Group 2

EEC 581 Computer Architecture, Spring 2014

Title:An Empirical Study to Identify Common Properties and Type-specific Behavior of Solid State Drive Goal:In this Paper, we accomplish empirical, database-centric performance measurements for different SSDs, explain the results and try to derive common characteristics. Overview:This Paper is based on the common properties and type specific behavior of different solid state drives. A solid state drive does not actually have any disk or motor to drive the disk. Due to its lack of mechanical parts it works faster as compared to regular hard disk drives. Solid State Drive technology uses electronic interfaces compatible with traditional block input/output hard disk drives. That makes them perfect candidates for incorporating into database systems to speed up data processing. In this paper they have performed the comparison of 5 different solid state drives from different companies and the data sheets provided by manufacturers. In this paper the problem addressed is that if solid state drives are completely reliable and can replaced the regular hard drives or not. Solid-state disks are using flash chips for persistently storing data. An abstraction layer on top of the chips provides a block device interface and hides the specific flash chip characteristics. Solid State drive research dates back in 1950s were they were trying to make solid state drives using Random access memory and similar technology. Later in the 1990s they introduced flash based solid state drives. That revolutionized the solid state drives and ever since then there have been a lot of research done on building solid state drives using the flash memory. Flash chips are used to store persistent data on Solid state drive in a matrix of storage cells. The cells can either embody NOR or NAND gates. The way this paper is different from the previous research is that they are comparing different manufactures Solid state drives as follows - SSD1 SuperTalent FSD32GC35M 32GB, SSD2 Mtron MSP-SATA7525, SSD3 Intel X25-M G1, SSD4 Intel X25-M G2, SSD5 Crucial RealSSD. The first test pattern they used is sequentially accessing n pages. This pattern is common in database servers when scanning through a table where no tailored access paths are available. The second pattern is randomly accessing all pages of the test file. This pattern is often seen in databases when accessing pre-selected leaf pages in a B-tree. The final pattern we tested is called skip-sequential, which accesses pages sequentially, but skips randomly over some pages. This pattern can be observed when a set of pages needs to be accessed, e.g., when upon reading scattered database pages the I/O requestsare performed using ascending/descending physical addresses. In this approach in comparison to the earlier approaches they mentioned the results interpretation like, Random access, unstable behavior, read/Write asymmetry, slower when full, Impact on queue depth, and energy consumption. In a conclusion of this paper, As the measurements clearly discover, each SSD exhibits a differing performance profile. We were able to identify some common patterns and outlined areas that are improving continuously, e.g., write performance. In this paper they have found problems with different Solid state drives as compared to each other. So different than previous approaches they have tried to focus on the problems they found in the solid state drive with their testing patterns. Also they have mentioned that nearly all the have ran discovered another fact for solid state drives and there is a lot more to detect. In this

Group 2

paper at the end they have proposed that hopefully the solid state drives will be able to put together the advantages of the hard disk drives combined with faster random access behavior. References: 1. Volker Hudlet, Daniel Schall SSD- An Empirical study to identify common properties and type-specific behavior Deparment of Computer Science, University of Kaiserslautern, Germany 2. Stephan Baumann, Giel de Nijs, Michel Strobel, and Kai-Uwe Sattler. Flashing Databases: Expectations and Limitations. In DaMon,2010.