Summary by Suman Nath. Track-Aligned Extents: Matching Access Patterns to Disk Drive Characteristics, Jiri Schindler, John Linwood Griffin, Chistopher R. Lumb, Gregory R. Ganger The following points about the paper are noteworthy: 1. Utilizing disk geometry to make the extents track-aligned, a system can avoid most rotational latency and track crossing overhead-- which are significant relative to other delays. This approach can be helpful to achieve high disk efficiency with smaller request sizes. While benefiting some special type of services (e.g., video server), it does not hurt the general cases. 2. Timing information to read blocks from disk can be used to find geometry of disk. 3. The paper gives good background material for general understanding of the overall area, the problem and solution. Take home message: 1. Sometimes a system can be made more efficient by looking at it in different level of abstraction. Higher-level abstraction may not always yield better performance -- details hidden by the abstraction may sometimes be a good place to optimize performance. 2. Whenever possible, try to avoid the more expensive components of the overhead -- even at the cost of more inexpensive resources. It is OK if there is some performance gain for some specific case if there is no performance degradation for more general case. 3. When changing an existing system, try to keep the user interface unmodified. Deficiencies (What could be done more): 1. No detailed description of traxtent allocation algorithm. It is not clear how it will find a suitable track to allocate for new data, and what are the options when no such track is found or multiple tracks of different sizes are available. This decision might be important for the overall performance. 2. No detailed analysis of wastage of space in the disk. 3. Experiments should have included I/O size bigger than the track size, to see how close it gets to the unaligned case for bigger files.