[
https://issues.apache.org/jira/browse/HADOOP-3442?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Chris Douglas updated HADOOP-3442:
----------------------------------
Attachment: 3442-4.patch
bq. I guess it is safe to choose a larger k. For example, with k = 8, the stack
depth is bounded by 250 in practice, but a lot fewer cases of input data
patterns will need to invoke the heapsort.
bq. I think the 2*log n heuristic not only bail out of worst cases but also
good cases since it is overly strict.
The IntroSort [paper|http://citeseer.ist.psu.edu/musser97introspective.html]
suggested that 2 * floor(log\(n)) produced good results. Given that the most
degenerate cases cascade, deferring the change to heapsort may effect longer
running times. Still, it does seem low. I ran some informal benchmarks with a
random set of elements in nearly sorted, sawtooth, and pipe organ patterns and
using k = 2 switched to heapsort pretty aggressively, so I increased k to 4.
The average running times were all within a few hundred milliseconds, though;
we probably don't want to get too carried away optimizing and tweaking an
operation typically accounting for no more than a few seconds out of every
spill.
> QuickSort may get into unbounded recursion
> ------------------------------------------
>
> Key: HADOOP-3442
> URL: https://issues.apache.org/jira/browse/HADOOP-3442
> Project: Hadoop Core
> Issue Type: Bug
> Components: mapred
> Affects Versions: 0.17.0
> Reporter: Runping Qi
> Assignee: Chris Douglas
> Priority: Blocker
> Fix For: 0.17.1, 0.18.0
>
> Attachments: 3442-0.patch, 3442-0v17.patch, 3442-1.patch,
> 3442-2.patch, 3442-3.patch, 3442-4.patch, CheckSortBuffer.java,
> HADOOP-3442.patch, overflow.zip, spillbuffers.patch
>
>
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
