Our test tag was using a HashMap.
You can view this performance quirk yourself using Struts bean:message tag.
To get orders of magnitude in the performance discrepancy (to make sure we
were measuring the tag performance difference on a much more significant
scale than server access variance times) we went to 200,000 tag exectutions
done a variety of ways:
One of our cases was a page with 20 tags, inside a loop 10,000 times.
Another case was same page, but done by 160 tags inside a 1250 loop.
Another case same page rendered by includes: page included another page 8
times inside a 1250 loop. The included page had 20 tags.
There were other varieties, but these three show the discrepancy in
performance with the increased number of tags very significantly. All
three cases output the same resulting html. All three cases executed the
tag code 200,000 times. The first and third cases, with 20 custom tag
occurrences, performed roughly the same (better performance). The second
case with 160 tags performed much much worse. I don't have our statistics
anymore, alas, but it was two or more orders of magnitude worse. We tried
an 80 tag variety as well, and it put the expected point on the slope
(significantly worse than same output rendered via 20 tags, not as bad as
160 tags).
Switching to IBM VM, or turning profiling flags on in the Sun VM before
running the tests, caused the performance discrepancy between these
approaches to vanish (roughly the better performance mark for all three).
Be wonderful if someone else could verify on their own. Two days of
cursing over it was enough for us ;-). As soon as we realized we could fix
the problem pages by restructuring them into separate JSPs, we declared
victory over the Mystery of the VM ;-).
Again this was 1.3, who knows if Sun's 1.4 vm still does this....
Jimbo
Build a simple JSP that contains
"Hajratwala,
Nayan (N.)" To: "'Struts Users Mailing List'"
<[EMAIL PROTECTED]>
<[EMAIL PROTECTED] cc:
m> Subject: RE: Improving performance by
splitting JSP?
09/10/2002 01:02
PM
Please respond to
"Struts Users
Mailing List"
Was it a Hashtable or HashMap?
It might be that since Hashtable is synchronized you were encountering a
bottleneck in your multithreaded environment, whereas just a regular loop
would not have this problem...
---
- Nayan Hajratwala
- Chikli Consulting LLC
- http://www.chikli.com
-----Original Message-----
From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED]]
Sent: Tuesday, September 10, 2002 11:48 AM
To: Struts Users Mailing List
Subject: RE: Improving performance by splitting JSP?
I had to go back and look at the test code we used, but yes - looks like we
did try a non-body tag. It did a hashtable lookup for whatever key values
was passed in as a parameter. I remember we found that we could only
recreate the performance discrepancy with tags that did some kind of memory
lookup; just adding loops around arithmetic statements are writing output
were tried, but did not exhibit the symptoms.
Jim Weaver
Software Developer - ThoughtWorks
"Martin Cooper"
<martin.cooper@tumb To: "'Struts Users
Mailing List'" <[EMAIL PROTECTED]>
leweed.com> cc:
Subject: RE: Improving
performance by splitting JSP?
09/09/2002 05:24 PM
Please respond to
"Struts Users
Mailing List"
> -----Original Message-----
> From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED]]
> Sent: Monday, September 09, 2002 2:51 PM
> To: Struts Users Mailing List
> Subject: RE: Improving performance by splitting JSP?
>
>
>
> Yup, we looked at overall size of the JSP as well, and the
> association with
> performance was definitely number of bm/bw tags within a
> single JSP rather
> than overall JSP size. We even tried editing the generated
> servlet code
> and adding big unused methods to see if the problem had to do with
> generated servlet file size. That would kind've made sense, but was a
> negative. I believe we also tried a tag that just did a
> sysout rather than
> any kind of memory lookup (hashtable or properties file) and
> found that
> this performance quirk in the sun vm did not appear in that case.
When you wrote your own tags (the message/write equivalents, I mean), did
you happen to play with body versus non-body tags? When I was wrestling
with
the "too many tags" problem a while ago, I ended up writing my own versions
of some of the Struts tags so that they were non-body tags, because they
caused noticeably less code to be generated. I didn't measure performance
differences, but I wouldn't be surprised to see a noticeable improvement
there too.
--
Martin Cooper
>
> We could never pin down a why, it seemed that the sun 1.3 vm
> just ran like
> a snail with a lot of bm or bw tags in a single page, so we
> stopped doing
> that ;-). Same number of tags split up into multiple JSPs,
> or a few tags
> called the same number of times via a loop - OK performance.
>
> It is very true also that splitting them up into smaller
> bites makes them
> more readable and maintainable, so it was a good solution all
> around. It's
> just that you are always nervous when you fix a problem and
> don't know the
> why of it ;-).
>
> Jimbo
>
>
>
> --
> To unsubscribe, e-mail:
> <mailto:[EMAIL PROTECTED]>
> For additional commands, e-mail:
> <mailto:[EMAIL PROTECTED]>
>
>
--
To unsubscribe, e-mail: <
mailto:[EMAIL PROTECTED]>
For additional commands, e-mail: <
mailto:[EMAIL PROTECTED]>
--
To unsubscribe, e-mail:
<mailto:[EMAIL PROTECTED]>
For additional commands, e-mail:
<mailto:[EMAIL PROTECTED]>
--
To unsubscribe, e-mail: <
mailto:[EMAIL PROTECTED]>
For additional commands, e-mail: <
mailto:[EMAIL PROTECTED]>
--
To unsubscribe, e-mail: <mailto:[EMAIL PROTECTED]>
For additional commands, e-mail: <mailto:[EMAIL PROTECTED]>
