Archives of the TeradataForum

Message Posted: Mon, 23 Feb 2004 @ 23:38:40 GMT

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

Way back in version 1 the only way to run EVL code was interpretive mode. That is, by executing pseudo instructions (virtual op-codes) on a stack-based virtual machine. In version two we compiled the pseudo code into machine code. This runs faster, but it makes the steps larger and if it gets too large we fall back to interpretive mode which doesn't included the complied code in the amp step and therefore tables less space.

By the way, you're probably asking, "What is EVL code?" about now. It's code the goes in the amp steps. The amps jump to this code when they have to do one of two things. The first is compare two rows. The second is build a row.

These are all segements gotten to store various things. Parser tree segments store tree nodes used in the syntaxer, resolver, and optimizer phases of the parser. Plastic steps are used in the final stage of the parser, the gen phase. This phase is dedicated to the job of transforming the "white tree" (output of the optimizer) into steps. The first kind of step is the plastic step. Plastic steps are missing using data and "real" spool numbers. They are the kind of steps we store in the steps cache. The second kind of step is the concrete step which is the kind of step that gets sent to the dispatcher which coordinates their execution on the amps. Concrete steps have using data and real spool numbers.

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