Large writes in Exadata Flashcache

In the Exadata storage management software 12.2.1.1.0, Flashcache began to support complex sequencing and a large number of hash joins to produce temporary data written to Flashcache, rather than directly to the SAS disk Tempfile, For subsequent reading of these temporary data from Flashcache, this flashcache enhancement will increase the performance of some special SQL statements by more than four times times (note: This feature must be based on the flashcache of writeback mode).

In fact, not only the temporary data of the hash join can be written to Flashcache, the large write operation supports write Flashcache, so-called "large write" refers to the following three major types:

(1). writes that is read once (read once writes)

For example: the brokered data generated by a sort or hash join.

(2). Writes that is unlikely to be read (never read writes)

For example: Write archive log, flashbacklog log, etc.

(3). Writes that is maybe read (maybe read writes)

Example: Index rebuild, direct path data loading, etc.

Although large write operations are allowed to use flashcache resources directly, the precondition is that these "large writes" Do not allow the impact of the current workload. For example, you need to prevent an increase in the average IO latency of the database by preventing the impact of an OLTP hit rate.

In addition, "large write" has the following limitations:

(1). Large writes can use 20% of total Flash Cache size (default)

(2). Prevent interference with DW caching (default 50% of total Flash Cache size)

(3). Non-temp Large writes only cached if disk is busy

Large writes in Exadata Flashcache