>
> On 10 Nov 2009, at 08:56, Martin Gerdes wrote:
>
>  First, how slow is slow? Time from pushing the send button until the new
>> webpage is loaded rises from 10.6 to 103 seconds.
>>
>
> 10.6 is already horrendously slow (unless perhaps it's a 20-year-old PC),
> which leads me to wonder what you're doing.  One thing is that with
> ProxyHTMLExtended On you could be parsing lots of text, so make
> sure you use flags to ensure you don't apply more rules to it than
> you absolutely need to.


Its a complex application, its running in a slow VM for testing purposes,
and is itself communicating with a slow test backend in another slow VM. I
am not programming the application, but I have no fear about the 10 seconds.
Especially since they have nothing to do with the proxy: It stays just as
slow if I connect directly to the application (without any encoding
transformations at all). So that simply isn't part of my problem (if it is a
problem, it's someone else's).


>
> Does it make any difference if lots of small rewrites have happened
> vs. if nothing matched any proxy_html rules?
>
> No. It stays just as slow if there isn't a single transformation rule at
all (mod_proxy_html is not doing anything).


>  Top on the server shows that the server is sitting idle, so it's not that
>> the recoding is taking up that much CPU (which would be ridiculous anyway).
>>
>
> Does it make any difference if you send an HTTP/1.0 request
> or force a connection close?

The version used in the connections is HTTP/1.1 (I see that in the content
of the transmitted packets).
I wasn't able to try out HTTP/1.0, and haven't looked into how to force a
connection close (what exactly should I try there?)

Instead I have other very interesting observations: I tried the application
out in IE (with the intent to force HTTP/1.0).
Once I press the login button, IE tells me I have no connection to the
internet, and stays with that opinion for any and all sites until I restart
it. Interestingly, IE shows this behaviour even if I comment the line
"ProxyHTMLCharsetOut *" out (which is enough to make firefox fast). [All of
this tried with and without forcing HTTP/1.0]
Commenting out all ProxyHTML directives makes the website work, and work
fast.

 Can you get mod_diagnostics
> output to track the data running through the filter?
>
> I'll try that after lunch. Ask if you want to know anything else. (I can
for example packet sniff the connection between IE and the proxy, and give
you the debug output of mod_proxy_html in that constallation. Provided you
are interested in that information.)

Kind regards,

Martin

Reply via email to