>
> Bruce Momjian <maillist@candle.pha.pa.us> writes:
> > Now that I believe the Mariposa code is good
>
> well...
>
> > best way to perform sort in memory (leftist sort vs. qsort)
>
> replacement/selection (e.g., the leftist tree thingy) is the usual
> technique for run generation. it can greatly reduce the number of
> passes required to sort. the main downside is complexity. there's
> still ongoing research on how to do r/s memory management
> efficiently. there are also memory system effects: qsort is
> more cache-friendly than r/s if implemented right.
>
> i'd just worry about the first-order effects for now. the extra
> disk copies (temp relations) are bad - fix that. the other stuff
> is probably noise.
This is very helpful. Thanks.
I will concentrate on intergrating the Mariposa code ( to eliminate the
temp tables) and prevent the seven tape files from being generated for
small sorts.
>
> some useful online papers:
>
> http://www.research.microsoft.com/research/barc/Gray/AlphaSort.doc
> ftp://ftp.cs.pdx.edu/pub/faculty/graefe/papers/sort.ps
> --
> Paul M. Aoki | University of California at Berkeley
> aoki@CS.Berkeley.EDU | Dept. of EECS, Computer Science Division #1776
> | Berkeley, CA 94720-1776
>
- --
Bruce Momjian
maillist@candle.pha.pa.us
------------------------------
End of hackers-digest V1 #399
*****************************
