Archives of the TeradataForum

Message Posted: Fri, 09 Jan 2004 @ 16:19:13 GMT

	<Prev	Next>		<<First	<Prev	Next>	Last>>

In general you want the cache to hold blocks from small, commonly referenced tables.

Teradata will not use the basic "Least Recently Used" cache algorithm when scanning this table, because it is large enough that doing so would likely flush out many other blocks that are more likely to be re-used.

SyncScan allows a scan request to join an equivalent scan of the same table "already in progress". Some amount of cache space is used to hold blocks temporarily in case another scan can use them. As long as the requests are processing rows at roughly the same rate, the physical I/O is only done once. When the first scan finishes, the next will "wrap back" and read whatever part of the table it missed, and so forth.

Since Teradata knows the spool file contents are going to be used in at least one later step, it would improve performance to keep the data in cache. But because the spool file is only going to be used by this one request, it is worth caching only if it is very small.

	<Prev	Next>		<<First	<Prev	Next>	Last>>