long hadoop run, stuck at reduce>reduce - hadoop

Question

I have hadoop run that basically just aggregate over keys, it's code:
(mapper is the identity mapper)
public void reduce(Text key, Iterator<Text> values,
OutputCollector<Text, Text> results, Reporter reporter) throws IOException {
String res = new String("");
while(values.hasNext())
{
res += values.next().toString();
}
Text outputValue = new Text("<all><id>"+key.toString()+"</id>"+res+"</all>");
results.collect(key, outputValue);
}
it stuck at this level:
12/11/26 06:19:23 INFO mapred.JobClient: Running job: job_201210240845_0099
12/11/26 06:19:24 INFO mapred.JobClient: map 0% reduce 0%
12/11/26 06:19:37 INFO mapred.JobClient: map 20% reduce 0%
12/11/26 06:19:40 INFO mapred.JobClient: map 80% reduce 0%
12/11/26 06:19:41 INFO mapred.JobClient: map 100% reduce 0%
12/11/26 06:19:46 INFO mapred.JobClient: map 100% reduce 6%
12/11/26 06:19:55 INFO mapred.JobClient: map 100% reduce 66%
I run it locally and saw this:
12/11/26 06:06:48 INFO mapred.LocalJobRunner:
12/11/26 06:06:48 INFO mapred.Merger: Merging 5 sorted segments
12/11/26 06:06:48 INFO mapred.Merger: Down to the last merge-pass, with 5 segments left of total size: 82159206 bytes
12/11/26 06:06:48 INFO mapred.LocalJobRunner:
12/11/26 06:06:54 INFO mapred.LocalJobRunner: reduce > reduce
12/11/26 06:06:55 INFO mapred.JobClient: map 100% reduce 66%
12/11/26 06:06:57 INFO mapred.LocalJobRunner: reduce > reduce
12/11/26 06:07:00 INFO mapred.LocalJobRunner: reduce > reduce
12/11/26 06:07:03 INFO mapred.LocalJobRunner: reduce > reduce
...
a lot of reduce > reduce ...
...
in the end , it finished the work. I want to ask:
1) what does it do in this reduce > reduce stage?
2) how can i improve this?

Answer 1

When looking at the percentages, 0-33% is shuffle, 34%-65% is sort, 66%-100% is the actual reduce function.
Everything looks fine in your code, but I'll take a stab in the dark:
You are creating and re-recreating the string res over and over. Every time you get a new value, Java is creating a new string object, then creating another string object to hold the concatenation. As you can see, this can get out of hand when the string gets pretty big. Try using a StringBuffer instead. Edit: StringBuilder is better than StringBuffer.
Whether or not this is the problem, you should change this to improve performance.

Answer 2

Using StringBuilder solves it. It improves the run time from 30 min to 30 sec. I didn't think it would make such a difference. Thanks a lot.