Re: Speed of HSSF

Isn't the real issue that the RPG to Java interface (JNI) does not
implement very well with HSSF? I'd like to defend the i, and say that it
does okay running Java, but the interface from RPG to Java performs poorly
over thousands of calls. I'm not arguing with Scott here. If we could
either write the spreadsheet in pure RPG as he is, or in pure Java, I think
the performance would be much better.




From: Scott Klement <rpg400-l@xxxxxxxxxxxxxxxx>
To: RPG programming on the IBM i / System i
<rpg400-l@xxxxxxxxxxxx>,
Date: 08/10/2012 10:38 PM
Subject: Re: Speed of HSSF
Sent by: rpg400-l-bounces@xxxxxxxxxxxx



Hiya John,

Here's what kills me: Imagine how long it takes to write the same file
to disk in Excel. How would you react if Excel took 15 minutes to write
the file?

It's hard to even imagine that, isn't it?

That was a big part of my incentive for rewriting mine in pure RPG. And
mine certainly isn't done (though, for someone who understands the
internals as I do, it's usable) but the difference between 30 seconds
and 30 minutes is an incredible improvement.

-SK



On 8/9/2012 6:49 PM, John Yeung wrote:

I hope you'll forgive me for hijacking an old thread, but I have
always wondered how long it actually takes to create Excel files with
HSSF compared to xlwt (using iSeries Python). I have heard some
people say Java is quite slow on the i, especially older ones; and of
course Python has a reputation for being rather slow as well. Which
is slower on the i?

On Thu, Jun 28, 2012 at 1:16 PM, <sjones@xxxxxxxxxxxxxxx> wrote:

The spread sheet has 25 columns & contains 37249 rows.

On Thu, Jun 28, 2012 at 2:56 PM, <sjones@xxxxxxxxxxxxxxx> wrote:

It runs for about 20 minutes - would love for it to run quicker but

don't

have any ideas on how to achieve that.

On Thu, Jun 28, 2012 at 3:02 PM, Scott Klement
<rpg400-l@xxxxxxxxxxxxxxxx> wrote:

You must have a faster computer than I do :-) My biggest spreadsheet is
14000 rows and 18 columns, and it takes about 30 minutes. And that's
using SXSSF (lower memory footprint of XSSF)

The reason I'm dredging this up now is that I've just generated a
spreadsheet with 8131 rows and 97 columns on a model 810 (processor
feature 7409) running V5R2 and iSeries Python 2.3.3, and it took about
15 minutes on a quiet (somewhat after-hours) system. The resulting
file is about 6 MB.

Very, very roughly speaking (as I have no idea what kind of hardware
the others were using), I'd say this is comparable to and quite
possibly better than HSSF. So I'm certainly feeling very little
incentive to switch from xlwt to HSSF.

Of course, later in that thread:

On Sun, Jul 1, 2012 at 1:33 PM, Scott Klement <rpg400-l@xxxxxxxxxxxxxxxx>

wrote:

I've been working on a service program that works very similarly to
HSSF, but doesn't call Java routines at all. Instead, builds an XLSX
file (the native Excel format in Excel 2007 and up) by building the
proper directory structure and the various XML files inside that
structure, then zipping the whole thing.

I have a ways to go -- but for the document I mentioned earlier (the one
that took 30 minutes with POI's XSSF) it only takes 30 seconds to build.
That's a 60-fold improvement in performance.

Yet again, RPG (I'm assuming he means his new program is completely
done in RPG) proves to be a great performer on the i.

John

--
This is the RPG programming on the IBM i / System i (RPG400-L) mailing list
To post a message email: RPG400-L@xxxxxxxxxxxx
To subscribe, unsubscribe, or change list options,
visit: http://lists.midrange.com/mailman/listinfo/rpg400-l
or email: RPG400-L-request@xxxxxxxxxxxx
Before posting, please take a moment to review the archives
at http://archive.midrange.com/rpg400-l.