Hello,
Firstly, let me describe a test setup I have in the lab.

I have two pcs with an automated browsing tool, they go thru a list of
websites and count the time it takes for each website to load. Then an
average (with only completely loaded websites) is done. This tool uses
IE 6.0.
I configure one of the tools to use a squid proxy (http 1.1 disabled)
and the other tool goes straight to the internet (http 1.1 enabled).
Before every run I stop the squid process and delete the whole cache
directories and recreate them ( -z) just to be sure I'm no caching
anything.
Just to clarify I run the tests simultaneously and they both use the
same T1 line to go out. The PC are clones and have identical hardware.
My squid config is "straight out of the box".

Well, I ran the test and *everytime* I did this the squid connected PC
loads the websites ~0.4s faster on average than the PC connected
straight to the internet. That's ~40 seconds faster for a 100 website
list.

I was intrigued so I did an Ethereal dump on both sides of the
connection. I was hoping to find that Squid was somehow caching most of
this entries. What I've found however is that just a few dozen Kb of
data (out of 10+ Mbs) had actually been cache HITs (probably a page
with multiple instances of the same file, or same AD). And it simply
cannot account for such a huge difference.

I'm completely confused as to how can going thru a proxy, with the
extra network hop and the overhead, beats going straight out to the
net. If anybody has any comments they will certainly be appreciated.
Right now it seems to me that squid is defying physics 

[ Post a follow-up to this message ]

Hi,

Suggest you investigate how many objects your browser is requesting in
a single http connection.

See http://www.mozilla.org/projects/net...lining-faq.html for
an explantion.

It could be squid requesting more of the page on each slow trip across
the internet, your browser then makes the same number of trips to the
squid box - but these are at lesser LAN latency.

A possible explanation, but really just a guess - I dont know SQUID too
well.

- Danny

Al wrote:
> Hello,
> Firstly, let me describe a test setup I have in the lab.
>
> I have two pcs with an automated browsing tool, they go thru a list
of
> websites and count the time it takes for each website to load. Then
an
> average (with only completely loaded websites) is done. This tool
uses
> IE 6.0.
> I configure one of the tools to use a squid proxy (http 1.1 disabled)
> and the other tool goes straight to the internet (http 1.1 enabled).
> Before every run I stop the squid process and delete the whole cache
> directories and recreate them ( -z) just to be sure I'm no caching
> anything.
> Just to clarify I run the tests simultaneously and they both use the
> same T1 line to go out. The PC are clones and have identical
hardware.
> My squid config is "straight out of the box".
>
> Well, I ran the test and *everytime* I did this the squid connected
PC
> loads the websites ~0.4s faster on average than the PC connected
> straight to the internet. That's ~40 seconds faster for a 100 website
> list.
>
> I was intrigued so I did an Ethereal dump on both sides of the
> connection. I was hoping to find that Squid was somehow caching most
of
> this entries. What I've found however is that just a few dozen Kb of
> data (out of 10+ Mbs) had actually been cache HITs (probably a page
> with multiple instances of the same file, or same AD). And it simply
> cannot account for such a huge difference.
>
> I'm completely confused as to how can going thru a proxy, with the
> extra network hop and the overhead, beats going straight out to the
> net. If anybody has any comments they will certainly be appreciated.
> Right now it seems to me that squid is defying physics 

[ Post a follow-up to this message ]