[reportlab-users] Report generation time is not optimized - ideas?

[J. R. Carroll]

>

> I'd say your next step is to run various benchmarks and timed operations

> on your embedded machine versus your usual one, and see if it's 6x slower

> at everything.   Does your app depend on

> - database queries?

> - reading files from disk?

> - writing files (including PDF?) to disk?

> - pure processing ?



It _will_ do a database query eventually - haven't hooked that part in yet.
 But, it's this slow just from reading a test file from disk (JSON object).
It is writing the file to disk, but it's built in memory right now for the
purposes of streaming it through Twisted (webserver).
Pure processing?  What do you mean by this?  If you mean is it 6sec of pure
processing time (not including the writing to disk) - then from what I can
tell, yes it is taking my machine about 6 sec to write to memory (with no
output - just closing the buffer).

Thank you for the pystone suggestion, I was not familiar with that.  I will
use that in the future.

Results of pystone are:

Local = 125000
Target = 8787

Which if I am understanding how pystone works then my local dev environment
should perform about 14x (rounded down) faster than the OE device.  (There
wasn't much documentation available on the web, only a few "bug reports"
from PSF suggesting that pystone might be outmoded now especially with
multithreading benchmark suites - so I am not sure if this is a valid
benchmark test?)

This was run using linux's "time".  I tested how long it took to write to
memory using one my scripts (didn't flush the buffer to disk).

Target/OE device

Real

6.42

User

5.68

Sys

0.15

Local/Dev PC

Real

.860

User

.436

Sys

.040


Expected results:

Real .860*14 = 12.04

User .436*14 = 6.104

Sys .040*14 = 0.56


Unless I screwed up my math, based on benchmarking, the target device is
actually performing better than expected -- across the board, as it were?

Next step, as I have indicated off the list, is to do some profiling now
that we have confirmed the accelerator is installed and is accessible.
 We'll optimize where possible and hopefully get those times down even
lower.  If anyone has additional thoughts on where to focus attention, I'm
all ears.

Thanks again everyone - your feedback is greatly appreciated!

-J


----


J. R. Carroll
Independent Researcher through Hurtz Labs
Research Methods, Test Development, and Statistics
www.jrcresearch.net
www.ontvp.com
Cell:  (650) 776-6613
Email: jrcarroll at jrcresearch.net
          jrcarroll at hurtzlab.com
          jrc.csus at gmail.com
<https://www.facebook.com/J.R.Car>
<https://twitter.com/jNammer><http://www.linkedin.com/in/jrcarroll>



On Thu, Jan 3, 2013 at 10:50 AM, Andy Robinson <andy at reportlab.com> wrote:


> I'd say your next step is to run various benchmarks and timed operations

> on your embedded machine versus your usual one, and see if it's 6x slower

> at everything.   Does your app depend on

> - database queries?

> - reading files from disk?

> - writing files (including PDF?) to disk?

> - pure processing ?

>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://two.pairlist.net/pipermail/reportlab-users/attachments/20130103/ad6e8817/attachment.htm>