|
Abstract
SSDs have shown promise as an alternative to disk-based storage devices because of their better random read performance and low energy consumption. However, they process random writes much slower than reads because of a costly erase operation that needs to precede before rewriting to a page. The erase operation can only be applied to a bunch of pages instead of a single page.
In view of such fundamentally different IO behaviour by SSDs, recent efforts have been made to design index structures which optimize for their read/write asymmetry. Most of these indexing schemes were designed without considering concurrency control. In this work, we look at enabling concurrency control in flash-based indexing schemes and discuss some involved challenges.