struts-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Pierre Thibaudeau" <pierre.thibaud...@gmail.com>
Subject Re: Performance issue
Date Tue, 06 Jan 2009 21:19:36 GMT
Yes the model is *exactly* the same in the two cases.  In fact, both
S1 and S2 exist concurrently in the same Tomcat application.

Each test is run using JMeter, with 10 users who randomly try to
access one of 42 different pages involving 10 different types of
results (that is, roughly speaking, 10 different JSPs, abstracting
from the fact that each page is really composed of 6 or 7 different
JSP tiles).  In total, roughly 1,600 pages were queried for each test.
 Moreover, each test was performed after re-starting Tomcat, once the
pages had been warmed up.

My last post in this thread was fairly extreme, and I felt the need to
make new tests that relied on fewer assumptions and more on hard
facts.  So I modified the pages tested as best I could, in order to
make the respective S1 pages and S2 pages as close to identical as
possible.  In particular, this meant beefing up my old S1 pages,
adding to them some of the more recent sections that existed only in
the S2 version:  mostly (but not exclusively) a lot of javascrip/dojo
stuff, which is not interpreted anyway by JMeter. (I don't rely on S2
for dojo, but provide my own; and yes, I package and shrinksafe it).

Since my original timing for S1 pages were so ridiculously low (often
under 10 ms!!!), the new beefed-up S1 pages were bound to be
significantly slower.
On average, with the beefed-up pages, S1 pages take 74 ms to load.

Since last running the S2 tests (about 4 days ago), I did a fair
amount of optimizing and I replaced the most obvious <s:tags> by
<c:ones> in about half of the JSPs. I therefore re-ran the tests just
before writing this current post.
There is a marked improvement and, on average, S2 pages load in 169 ms.

In other words, I may have been a little (!) hasty in posting my
previous statement! (*blushing*)  My apologies to all the people who
care about S2!!!
We are happily very far from the apocalyptic 30x factor that I had
previously observed.  (Some tests last week were showing averages of
580 ms per S2-page!)

As I am under pressure, I may not have time immediately to set up a
new test to compare the most recent result with the case of S2 pages
that entirely rely on <s:tags>, though that would be very
enlightening.  If I can get round to it, I promise to post the
results!

---------------------------------------------------------------------
To unsubscribe, e-mail: user-unsubscribe@struts.apache.org
For additional commands, e-mail: user-help@struts.apache.org


Mime
View raw message