There's one study that shows that if users are able to complete their task (e.g. make a reservation) they perceive the site as faster, even though the roundtrip time may be slower than the competition :)

But then there's Google maps that got 20% more traffic when they improved the loading time by 20%

--

Stoyan Stefanov

http://phpied.com

On Feb 21, 2009, at 11:40 AM, "beinwal" <beinwal@gmail.com> wrote:

Will they also be differentiating between perceived and actual
response times?

This is another grey area along with the impact of performance to
overall customer satisfaction.

We just redesigned our product pages and they have increased in size
(a.k.a. slower), but the time where the page begins to render is much
improved. I am hoping that helps to balance it out, but it is too
soon to tell.

--- In exceptional-performance@yahoogroups.com, Steve Souders
<steve@...> wrote:
>
> This year's Velocity conference will have at least two talks from
big
> (alexa top 100) websites that quantify the impact of page load times
to
> revenue.
>
> -Steve
>
> On 2/19/2009 1:07 PM, Peter Booth wrote:
> >
> > That's what got me thinking about A/B testing of performance
changes.
> >
> >
> > I was talking with a potential client two weeks ago and had an
"Aha"
> > moment.
> > I realized that this guy really didn't want to hire me by the
hour.
> > He wanted to buy the results of performance engineering.
> >
> > If I could have told him
> >
> > "For X thousand dollars I can improve your typical Los Angeles 1.5
> > MB/s DSL user's mean response time from seven sec to five sec"
> >
> > then he would have been delighted.
> >
> > If I could go further and say "This should increase your
conversion
> > rate from 17% to 21%"
> > then he could easily see whether the work was worthwhile.
> >
> >
> >
> > On Feb 19, 2009, at 3:47 PM, Ernest Mueller wrote:
> >
> >> Yeah, this is hard. We have a complex site and so when we do
> >> optimizations, it's usually around the same time as 50 different
Web app
> >> releases, all of which not only affect performance but also
ideally are
> >> targeted at driving conversion themselves.
> >>
> >> Ernest
> >> ______________________
> >> UN-altered REPRODUCTION and DISSEMINATION of
> >> this IMPORTANT information is ENCOURAGED.
> >>
> >> From: Peter Booth <peter_booth@...
<mailto:peter_booth%40mac.com>>
> >>
> >> To: exceptional-performance@yahoogroups.com
> >> <mailto:exceptional-performance%40yahoogroups.com>,
yslow@yahoogroups.com
> >> <mailto:yslow%40yahoogroups.com>
> >>
> >> Date: 02/19/2009 02:43 PM
> >>
> >> Subject: [exceptional-performance] Show me the numbers. How has
> >> performance improvement impacted conversion rate?
> >>
> >> Sent by: exceptional-performance@yahoogroups.com
> >> <mailto:exceptional-performance%40yahoogroups.com>
> >>
> >>
> >> I take it on faith that, all things being equal, a faster website
> >> will hold
> >> peoples attention longer, and consequently have higher conversion
> >> rates. I just
> >> moved a domain from register.com to namecheap.com because I was
> >> frustrated by
> >> register's slow admin application.
> >>
> >>
> >>
> >>
> >> I've also seen two quotes from Google and Amazon that are
frequently
> >> repeated, but
> >> never with specific details connecting them to specific tests or
> >> projects . I
> >> don't have any personal experience of seeing conversion rates
change
> >> as a result
> >> of changes in performance.
> >>
> >> Has anyone in this group seen this, either as A/B testing or
> >> before/after?
> >>
> >> Can anyone quote, or even hint, at profit/sales/revenue/signups
rates
> >> correlated
> >> with performance?
> >>
> >> I'm contemplating finding a local online store and doing a
project,
> >> even as a
> >> freebie, to get some data that I can use for marketing.
> >>
> >> Peter
> >>
> >>
> >>
> >> <pic12382.gif><pic17421.gif>
> >
> >
>

When launched, Google Maps was pretty slow, but a speed improvement brought a lot more traffic. "When the Google Maps home page was put on a diet, shrunk from 100K to about 70K to 80K, traffic was up 10 percent the first week and in the following three weeks, 25 percent more."

 

Stoyan,

Do you have a reference to the Google datapoint?

Peter Booth
(917) 445 5663 peter_booth@... <mailto:peter_booth@...>

On Feb 22, 2009, at 3:03 PM, Stoyan Stefanov <stoyan@...> wrote:

 

There's one study that shows that if users are able to complete their task (e.g. make a reservation) they perceive the site as faster, even though the roundtrip time may be slower than the competition :)

But then there's Google maps that got 20% more traffic when they improved the loading time by 20%

--
Stoyan Stefanov
http://phpied.com <http://phpied>

On Feb 21, 2009, at 11:40 AM, "beinwal" <beinwal@...> wrote:

 

Will they also be differentiating between perceived and actual
response times?  

This is another grey area along with the impact of performance to
overall customer satisfaction.  

We just redesigned our product pages and they have increased in size
(a.k.a. slower), but the time where the page begins to render is much
improved.  I am hoping that helps to balance it out, but it is too
soon to tell.

