Hm, what's up with the result formatting in DHTML?
Oh, I know. I have a much improved formatter in the debugger, when I
port that to LFC, I will fix measure to use that too. I also want
Debug.trace to use the meters from measure so that you can say
Debug.trace(foo, 'bar') and it will keep a running tally of the call
statistics. I guess I should file improvements for these ideas...
Finally, I do agree with Jim that these particular benchmarks are too
low level to measure performance. I wrote these mostly to test JS
implementations, to test optimizations the compiler could make, and
to guide us in how to write JS. The measure framework is meant to be
applicable to larger benchmarks, I just never wrote any.
On 2006-10-17, at 21:58 EDT, Benjamin Shine wrote:
Tell me this isn't the most fascinating data you've seen in a long
time. The executive summary: function calls and array access are
faster in dhtml than in flash, on my particular setup.
Rather than constructing reports with copy and paste, I think I
will bastardize Henry's logger.jsp. Run tests, calculate results,
post to server via a jsp. Then in a separate step we can analyze
the results.
http://localhost:8080/legals/test/performance/adding-elements-to-
array.lzx?lzr=swf7
iterations = 200
empty : 0.83us ± 0.02 [0.81..0.93]/100
Array.push : 3.51us ± 3.71 [2.50..27.40]/100
Array[Array.length] = : 2.02us ± 0.12 [1.93..3.02]/100
Array[index] = : 1.63us ± 0.06 [1.54..1.84]/100
Array[index] = (preallocated) : 1.67us ± 0.05 [1.59..1.82]/100
Object[key] = : 1.62us ± 0.88 [1.47..10.36]/100
http://localhost:8080/legals/test/performance/adding-elements-to-
array.lzx?lzr=dhtml
iterations = 200
empty : 0.65us ± 1.69 [0.00..5.00]/100
Array.push : 3.45us ± 2.32 [0.00..5.00]/100
Array[Array.length] = : 2.20us ± 2.49 [0.00..5.00]/100
Array[index] = : 2.40us ± 2.51 [0.00..5.00]/100
Array[index] = (preallocated) : 2.00us ± 2.46 [0.00..5.00]/100
Object[key] = : 1.90us ± 2.44 [0.00..5.00]/100
http://localhost:8080/legals/test/performance/cost-of-a-function-
call.lzx?lzr=swf7
iterations = 200
empty : 1.06us ± 0.04 [1.05..1.49]/100
assignment : 2.06us ± 0.22 [1.77..2.77]/100
function call : 8.73us ± 5.37 [7.32..35.08]/100
function call with 1 param : 8.62us ± 0.48 [8.21..10.29]/100
function call with 2 params : 10.45us ± 0.35 [9.93..11.94]/100
function call with 3 params : 9.65us ± 0.41 [9.23..11.16]/100
function call with 4 params : 14.01us ± 21.89 [9.70..141.19]/
100
method call : 11.64us ± 12.68 [8.78..137.09]/
100
method call with 1 param : 9.38us ± 0.37 [9.04..12.05]/100
proto method call : 10.66us ± 0.26 [10.35..11.35]/
100
proto method call with 1 param: 9.60us ± 0.22 [9.29..10.31]/100
http://localhost:8080/legals/test/performance/cost-of-a-function-
call.lzx?lzr=dhtml
iterations = 200
empty : 0.65us ± 1.69 [0.00..5.00]/100
assignment : 2.40us ± 2.51 [0.00..5.00]/100
function call : 2.65us ± 2.51 [0.00..5.00]/100
function call with 1 param : 2.45us ± 2.51 [0.00..5.00]/100
function call with 2 params : 3.05us ± 2.45 [0.00..5.00]/100
function call with 3 params : 2.55us ± 2.51 [0.00..5.00]/100
function call with 4 params : 2.70us ± 2.50 [0.00..5.00]/100
method call : 2.75us ± 2.50 [0.00..5.00]/100
method call with 1 param : 2.85us ± 2.49 [0.00..5.00]/100
proto method call : 2.75us ± 2.50 [0.00..5.00]/100
proto method call with 1 param: 3.15us ± 2.43 [0.00..5.00]/100
http://localhost:8080/legals/test/performance/iteration-forms.lzx?
lzr=swf7
iterations = 200
empty : 0.10us ± 0.03 [0.07..0.28]/100
for : 2.44us ± 0.23 [2.14..3.21]/100
for in : 2.06us ± 2.57 [1.62..25.35]/100
while : 2.14us ± 0.05 [2.11..2.44]/100
http://localhost:8080/legals/test/performance/iteration-forms.lzx?
lzr=dhtml
iterations = 200
empty : 0.05us ± 0.50 [0.00..5.00]/100
for : 1.95us ± 2.45 [0.00..5.00]/100
for in : 2.80us ± 2.49 [0.00..5.00]/100
while : 1.95us ± 2.45 [0.00..5.00]/100
On Oct 17, 2006, at 6:34 PM, Benjamin Shine wrote:
I just ran across test/performance and components/utils/performance.
This
http://localhost:8080/legals/test/performance/adding-elements-to-
array.lzx?debug=true
gives
empty : 0.87us ± 0.02 [0.84..0.95]/100
Array.push : 3.89us ± 1.39 [3.66..17.63]/
100
Array[Array.length] = : 3.47us ± 0.06 [3.40..3.70]/100
Array[index] = : 2.41us ± 0.04 [2.35..2.53]/100
Array[index] = (preallocated) : 2.46us ± 0.06 [2.38..2.80]/100
Object[key] = : 2.83us ± 5.14 [2.26..53.75]/
100
Seems to work in dhtml, too. Darn it's nice -- any reason
components/utils/performance/measure.lzx shouldn't be the basis
for our updated performance benchmarks?
We need a way to track the results in different versions &
runtimes of course. Someone's going to say "how about the wiki?"
in response to which I am going to shoot myself. XML checked into
the source tree would work better, I think... and if Josh Crowley
stays up late he could write a charting app to visualize the
performance results. (Not that Josh Crowley would ever stay up late.)
-ben
On Oct 17, 2006, at 4:06 PM, Max Carlson wrote:
Yes it does.
-Max
Henry Minsky wrote:
Does the getTimer() value that is sent in LzIdle events start
from zero when the LFC is loaded?
On 10/16/06, *Henry Minsky* < [EMAIL PROTECTED]
<mailto:[EMAIL PROTECTED]>> wrote:
At the bug scrub we talked about getting some updated
performance
benchmarks , are these assigned with some high priorty?
There's a number of proposed optimizations that are in JIRA
which
would be easier to attack if we had a simple set of metrics we
could run. A couple of issues involve app startup time, is
that
adequately covered by the existing inittimer tag ?
-- Henry Minsky
Software Architect
[EMAIL PROTECTED] <mailto:[EMAIL PROTECTED]>
--
Henry Minsky
Software Architect
[EMAIL PROTECTED] <mailto:[EMAIL PROTECTED]>
Benjamin Shine
Software Engineer, Open Laszlo / Laszlo Systems
[EMAIL PROTECTED]
Benjamin Shine
Software Engineer, Open Laszlo / Laszlo Systems
[EMAIL PROTECTED]
