Subject: Re: Re: [shell-discuss] More ksh93 builtins [PSARC/2010/095 FastTrack
timeout 03/25/2010]
--------
The focus of this discussion so far has been about the performance
advantage of the AST tools because they can be run as built-ins.
However, the AST toolkit has advantages over the GNU toolkit with respect
to performance when processing large amounts of data for which that
time to start a command is negligble.
Here is an example of sorting 20 million randomly generated 100 byte
records.
================AST TOOLS=====================
$ time /usr/ast/bin/wc rec2000
20000000 42066033 2000000000 rec2000
real 0m15.55s
user 0m4.45s
sys 0m1.91s
$ time /usr/ast/bin/sort rec2000 > sort1
real 3m41.59s
user 0m34.98s
sys 0m30.20s
================end AST TOOLS=====================
================GNU TOOLS=====================
$ time /usr/gnu/bin/wc rec2000
20000000 42066033 2000000000 rec2000
real 0m35.10s
user 0m31.03s
sys 0m1.08s
$ time /usr/gnu/bin/sort rec2000 > sort2
real 3m39.65s
user 1m11.98s
sys 0m21.59s
================end GNU TOOLS=====================
The results were identical.
As you can see, wc is about 7 times faster and sort is about twice as
fast. Moreover, the AST sort has several extensions found on mainframe
sorts, but not found on UNIX sorts.
David Korn
dgk at research.att.com