--- In exceptional-performance@yahoogroups.com <mailto:exceptional-performance%40yahoogroups.com> , Steve Souders
<steve@...> wrote:
>
> This year's Velocity conference will have at least two talks from
big
> (alexa top 100) websites that quantify the impact of page load times
to
> revenue.
>
> -Steve
>
> On 2/19/2009 1:07 PM, Peter Booth wrote:
> >
> > That's what got me thinking about A/B testing of performance
changes.
> >
> >
> > I was talking with a potential client two weeks ago and had an
"Aha"
> > moment.
> > I realized that this guy really didn't want to hire me by the
hour.
> > He wanted to buy the results of performance engineering.
> >
> > If I could have told him
> >
> > "For X thousand dollars I can improve your typical Los Angeles 1.5
> > MB/s DSL user's mean response time from seven sec to five sec"
> >
> > then he would have been delighted.
> >
> > If I could go further and say "This should increase your
conversion
> > rate from 17% to 21%"
> > then he could easily see whether the work was worthwhile.
> >
> >
> >
> > On Feb 19, 2009, at 3:47 PM, Ernest Mueller wrote:
> >
> >> Yeah, this is hard. We have a complex site and so when we do
> >> optimizations, it's usually around the same time as 50 different
Web app
> >> releases, all of which not only affect performance but also
ideally are
> >> targeted at driving conversion themselves.
> >>
> >> Ernest
> >> ______________________
> >> UN-altered REPRODUCTION and DISSEMINATION of
> >> this IMPORTANT information is ENCOURAGED.
> >>
> >> From: Peter Booth <peter_booth@...
<mailto:peter <mailto:peter> _booth%40mac.com>>
> >>
> >> To: exceptional-performance@yahoogroups.com <mailto:exceptional-performance%40yahoogroups.com>  
> >> < <mailto:exceptional> mailto:exceptional <mailto:exceptional> -performance%40yahoogroups.com>,
yslow@yahoogroups.com <mailto:yslow%40yahoogroups.com>  
> >> <mailto:yslow <mailto:yslow> %40yahoogroups.com>
> >>
> >> Date: 02/19/2009 02:43 PM
> >>
> >> Subject: [exceptional-performance] Show me the numbers. How has
> >> performance improvement impacted conversion rate?
> >>
> >> Sent by: exceptional-performance@yahoogroups.com <mailto:exceptional-performance%40yahoogroups.com>  
> >> < <mailto:exceptional> mailto:exceptional <mailto:exceptional> -performance%40yahoogroups.com>
> >>
> >>
> >> I take it on faith that, all things being equal, a faster website
> >> will hold
> >> peoples attention longer, and consequently have higher conversion
> >> rates. I just
> >> moved a domain from register.com to namecheap.com because I was
> >> frustrated by
> >> register's slow admin application.
> >>
> >>
> >>
> >>
> >> I've also seen two quotes from Google and Amazon that are
frequently
> >> repeated, but
> >> never with specific details connecting them to specific tests or
> >> projects . I
> >> don't have any personal experience of seeing conversion rates
change
> >> as a result
> >> of changes in performance.
> >>
> >> Has anyone in this group seen this, either as A/B testing or
> >> before/after?
> >>
> >> Can anyone quote, or even hint, at profit/sales/revenue/signups
rates
> >> correlated
> >> with performance?
> >>
> >> I'm contemplating finding a local online store and doing a
project,
> >> even as a
> >> freebie, to get some data that I can use for marketing.
> >>
> >> Peter
> >>
> >>
> >>
> >> <pic12382.gif><pic17421.gif>
> >
> >
>

  
   

 
   

 
    

I was wondering if anyone could go through how the Performance team
came to the conclusion that:

pngcrush image.png -rem alla -reduce -brute result.png

(even though it should read "pngcrush -rem alla -reduce -brute
image.png result.png")

is the best method for crushing .png files.  Are there any other
options that could lead to even further reduced image files?

Hi Stoyan,

    Thanks for the info. Could you share some of the methodologies
used by the Yahoo experiment? Delaying a page by 400ms is hardly
noticable by the users, how could that result in 5-9% drop in total
page traffic?

Thanks.

Yubin


--- In exceptional-performance@yahoogroups.com, Stoyan Stefanov
<stoyan@...> wrote:
>
> From
> http://googlesystem.blogspot.com/2006/11/need-for-speed-in-web-
applications.
> html
>
> > When launched, Google Maps was pretty slow, but a speed
improvement brought a
> > lot more traffic. "When the Google Maps home page was put on a
diet, shrunk
> > from 100K to about 70K to 80K, traffic was up 10 percent the
first week and in
> > the following three weeks, 25 percent more."
>
> There?s also the Yahoo experiment that delaying a page artificially
with
> 400ms caused the abandonment rate (people go away before onload) to
jump
> with 5-9%. So perceived performance may be different than roundtrip
> performance, but there?s a correlation between a slow site and
people just
> giving up. In addition, once you leave a site because it?s slow,
you often
> don?t bother to come back.
>
> I don?t have a URL that describes this experiment, just mentions in
> presentations from Yahoos here and there
> http://www.slideshare.net/stoyan/yslow-20-presentation
> http://www.slideshare.net/stubbornella/7-habits-of-exceptional-
performance-p
> resentation
>
> Best,
> Stoyan
>
>
> On 2/23/09 5:03 AM, "Peter Booth" <peter_booth@...> wrote:
>
> >
> >
> >
> > Stoyan,
> >
> > Do you have a reference to the Google datapoint?
> >
> > Peter Booth
> > (917) 445 5663 peter_booth@... <mailto:peter_booth@...>
> >
> > On Feb 22, 2009, at 3:03 PM, Stoyan Stefanov <stoyan@...> wrote:
> >
> >>
> >>
> >>
> >> There's one study that shows that if users are able to complete
their task
> >> (e.g. make a reservation) they perceive the site as faster, even
though the
> >> roundtrip time may be slower than the competition :)
> >>
> >> But then there's Google maps that got 20% more traffic when they
improved the
> >> loading time by 20%
> >>
> >> --
> >> Stoyan Stefanov
> >> http://phpied.com <http://phpied>
> >>
> >> On Feb 21, 2009, at 11:40 AM, "beinwal" <beinwal@...> wrote:
> >>
> >>>
> >>>
> >>>
> >>> Will they also be differentiating between perceived and actual
> >>> response times?
> >>>
> >>> This is another grey area along with the impact of performance
to
> >>> overall customer satisfaction.
> >>>
> >>> We just redesigned our product pages and they have increased in
size
> >>> (a.k.a. slower), but the time where the page begins to render
is much
> >>> improved.  I am hoping that helps to balance it out, but it is
too
> >>> soon to tell.
> >>>
> >>> --- In exceptional-performance@yahoogroups.com
> >>> <mailto:exceptional-performance%40yahoogroups.com> , Steve
Souders
> >>> <steve@> wrote:
> >>>> >
> >>>> > This year's Velocity conference will have at least two talks
from
> >>> big
> >>>> > (alexa top 100) websites that quantify the impact of page
load times
> >>> to
> >>>> > revenue.
> >>>> >
> >>>> > -Steve
> >>>> >
> >>>> > On 2/19/2009 1:07 PM, Peter Booth wrote:
> >>>>> > >
> >>>>> > > That's what got me thinking about A/B testing of
performance
> >>> changes.
> >>>>> > >
> >>>>> > >
> >>>>> > > I was talking with a potential client two weeks ago and
had an
> >>> "Aha"
> >>>>> > > moment.
> >>>>> > > I realized that this guy really didn't want to hire me by
the
> >>> hour.
> >>>>> > > He wanted to buy the results of performance engineering.
> >>>>> > >
> >>>>> > > If I could have told him
> >>>>> > >
> >>>>> > > "For X thousand dollars I can improve your typical Los
Angeles 1.5
> >>>>> > > MB/s DSL user's mean response time from seven sec to five
sec"
> >>>>> > >
> >>>>> > > then he would have been delighted.
> >>>>> > >
> >>>>> > > If I could go further and say "This should increase your
> >>> conversion
> >>>>> > > rate from 17% to 21%"
> >>>>> > > then he could easily see whether the work was worthwhile.
> >>>>> > >
> >>>>> > >
> >>>>> > >
> >>>>> > > On Feb 19, 2009, at 3:47 PM, Ernest Mueller wrote:
> >>>>> > >
> >>>>>> > >> Yeah, this is hard. We have a complex site and so when
we do
> >>>>>> > >> optimizations, it's usually around the same time as 50
different
> >>> Web app
> >>>>>> > >> releases, all of which not only affect performance but
also
> >>> ideally are
> >>>>>> > >> targeted at driving conversion themselves.
> >>>>>> > >>
> >>>>>> > >> Ernest
> >>>>>> > >> ______________________
> >>>>>> > >> UN-altered REPRODUCTION and DISSEMINATION of
> >>>>>> > >> this IMPORTANT information is ENCOURAGED.
> >>>>>> > >>
> >>>>>> > >> From: Peter Booth <peter_booth@
> >>> <mailto:peter <mailto:peter> _booth%40mac.com>>
> >>>>>> > >>
> >>>>>> > >> To: exceptional-performance@yahoogroups.com
> >>>>>> <mailto:exceptional-performance%40yahoogroups.com>
> >>>>>> > >> < <mailto:exceptional> mailto:exceptional
<mailto:exceptional>
> >>>>>> -performance%40yahoogroups.com>,
> >>> yslow@yahoogroups.com <mailto:yslow%40yahoogroups.com>
> >>>>>> > >> <mailto:yslow <mailto:yslow> %40yahoogroups.com>
> >>>>>> > >>
> >>>>>> > >> Date: 02/19/2009 02:43 PM
> >>>>>> > >>
> >>>>>> > >> Subject: [exceptional-performance] Show me the numbers.
How has
> >>>>>> > >> performance improvement impacted conversion rate?
> >>>>>> > >>
> >>>>>> > >> Sent by: exceptional-performance@yahoogroups.com
> >>>>>> <mailto:exceptional-performance%40yahoogroups.com>
> >>>>>> > >> < <mailto:exceptional> mailto:exceptional
<mailto:exceptional>
> >>>>>> -performance%40yahoogroups.com>
> >>>>>> > >>
> >>>>>> > >>
> >>>>>> > >> I take it on faith that, all things being equal, a
faster website
> >>>>>> > >> will hold
> >>>>>> > >> peoples attention longer, and consequently have higher
conversion
> >>>>>> > >> rates. I just
> >>>>>> > >> moved a domain from register.com to namecheap.com
because I was
> >>>>>> > >> frustrated by
> >>>>>> > >> register's slow admin application.
> >>>>>> > >>
> >>>>>> > >>
> >>>>>> > >>
> >>>>>> > >>
> >>>>>> > >> I've also seen two quotes from Google and Amazon that
are
> >>> frequently
> >>>>>> > >> repeated, but
> >>>>>> > >> never with specific details connecting them to specific
tests or
> >>>>>> > >> projects . I
> >>>>>> > >> don't have any personal experience of seeing conversion
rates
> >>> change
> >>>>>> > >> as a result
> >>>>>> > >> of changes in performance.
> >>>>>> > >>
> >>>>>> > >> Has anyone in this group seen this, either as A/B
testing or
> >>>>>> > >> before/after?
> >>>>>> > >>
> >>>>>> > >> Can anyone quote, or even hint, at
profit/sales/revenue/signups
> >>> rates
> >>>>>> > >> correlated
> >>>>>> > >> with performance?
> >>>>>> > >>
> >>>>>> > >> I'm contemplating finding a local online store and
doing a
> >>> project,
> >>>>>> > >> even as a
> >>>>>> > >> freebie, to get some data that I can use for marketing.
> >>>>>> > >>
> >>>>>> > >> Peter
> >>>>>> > >>
> >>>>>> > >>
> >>>>>> > >>
> >>>>>> > >> <pic12382.gif><pic17421.gif>
> >>>>> > >
> >>>>> > >
> >>>> >
> >>>
> >>>
> >>>
> >>
> >>
> >
> >
> >
> >>
>

 

I was wondering if anyone could go through how the Performance team
came to the conclusion that:

pngcrush image.png -rem alla -reduce -brute result.png

(even though it should read "pngcrush -rem alla -reduce -brute
image.png result.png")

is the best method for crushing .png files.  Are there any other
options that could lead to even further reduced image files?

