FTL with Flash-based Database Storage

4 Jul

In recent years, the development of the disk has reached a bottleneck, and in the various computer systems in the process of flash memory instead of the disk faster. Flash memory is a magnetic hard disc has the completely different nature of the new read-write storage medium, the first electrical device of flash memory is, no mechanical seek to be positioned relative to the disk can be provided a high random access performance: Second, because the flash memory chip before overwriting relatively large area must be erased, flash write operation and a read operation is very slow compared to having to read and write asymmetry. Directly to the traditional database system set up in the flash memory can not give full play to the optimal performance, the reason is, the database storage engine associated with the external memory data structures and algorithms are based on the performance characteristics of the disk as a starting point, such as the page is excluded from the memory unconditional write-back, read a lot of data to give priority to the needle orderly movement. These structures and algorithms are no longer suitable for flash memory, database systems can easily become a bottleneck. Therefore, the flash-based database storage research is imperative. FTL Flash Translation Layer can be applied, in this context, this paper to the specific performance characteristics of flash memory and database access logic, based on a more complete study of the flash-based database storage engine technology, has made some achievements. Specifically include:

 This paper presents a complete flash database storage engine framework called CRL (Compressed, Redo-based version control, Later-grouped materialize), the physical characteristics of flash memory for a series of specialized storage and transaction level design, including support partial decompression compression algorithm, based on the version control protocol redo logs and delay materialized into groups, etc., which greatly improved the database in an online transaction processing applications performance.

This paper presents an application for online transactional index tree structure called a UM-B + tree (Update Migration B + Tree), update records in the index by migrating between layers, thereby reducing the number of random writes the index, the index increased IO performance. Meanwhile this paper for the first time UM-B + tree extend to transactional engine, discussed in a highly concurrent environments access control and recovery mechanisms, increasing the UM-B + tree availability.

This paper presents an application for online analytical massive data parallel scan scheduling framework S ~ 3 (Semi-Share Scan). S ~ 3 uses a new semi-shared scan mode, through a set of shared reading to reduce bandwidth consumption while reading by compensating for the scan faster acceleration, better use of the I / O bandwidth and CPU resources.


Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

%d bloggers like this: