Citation Context

...cessor and disk [1]. Further, such a disk system has and can exploit low-level information not typically available at the file-system level, including exact head position and blockmapping information =-=[26, 35]-=-. Finally, unmodified file systems can leverage these optimizations, enabling deployment across a broad range of systems. Unfortunately, while smart disk systems have great promise, realizing their fu...

Citation Context

...etic media. The best case, sequential streaming of data, achieves disk head efficiency of approximately 0.9, falling below 1.0 because no data is transferred when switching from one track to the next =-=[28]-=-. Non-sequential access patterns usually achieve efficiencies well below 0.1, as seek time and rotational latency dominate data transfer time. For example, a disk with an average seek time of 5 ms tha...

Citation Context

... access latency. With correlation-directed disk layouts, the system only needs to pay a one-time seek and rotational delay to get multiple blocks that are likely to be accessed soon. Previous studies =-=[52, 54]-=- have shown promising results in allocating correlated file blocks on the same track to avoid trackswitching costs. Correlations can also be used to direct storage caching. For example, a storage cach...

Citation Context

...until they are swapped out and are associated with some disk locations. To map the LBN (Logical Block Number) of a block into a onedimensional physical disk address, we use the technique described in =-=[33]-=- to extract track boundaries. To characterize accurately block location sequentiality, all the defective and spare blocks on disk are counted. We also artificially place a dummy block between the bloc...

Citation Context

...death within storage removes this overhead. Intelligent prefetching: Modern disks perform aggressive prefetching; when a block is read, the entire track in which the block resides is often prefetched =-=[22]-=-, and cached in the internal disk cache. In an aged (and thus, fragmented) file system, only a subset of blocks within a track may be live, and thus, caching the whole track may result in suboptimal c...

Citation Context

...ize its behavior. For example, with simple probes in microbenchmarks, one can determine parameters of the the memory hierarchy [2, 24], processor cycle time [28], and characteristics of disk geometry =-=[26, 30]-=-. In our view, the key difference between fingerprinting and microbenchmarking is that a fingerprint is used to discover the policy or algorithm employed by the underlying layer, whereas a microbenchm...

Citation Context

...rating storage vendors and file-system implementors. Another example of the benefits of low-level knowledge of disk characteristics is found in Schindler et al.'s recent work on track-aligned extents =-=[36]-=-. Therein, the authors explore the range of performance improvements possible when allocating and accessing data on disktrack boundaries, thereby avoiding rotational latency and track-crossing overhea...

Citation Context

...r disk, the more accurate an estimate a diskaware policy can make about access costs. For example, knowing that blocks span a track boundary informs that access would incur the track crossing penalty =-=[24]-=-. As another example, knowing that a set of non-contiguous blocks have some spatial locality, the scheduler could infer that access of these blocks would incur the cost of semi-sequential access, inte...

Citation Context

...s not provide a generic solution to improve overall disk I/O performance of the system. For throughput improvement, Schindler et al. have proposed free-block scheduling [18] and track-aligned extents =-=[31]-=- which use intelligent I/O scheduling rather than block reorganization. These are complementary techniques that can be used in conjunction with BORG. Among block level approaches, our work is closest ...