  
    

Hello,

I wonder if the number of flash object used on a page increase a lot
the display render time of the page.

Regards

Mathieu

Hello,

In my opinion, "-rem alla" is a bad idea, since it will strip out bKGD
chunks.

bKGD chunks are USEFUL for GRACEFUL DEGRADATION WITH IE6. In some
cases, using a custom opaque background (rather than the default light
gray one) for IE6 is an acceptable compromise, which saves you the
trouble of having to author a different GIF or PNG-8 image just for IE6.

(And I know that AlphaImageLoader exists, or that serving PNG-8 for
everyone might be better in most cases, but let's not discuss these,
it's not the point.)

Right now, when using smush.it, I must separate the images I will
optimize with smush.it and the ones I will optimize separately with my
own install of pngcrush and my own parameters. Just so that it won't
remove the bKGD chunk from some files.

The options I use are basically this:
"-rem gAMA -rem cHRM -rem iCCP -rem sRGB -rem text"

Not removing the bKGD chunk is, in my opinion, the way to go since:
1. Most authoring tools won't write it if the user doesn't request it.
2. Some authoring tools don't write it at all anyway (Photoshop?).
3. If it's there, it's quite likely that it's on purpose.

I remember writing to Stoyan on that topic, saying that providing an
option for this in smush.it would be great, and he replied he would
consider it. Any news on that?

--
Florent Verschelde
http://covertprestige.net

I was running a webpagetest from
http://performance.webpagetest.org:8080/ (which is awesome btw) and I
noticed a component that was taking 2251ms to load. The component is
labeled 'undefined'. Does anyone know what this could be? I scanned
the html and didn't notice any empty 'src' tags for images, js or css.
Could this be coming from a flash file?

Here is the actual test I ran. The undefined component is #64.
http://performance.webpagetest.org:8080/result/090225_140P/1/details/

Hi Florent,

If you are using bKGD to provide a background color for IE6 than your image
doesn't
require truecolor transparency and you are using the wrong file format.  You
would get
better compression with PNG8 and get true progressive enhancement by adding in a
few
alpha transparent pixels [1].

Do you have an example where this technique is used on a live website?

Thanks,
Nicole

1. http://www.sitepoint.com/blogs/2007/09/18/png8-the-clear-winner/

--- In exceptional-performance@yahoogroups.com, "fversch" <florent@...> wrote:
>
> Hello,
>
> In my opinion, "-rem alla" is a bad idea, since it will strip out bKGD
> chunks.
>
> bKGD chunks are USEFUL for GRACEFUL DEGRADATION WITH IE6. In some
> cases, using a custom opaque background (rather than the default light
> gray one) for IE6 is an acceptable compromise, which saves you the
> trouble of having to author a different GIF or PNG-8 image just for IE6.
>
> (And I know that AlphaImageLoader exists, or that serving PNG-8 for
> everyone might be better in most cases, but let's not discuss these,
> it's not the point.)
>
> Right now, when using smush.it, I must separate the images I will
> optimize with smush.it and the ones I will optimize separately with my
> own install of pngcrush and my own parameters. Just so that it won't
> remove the bKGD chunk from some files.
>
> The options I use are basically this:
> "-rem gAMA -rem cHRM -rem iCCP -rem sRGB -rem text"
>
> Not removing the bKGD chunk is, in my opinion, the way to go since:
> 1. Most authoring tools won't write it if the user doesn't request it.
> 2. Some authoring tools don't write it at all anyway (Photoshop?).
> 3. If it's there, it's quite likely that it's on purpose.
>
> I remember writing to Stoyan on that topic, saying that providing an
> option for this in smush.it would be great, and he replied he would
> consider it. Any news on that?
>
> --
> Florent Verschelde
> http://covertprestige.net
>

Thanks Pat. I see the undefined value in the swfobject.js file you are
referring to.

I'm still a little unclear as to why both YSlow and webpagetester are
recording this as a component that is being downloaded. I'll look at
it a little more and see if this is making a call at some point.

--- In exceptional-performance@yahoogroups.com, "Patrick Meenan"
<PatMeenan@...> wrote:
>
> Looks like /shared/components/js/lib/swfobject.js defines it right
at the
> beginning - var swfobject=function(){var Z="undefined" and also
later in the
> code when it is detecting the flash player version:
> hasFlashPlayerVersion:O,createSWF:function(k,j,i){if(a.w3cdom&&S){return
> R(k,j,i)}else{return undefined}}
>
> If you google for " swfobject undefined" you should get a bunch of
> information.
>
> -Pat
>
>
>
>
>
> From: exceptional-performance@yahoogroups.com
> [mailto:exceptional-performance@yahoogroups.com] On Behalf Of robfaraj
> Sent: Thursday, February 26, 2009 10:00 AM
> To: exceptional-performance@yahoogroups.com
> Subject: [exceptional-performance] Webpagetester: Undefined Component
>
>
>
>
> I was running a webpagetest from
> http://performance.webpagetest.org:8080/ (which is awesome btw) and I
> noticed a component that was taking 2251ms to load. The component is
> labeled 'undefined'. Does anyone know what this could be? I scanned
> the html and didn't notice any empty 'src' tags for images, js or css.
> Could this be coming from a flash file?
>
> Here is the actual test I ran. The undefined component is #64.
> http://performance.webpagetest.org:8080/result/090225_140P/1/details/
>

when anayzing the components loading on my webpage I notice the
following:
1 html loads
2. css loads (some in parallel,etc all good although these will be
combined in future.
3. images load (ok)
4. couple of js files load

so far so good.
However  for a particular js it takes 24 ms to load it, but after that
the waterfall view shows no activity for about 300ms (there's not light
grey or dark grey bar )
Then after that 300ms another couple of javascripts start loading.

