What is the problem the authors are trying to solve?
The authors are interested in determining, and providing, a variety of design choices for SSDs (solid-state disks) that optimize performance. Up until this paper, there were few publications detailing the designs of SSD hardware.
What other approaches or solutions existed at the time that this work was done?
The authors claim that existing knowledge of SSD design primarily existed as intellectual property of various vendors. Thus, SSD design was not well understood to the community, and other vendors interested in designing their own SSDs were required to derive their own understanding of NAND-based SSDs and how they were architected.
What was wrong with the other approaches or solutions?
Previous SSD designs showed issues that previously occurred at a higher level of abstraction–above HDDs (hard drive disks). Worse than the lack of performance found in existing SSD designs was that information regarding SSD designs were not public.
What is the authors’ approach or solution?
The authors of this paper believed that discussion of various design tradeoffs would allow for future SSD designs to be guided in better directions. The design tradeoffs discussed were then analyzed using a simulator which was based on a simulator produced by a research lab at CMU. So, not only were the analyses for proposed tradeoffs made available, but the simulator itself could be re-used by other groups.
Why is it better than the other approaches or solutions?
The author’s contributions were better than previous SSD designs because they accommodated information from software designs for SSDs in their simulator and in the discussion of SSD design tradeoffs. Then, after collecting actual traces they used the simulator and their expertise to design better SSD architectures.
How does it perform?
In general, it does not seem like the author’s detailed design tradeoffs are not compared to any work from other research groups or projects. Instead, the variations of each design and their inclusion and exclusion are benchmarked. For some things there are clear benefits–copy-back, interleaving plane-pairs, greedy + rate-limiting + migration wear-leveling algorithm. Meanwhile, some designs seem to have minimal tradeoffs between the variations or have minimal improvement over not including the design.
Why is this work important?
This work is important because it provides a survey-like breadth of SSD design choices. Even if these particular design choices may not be desirable, the paper certainly paints a quality picture of how to benchmark and analyze future design choices.