thus, the models we describe can be tuned with
different input parameters to describe these systems.
Disk modeling has been traditionally used to design storage systems that use hard drives, and a complete survey of this work is beyond the scope of this paper. Because disk performance is so workload-dependent, most simplifying assumptions cause large errors [16]. Shriver [19] developed some analytic models to incorporate effects of disk caching and I/O workload variation.
Because probe-based storage devices do not yet exist, it is particularly useful to create models of them that can yield insights into performance. Yang and Madhyastha created several physical models for seek time of probe-based storage, defining a performance range for a specific hardware configuration [10]. A different model was presented by Schlosser and Griffin et al. [6,17], who conclude that a probe-based storage device can improve applications performance by a factor of three over disks. They compare and contrast probe-based storage and traditional disk drives, and study how aspects of the operating system need to be changed when a system is built with probe-based storage [7]. Uysal et al [24] evaluate several hybrid MEMS/disk architectures, showing that hybrid architectures can give performance benefits similar to replacing disks with probe-storage devices (at lower price-performance). Ying et al [9] used this seek-time model to devise policies for power conservation. However, these studies all rely upon trace-driven simulation of traces. In contrast, our focus here is to develop an analytical model for a wide range of probe-based storage characteristics that can be used for such performance research.
Probe-based storage devices are much faster than traditional disk drives, making the question of how probe-based storage may be integrated into the memory hierarchy very important. Griffin et al [18] show that using probe-based storage as a disk replacement will improve overall application runtime by a factor of 1.9-4.4, and when used as a disk cache can improve I/O response time by up to 3.5 times.