In other words there's a gap in loading of components where the browser
just seems to sit idle!?

What's happening here?
I thought perhaps some form of javascript processing going on,
interpreting the loaded file of 24ms, but the firebug js profiler shows
no activity in the file on page load. (it's a js of a framework, so I
have not really a clue what's going on in there)

To test that this isn't normal I loaded some other pages from various
websites: I see no gap in the loading of components in any of them.

Anyone?
Thanks,

Britske

when anayzing the components loading on my webpage I notice the
following:
1 html loads
2. css loads (some in parallel,etc all good although these will be
combined in future.
3. images load (ok)
4. couple of js files load

so far so good.
However  for a particular js it takes 24 ms to load it, but after that
the waterfall view shows no activity for about 300ms (there's not light
grey or dark grey bar )
Then after that 300ms another couple of javascripts start loading.

In other words there's a gap in loading of components where the browser
just seems to sit idle!?

What's happening here?
I thought perhaps some form of javascript processing going on,
interpreting the loaded file of 24ms, but the firebug js profiler shows
no activity in the file on page load. (it's a js of a framework, so I
have not really a clue what's going on in there)

To test that this isn't normal I loaded some other pages from various
websites: I see no gap in the loading of components in any of them.

Anyone?
Thanks,

Britske

Hi Britske,

here a few ideas what might issue the long "idle" times:
- connection time to the server (time that is needed to establish the
connection to the server that hosts the js files)
- dns time (time that is needed to resolve the ip adress of the host
of the js files from your isps dns server)
- first byte time (time that is needed to fullfill the request from
your client (i. e. FF3) by the webserver that already knows what you want)

Connection time and First Byte Time issues mainly can be adressed
directly to the infrastructure.
So you should check: Are the js files located on an application server
that gets requests from more than one webserver. Is the loadbalancer
configured well, are the js files created dynamically on user request
(maybe internal database connection issues). and a lot of things more
that might issue the long "idle" times.
It normally has nothing to do with the time that is needed to render
the js files. those are not captured in Firebug.

Hope that enlights the "idle" time.

--- In exceptional-performance@yahoogroups.com, "gbrits27"
<gbrits27@...> wrote:
>
> when anayzing the components loading on my webpage I notice the
> following:
> 1 html loads
> 2. css loads (some in parallel,etc all good although these will be
> combined in future.
> 3. images load (ok)
> 4. couple of js files load
>
> so far so good.
> However  for a particular js it takes 24 ms to load it, but after that
> the waterfall view shows no activity for about 300ms (there's not light
> grey or dark grey bar )
> Then after that 300ms another couple of javascripts start loading.
>
> In other words there's a gap in loading of components where the browser
> just seems to sit idle!?
>
> What's happening here?
> I thought perhaps some form of javascript processing going on,
> interpreting the loaded file of 24ms, but the firebug js profiler shows
> no activity in the file on page load. (it's a js of a framework, so I
> have not really a clue what's going on in there)
>
> To test that this isn't normal I loaded some other pages from various
> websites: I see no gap in the loading of components in any of them.
>
> Anyone?
> Thanks,
>
> Britske
>

Oh, sorry I didn't see this till now.

Yeah, we use the Citrix Netscaler devices to front our Web servers.  We
switched from a software-based load balancer to them in 2006.  We get a lot
of mileage out of them; the compression, caching, multiplexing, SSL
offload, and keepalives all together in one package was compelling.

In fact - I don't see any proprietary info in it after review and a little
redaction, so here, I'm going to share our before/after metrics from our
site after our Netscaler implementation.  Because we got all these benefits
at once, we couldn't always separate the effects of one from another, so it
doesn't clearly say "here's the benefit of multiplexing" specifically, but
there's enough data here you can do that somewhat.  Props to Mark Osborne
on my team here at NI for putting all this together.

(See attached file: sharable success metrics.ppt)

I can't share the matrix from our eval, but the short form is we really
liked Citrix and F5 for this.  F5 adds more functionality over and above
the Netscalers with their iRules but is also way, way more expensive.  So
we went Netscaler.  Looked at Juniper and Radware too.

Ernest
______________________
UN-altered REPRODUCTION and DISSEMINATION of
this IMPORTANT information is ENCOURAGED.




   From:       Peter Booth <peter_booth@...>

   To:         "exceptional-performance@yahoogroups.com"
<exceptional-performance@yahoogroups.com>

   Cc:         "exceptional-performance@yahoogroups.com"
<exceptional-performance@yahoogroups.com>

   Date:       02/13/2009 11:21 AM

   Subject:    Re: [exceptional-performance] RFC: multipart HTTP sessions

   Sent by:    exceptional-performance@yahoogroups.com











  I recall seeing a product six or seven years ago that sat in front of a pool of
  webservers acting as a concentrator so that say 2000 concurrent short-lived
  browser connections would be demultiltiplexed by this appliance and the
appliance
  would maintain a handful of longlasting tcp connections to the web server. I
  didn't get a chance to evaluate it and haven't seen it since.



  Does anyone recall this product and even have experience with it?

  Peter Booth
  (917) 445 5663 peter_booth@...

  On Feb 12, 2009, at 6:50 PM, Patrick Meenan <PatMeenan@...> wrote:




        The problem with pipelining is that it returns the requests in the order
        asked and the client doesn't necessarily know the best order to ask for
        things to deliver the best user experience (a really large image could
hold
        everything up for example).  Requesting the objects one at a time from
        multiple connections allows for a large object to block one of the
        connections but nothing gets stuck behind it and the next request can go
on
        the other connection and download in parallel.


