I have a question on DateTime performance
I really want to use DateTime in most of my projects, especially when
I'm working with the likes of Class::DBI, to give the proper abstraction
for handling TIMESTAMP/DATETIME fields, as well as the free validation I
get from DateTime.
but DateTime's performance really bugs me. This is especially true when
I'm trying to use these modules in web applications.
As an example, Miyagawa-san had pointed out in his blog:
http://blog.bulknews.net/mt/archives/000367.html
Here's my version of the above benchmark
[timethese]
Benchmark: timing 100000 iterations of datetime, posix...
datetime: 139 wallclock secs (138.96 usr + 0.15 sys = 139.11 CPU) @
718.86/s (n=100000)
posix: 0 wallclock secs ( 1.53 usr + 0.09 sys = 1.62 CPU) @
61728.40/s (n=100000)
[cmpthese]
Rate datetime posix
datetime 719/s -- -99%
posix 62112/s 8533% --
Ouch.
I profiled this a long time ago, and Params::Validate seemed like a
problem. Anyway, here's fresh profile:
[EMAIL PROTECTED]:~$ ~/test/bin/dprofpp tmon.out
Total Elapsed Time = 11.44324 Seconds
User+System Time = 11.42324 Seconds
Exclusive Times
%Time ExclSec CumulS #Calls sec/call Csec/c Name
34.6 3.959 4.389 60277 0.0001 0.0001 Params::Validate::_validate
12.3 1.410 7.609 20000 0.0001 0.0004 DateTime::new
7.53 0.860 8.129 10000 0.0001 0.0008 DateTime::from_epoch
6.92 0.790 0.940 60000 0.0000 0.0000
DateTime::TimeZone::_spans_binary_
search
6.57 0.750 1.780 10000 0.0001 0.0002 DateTime::strftime
6.30 0.720 4.880 10000 0.0001 0.0005 DateTime::set
6.04 0.690 1.640 20002 0.0000 0.0001 DateTime::TimeZone::new
5.25 0.600 1.460 190000 0.0000 0.0000 DateTime::__ANON__
4.81 0.550 0.630 30002 0.0000 0.0000
DateTime::_calc_local_components
3.59 0.410 1.650 60000 0.0000 0.0000
DateTime::TimeZone::_span_for_date
time
3.15 0.360 1.770 30002 0.0000 0.0001 DateTime::_calc_local_rd
3.06 0.350 1.150 30000 0.0000 0.0000
DateTime::_handle_offset_modifier
2.36 0.270 0.280 20001 0.0000 0.0000 DateTime::Locale::load
2.28 0.260 7.049 10000 0.0000 0.0007
DateTime::Format::Mail::format_dat
etime
1.84 0.210 0.210 60000 0.0000 0.0000 DateTime::TimeZone::max_span
Now let me try to profile this.
Ouch. Hmm, let me try this with PERL_NO_VALIDATION = 1
[timethese]
Benchmark: timing 100000 iterations of datetime, posix...
datetime: 136 wallclock secs (135.06 usr + 0.12 sys = 135.18 CPU) @
739.75/s (n=100000)
posix: 2 wallclock secs ( 1.48 usr + 0.10 sys = 1.58 CPU) @
63291.14/s (n=100000)
[cmpthese]
Rate datetime posix
datetime 740/s -- -99%
posix 63694/s 8513% --
Argh. No luck.
I really really want to make it faster. I know that it's impossible to
match sprintf + localtime, but wouldn't it be nice if the difference was
an order of magnitude lower... ;)
So ... are we going to address this?
--d
