i have often though about this myself.. as i run squid
behind a bandwidth manager.. ovr the past few months,
i have come to this conclusion..
when a user sends an HTTP request, it goes to the
router at the bandwidth assigned to the user for his
interface to the router... i use transparent caching..
so, the router deflects the request to the cache
server, using the internal bandwidth of over
100Mbps... but the bandwidth manager will release the
request to the cache server over the 100Mbps ethernet
cable attached to the interfaces of both machines....
now, squid will begin to process the packet, and fetch
the information for the user from the net.. squid here
will use all bandwidth that it has been assigned when
going through the router.... for instance, if your
total net bandwidth was 1Mbps, and you assigned
800Kbps for HTTP, and the rest for anything else,
squid would use the entire 800Kbps to fetch the user's
request... upon reply of the request to squid, from
the remote server, squid now sends the response for
the request back to the user, bypassing the router,
but using values assigned on the bandwidth manager
interface that squid is connected to.... so, if a user
has purchased 64Kbps from the ISP, he can only have
his request[s] travel at that speed, from the cache
server..
what i am saying is, there are two bandwidth limit
assignments, one on the interface to the router, and
another on the interface to the squid box... as HTTP
is the more popular traffic, u will see more packets
for the squid interface than for the router
interface...
so, even though squid can download the web site for
the user at over 800Kbps, the user will not take
advantage of this, and will only receive the
information at his CIR.. which is 64Kbps...
of course, if u don't have a bandwidth manager, u will
respond to the user's requests at the speed at which
he connects to your NOC, or squid server... if he has
a 10Mbps wireless connection, and he connects direct
into your switch, this is the speed at which he will
download info from your squid.. and to him, u will be
the best squid server there is...
good luck
AKNIT
--- Henrik Nordstrom <hno@marasystems.com> wrote: >
Squid is throttled by the client speed.
>
> Small objects that fit in the TCP windows, socket
> buffers and internal
> buffers (total about 50-128KB depending on the OS)
> are fetched quicker.
>
> The delay_pools feature of Squid can be used to
> further throttle the download
> speed of Squid.
>
> Regards
> Henrik Nordström
> Squid Developer
> MARA Systems AB, Sweden
>
>
>
> On Friday 28 December 2001 00.37, Evren Yurtesen
> wrote:
> > I looked at many pages for hours now including the
> mailing list archives
> > but can't find a clear answer to the question in
> my mind.
> >
> > some information about the question:
> > the clients have 64kbit physical bandwidth(ISDN
> line)
> > and the server has 1mbit connection.
> >
> > actual question:
> > when a client requests a file, does the squid
> request the file using the
> > whole bandwidth and finish the fetching as fast as
> possible
> > or
> > it fetches the file as fast as the client can
> accept? In other words, as
> > fast as the data sent to the client is
> acknowledged.
> >
> > Evren
>
> --
> MARA Systems AB, Giving you basic free Squid support
> Customized solutions, packaged solutions and
> priority support
> available on request
__________________________________________________
Do You Yahoo!?
Everything you'll ever need on one web page
from News and Sport to Email and Music Charts
http://uk.my.yahoo.com
Received on Fri Dec 28 2001 - 05:50:02 MST
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 17:05:30 MST