        If pipelining supported interleaving the responses (giving 2 virtual
"pipes"
        over a single connection) then it would be a lot more similar but that
adds
        a lot of complexity.


        -Pat





        From: exceptional-performance@yahoogroups.com [mailto:exceptional
        -performance@yahoogroups.com] On Behalf Of Gerald Oskoboiny
        Sent: Thursday, February 12, 2009 5:05 PM
        To: exceptional-performance@yahoogroups.com
        Subject: Re: [exceptional-performance] RFC: multipart HTTP sessions





        * steamer2k <steamer25@...> [2009-02-11 21:58-0000]
        > Given the number one performance rule in the YSlow guidelines
        > (http://developer.yahoo.com/yslow/help/#guidelines) is "Make Fewer
        > HTTP Requests". I thought this might be a good forum to start a
        > conversation about a potentially very useful feature.
        >
        > This feature would require buy-in from browser and server developers*
        > but even partial implementation could be useful. Here goes:
        >
        > Given cases where I need several files including dependencies, e.g.,
        >
        > yahoo-dom-event.js
        > connection-min.js
        > mystuff-min.js
        >
        > ...it would be useful to make a single, batched request for all three

        This sounds similar to HTTP/1.1 Pipelining:
        http://en.wikipedia..org/wiki/HTTP_pipelining

        > *This would require special HTML markup and an HTTP contract. But as
        > Crockford has inferred in the past, you don't need a standard to
        > innovate. There are plenty of open source browsers and servers. If we
        > had agreement between say, Apache and Mozilla, something like this
        > could have an appreciable impact.

        I think pipelining was added to Apache about 12 years ago, and
        Mozilla has supported it for ages as well, but apparently it's
        turned off by default in Firefox.

        Many of the Firefox performance tuning articles I have seen
        recommend enabling pipelining; I don't know why it isn't enabled
        by default. Mozilla's Pipelining FAQ doesn't say why.
        http://www.mozilla.org/projects/netlib/http/pipelining-faq.html

        --
        Gerald Oskoboiny <gerald@...>
        http://impressive.net/people/gerald/

Hi Britske,

You may want to try IBM page detailler as a complement to firebug. You'll see a
lot more
gaps as the waterfall charts are created using a packet sniffer rather than DOM
events.
The results are much more accurate. Also, you can run it with multiple browsers
to
understand differences in how they handle component types.

In the case of JavaScript, some browsers block downloads until the file has been
downloaded and _executed_. It is kind of logical, after all the script could do
a
document.write and all of those components would be irrelevant.

The gaps you are seeing are probably the execution time of the JS.

Cheers,
Nicole

--- In exceptional-performance@yahoogroups.com, "gbrits27" <gbrits27@...> wrote:
>
> when anayzing the components loading on my webpage I notice the
> following:
> 1 html loads
> 2. css loads (some in parallel,etc all good although these will be
> combined in future.
> 3. images load (ok)
> 4. couple of js files load
>
> so far so good.
> However  for a particular js it takes 24 ms to load it, but after that
> the waterfall view shows no activity for about 300ms (there's not light
> grey or dark grey bar )
> Then after that 300ms another couple of javascripts start loading.
>
> In other words there's a gap in loading of components where the browser
> just seems to sit idle!?
>
> What's happening here?
> I thought perhaps some form of javascript processing going on,
> interpreting the loaded file of 24ms, but the firebug js profiler shows
> no activity in the file on page load. (it's a js of a framework, so I
> have not really a clue what's going on in there)
>
> To test that this isn't normal I loaded some other pages from various
> websites: I see no gap in the loading of components in any of them.
>
> Anyone?
> Thanks,
>
> Britske
>

Really?  I would think that the connection, dns, and first byte time are
included in the individual object time in the waterfall, at least I was
assuming they were because they are in Keynote waterfalls etc.

We see idle time a good bit and usually it is JS execution, CSS rendering,
or other purely "inside the browser" activity.

Ernest
______________________
UN-altered REPRODUCTION and DISSEMINATION of
this IMPORTANT information is ENCOURAGED.




   From:       "Heiko Specht" <specht_heiko@...>

   To:         exceptional-performance@yahoogroups.com

   Date:       02/27/2009 05:58 AM

   Subject:    [exceptional-performance] Re: firebug net panel question: files
being loaded with large gap in between

   Sent by:    exceptional-performance@yahoogroups.com











  Hi Britske,

  here a few ideas what might issue the long "idle" times:
  - connection time to the server (time that is needed to establish the
  connection to the server that hosts the js files)
  - dns time (time that is needed to resolve the ip adress of the host
  of the js files from your isps dns server)
  - first byte time (time that is needed to fullfill the request from
  your client (i. e. FF3) by the webserver that already knows what you want)

  Connection time and First Byte Time issues mainly can be adressed
  directly to the infrastructure.
  So you should check: Are the js files located on an application server
  that gets requests from more than one webserver. Is the loadbalancer
  configured well, are the js files created dynamically on user request
  (maybe internal database connection issues). and a lot of things more
  that might issue the long "idle" times.
  It normally has nothing to do with the time that is needed to render
  the js files. those are not captured in Firebug.

  Hope that enlights the "idle" time.

