[AVRO-316] Optiminzing inner loop functions of Avro io

• Type: Improvement
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 1.3.0
• Component/s: java
• Hadoop Flags: Reviewed
• Assignee: Thiruvalluvan M. G.
• Reporter: Thiruvalluvan M. G.
• Created: 13/Jan/10 12:41
• Updated: 01/Mar/10 17:09
• Resolved: 15/Jan/10 17:20

Description

The methods advance() and pushProduction() of org.apache.avro.io.parsing.Parser class are invoked at least once per read/write call on ValidatingDecoder, ResolvingDecoder and ValidatingEncoder. Any optimization in these functions will improve overall performance when using these classes.

Try org.apacge.avro.io.Perf -V to see the effect of this patch on the performance. On my machine it gives about 5 to 8% improvement.

The optimizations are:

• Reorder the code within advance() so that the most frequent paths involve minimal comparisons.
• Repleace Symbol.CONTINUE with null to indicate that next symbol on the stack needs to be looked at.
• Get rid of a parameter to pushProduction. This parameter was used only in one call site of that function; others pass null. Refactored the code so that these invocations do not pay for that piece of logic that they don't use.