  --- In exceptional-performance@yahoogroups.com, "gbrits27"
  <gbrits27@...> wrote:
  >
  > when anayzing the components loading on my webpage I notice the
  > following:
  > 1 html loads
  > 2. css loads (some in parallel,etc all good although these will be
  > combined in future.
  > 3. images load (ok)
  > 4. couple of js files load
  >
  > so far so good.
  > However for a particular js it takes 24 ms to load it, but after that
  > the waterfall view shows no activity for about 300ms (there's not light
  > grey or dark grey bar )
  > Then after that 300ms another couple of javascripts start loading.
  >
  > In other words there's a gap in loading of components where the browser
  > just seems to sit idle!?
  >
  > What's happening here?
  > I thought perhaps some form of javascript processing going on,
  > interpreting the loaded file of 24ms, but the firebug js profiler shows
  > no activity in the file on page load. (it's a js of a framework, so I
  > have not really a clue what's going on in there)
  >
  > To test that this isn't normal I loaded some other pages from various
  > websites: I see no gap in the loading of components in any of them.
  >
  > Anyone?
  > Thanks,
  >
  > Britske
  >

Hi Britske,

You may want to try IBM page detailler as a complement to firebug. You'll see a lot more
gaps as the waterfall charts are created using a packet sniffer rather than DOM events.
The results are much more accurate. Also, you can run it with multiple browsers to
understand differences in how they handle component types.

In the case of JavaScript, some browsers block downloads until the file has been
downloaded and _executed_. It is kind of logical, after all the script could do a
document.write and all of those components would be irrelevant.

The gaps you are seeing are probably the execution time of the JS.

Cheers,
Nicole

--- In exceptional-performance@yahoogroups.com, "gbrits27" <gbrits27@...> wrote:
>
> when anayzing the components loading on my webpage I notice the
> following:
> 1 html loads
> 2. css loads (some in parallel,etc all good although these will be
> combined in future.
> 3. images load (ok)
> 4. couple of js files load
>
> so far so good.
> However for a particular js it takes 24 ms to load it, but after that
> the waterfall view shows no activity for about 300ms (there's not light
> grey or dark grey bar )
> Then after that 300ms another couple of javascripts start loading.
>
> In other words there's a gap in loading of components where the browser
> just seems to sit idle!?
>
> What's happening here?
> I thought perhaps some form of javascript processing going on,
> interpreting the loaded file of 24ms, but the firebug js profiler shows
> no activity in the file on page load. (it's a js of a framework, so I
> have not really a clue what's going on in there)
>
> To test that this isn't normal I loaded some other pages from various
> websites: I see no gap in the loading of components in any of them.
>
> Anyone?
> Thanks,
>
> Britske
>

Hello Nicole and thanks for the answer.

> If you are using bKGD to provide a background color for
> IE6 then your image doesn't require truecolor transparency
> and you are using the wrong file format.

Could you please define “require”?
(Sorry, rhetorical question. ;))

The thing is, the stated philosophy of smush.it is to let the
designer/front-end developer choose the format, and to work on that
choosen format to optimize it.

True enough, PNG-8 is much lighter than PNG-24/32, and using PNG-8
makes a lot of sense in many cases. I use PNG-8 a lot.

But sometimes using truecolor transparency (aka PNG-32) is interesting
too. (We could talk of abusive uses of PNG-32, but that's not really
the point here. Let's just say we're in a case where truecolor
transparency makes sense.)

So how do you handle IE6? AlphaImageLoader is limited, has many
secondary effects that make it impossible to use in some cases,
requires some DOM manipulations if you want to use it for IMG elements
(which may hamper accessibility), must be used with CSS hacks or
conditional comments and thus make maintenance harder, etc. I tend to
avoid it as much as possible. So, what are the other options? There
are three of them:
1. Give specific styles to IE6 to make it use a different image (JPEG
or PNG-8).
2. Use a PNG-8 image with paletted transparency (you will lose all
non-opaque pixels on IE6, but it may work decently for some
carefully-crafted images).
3. Accept that the image will be opaque in IE6, but use a custom
opaque color on the image to make it blend in its surroundings. That's
the bKGD chunk.

I think all three (all four with AlphaImageLoader) are far from
perfect but interesting solutions.

> get true progressive enhancement by adding in a few
> alpha transparent pixels [1].

The image must be specifically crafted with that solution in mind if
you want it to look decently in IE6. For instance, it could be an
opaque box, with an opaque 1px border that makes it clearly visible,
then with a transparent drop shadow. You will lose the drop shadow in
IE6 (and get it on every other browser), but if the pixels that remain
form a decent image then you're fine.

That's one limit (to an otherwise great solution). Another one is that
Fireworks is not always good at creating those images. For instance:
if you have an object with 30 colors, then a shadow that is yet
another color (with varying alpha), trying to export this as PNG-8
with paletted transparency won't work that well. Even if you tell
Fireworks (CS3 and CS4) to use 256 colors, or maybe just 50, or 90,
whatever, well it will restrict itself to a quota of 31 colors (the
number of actual colors that are in the original object). So you end
up with an image that's corrupted even though you would only need
something like 50 palette slots to express it perfectly. And in some
cases the loss is really clearly visible. That's obviously a bug in
Fireworks, but since FW is the main if not only authoring tool for
PNG-8 with paletted alpha… it makes the format less useful than it
could have been.
(Well, at least that's my experience. I may need to tweak the FW
settings to get things working as expected…)

> Do you have an example where this technique is
> used on a live website?

See for instance:
- http://microplaza.com/static/images/middle-bg.png (358 bytes)
- http://microplaza.com/static/images/sidebar-tab-white.png (515 bytes)
(go to the domain root to see the website)
Here using PNG-32 is not absolutely necessary, it just helps making
the very discrete shadows work well, whatever the colour behind it
(notice how there's a gradient on the BODY?). The bKGD color is either
white or very light blue. The result on IE6 is as good as a
corresponding PNG-8 would have been, and way better than with no bKGD
chunk.

I have two or three other examples, but they're not public.

Admittedly, the cases where you can use PNG-32 with a bKGD chunk are
quite rare. But it's still useful. I would argue that it's a parallel
solution to the paletted alpha in PNG-8. In one case you lose parts of
your image in IE6 (everything that's not opaque); in the other one you
don't lose those parts but the image itself is opaque (only not with
the usual ugly gray background, but with a custom color). Both are
limited but interesting and sometimes useful solutions.

Hi Peter,

I gave a performance bootcamp workshop at Web Directions North earlier this
month and
we were able to download and install it, but now I can't find the link. Hmm...

Nicole

--- In exceptional-performance@yahoogroups.com, Peter Booth <peter_booth@...>
wrote:
>
> Is PageDetailer still available? I remember it being available as part
> of a specific websphere bundle that was withdrawn, and also from
> alphaworks.
>
> Peter Booth
> (917) 445 5663 peter_booth@...
>
> On Feb 27, 2009, at 1:14 PM, Nicole Sullivan <nicole@...>
> wrote:
>
> > Hi Britske,
> >
> > You may want to try IBM page detailler as a complement to firebug.
> > You'll see a lot more
> > gaps as the waterfall charts are created using a packet sniffer
> > rather than DOM events.
> > The results are much more accurate. Also, you can run it with
> > multiple browsers to
> > understand differences in how they handle component types.
> >
> > In the case of JavaScript, some browsers block downloads until the
> > file has been
> > downloaded and _executed_. It is kind of logical, after all the
> > script could do a
> > document.write and all of those components would be irrelevant.
> >
> > The gaps you are seeing are probably the execution time of the JS.
> >
> > Cheers,
> > Nicole
> >
> > --- In exceptional-performance@yahoogroups.com, "gbrits27"
> > <gbrits27@> wrote:
> > >
> > > when anayzing the components loading on my webpage I notice the
> > > following:
> > > 1 html loads
> > > 2. css loads (some in parallel,etc all good although these will be
> > > combined in future.
> > > 3. images load (ok)
> > > 4. couple of js files load
> > >
> > > so far so good.
> > > However for a particular js it takes 24 ms to load it, but after
> > that
> > > the waterfall view shows no activity for about 300ms (there's not
> > light
> > > grey or dark grey bar )
> > > Then after that 300ms another couple of javascripts start loading.
> > >
> > > In other words there's a gap in loading of components where the
> > browser
> > > just seems to sit idle!?
> > >
> > > What's happening here?
> > > I thought perhaps some form of javascript processing going on,
> > > interpreting the loaded file of 24ms, but the firebug js profiler
> > shows
> > > no activity in the file on page load. (it's a js of a framework,
> > so I
> > > have not really a clue what's going on in there)
> > >
> > > To test that this isn't normal I loaded some other pages from
> > various
> > > websites: I see no gap in the loading of components in any of them.
> > >
> > > Anyone?
> > > Thanks,
> > >
> > > Britske
> > >
> >
> >
>

It is still available on the IBM AlphaWorks site:

http://www.alphaworks.ibm.com/tech/pagedetailer

...Bryan

--- In exceptional-performance@yahoogroups.com, "Nicole Sullivan"
<nicole@...> wrote:
>
> Hi Peter,
>
> I gave a performance bootcamp workshop at Web Directions North
earlier this month and
> we were able to download and install it, but now I can't find the
link. Hmm...
>
> Nicole
>
> --- In exceptional-performance@yahoogroups.com, Peter Booth
<peter_booth@>
> wrote:
> >
> > Is PageDetailer still available? I remember it being available as
part
> > of a specific websphere bundle that was withdrawn, and also from
> > alphaworks.
> >
> > Peter Booth
> > (917) 445 5663 peter_booth@
> >
> > On Feb 27, 2009, at 1:14 PM, Nicole Sullivan <nicole@>
> > wrote:
> >
> > > Hi Britske,
> > >
> > > You may want to try IBM page detailler as a complement to
firebug.
> > > You'll see a lot more
> > > gaps as the waterfall charts are created using a packet sniffer
> > > rather than DOM events.
> > > The results are much more accurate. Also, you can run it with
> > > multiple browsers to
> > > understand differences in how they handle component types.
> > >
> > > In the case of JavaScript, some browsers block downloads until
the
> > > file has been
> > > downloaded and _executed_. It is kind of logical, after all the
> > > script could do a
> > > document.write and all of those components would be irrelevant.
> > >
> > > The gaps you are seeing are probably the execution time of the
JS.
> > >
> > > Cheers,
> > > Nicole
> > >
> > > --- In exceptional-performance@yahoogroups.com, "gbrits27"
> > > <gbrits27@> wrote:
> > > >
> > > > when anayzing the components loading on my webpage I notice
the
> > > > following:
> > > > 1 html loads
> > > > 2. css loads (some in parallel,etc all good although these
will be
> > > > combined in future.
> > > > 3. images load (ok)
> > > > 4. couple of js files load
> > > >
> > > > so far so good.
> > > > However for a particular js it takes 24 ms to load it, but
after
> > > that
> > > > the waterfall view shows no activity for about 300ms (there's
not
> > > light
> > > > grey or dark grey bar )
> > > > Then after that 300ms another couple of javascripts start
loading.
> > > >
> > > > In other words there's a gap in loading of components where
the
> > > browser
> > > > just seems to sit idle!?
> > > >
> > > > What's happening here?
> > > > I thought perhaps some form of javascript processing going on,
> > > > interpreting the loaded file of 24ms, but the firebug js
profiler
> > > shows
> > > > no activity in the file on page load. (it's a js of a
framework,
> > > so I
> > > > have not really a clue what's going on in there)
> > > >
> > > > To test that this isn't normal I loaded some other pages from
> > > various
> > > > websites: I see no gap in the loading of components in any of
them.
> > > >
> > > > Anyone?
> > > > Thanks,
> > > >
> > > > Britske
> > > >
> > >
> > >
> >
>