Re: [squid-users] squid + axel - netiquete idea

i just wondered about your worries about netiquette, and concluded that :
squid shouldn't open multiple connections if it is using only one internet
connection (if squid contacts with proxy doing round-robin balancing it
can't know how many internet gateways are available... so this must be an
option) _and_ remote end have no aliases (is single ip, with no load
balancing).

and if we _do_ have multiple ISP's and want to download faster

if remote host doesn't have any aliases :
squid should abort obsolete connections, if their overall transfer rate is
lower than with less connections (indicating that remote host isn't on
faster pipe than sum of our pipes are) and - after short period of 'maxing
out' remote host (testing) squid should limit download to about 50% of
remote's end capablity of upload. (this % should be user-definable).
this check should be repeated after some time (to check wheter we shouldn't
reduce or increase our download speed)

sometimes we download from host which is much faster than sum of all of our
connection is , i.e. 100Mbit server and 10x 100kbit. in this case even if
we'll use all our connections , ablity of upload of server will not be
impacted. thus above scheme of limiting speed apply only if squid can detect
degradation of server's ablity of upload over , lets say - 90% of
connections _and_ server doesn't have aliases.

i think such policy would protect weak hosts from being exploited by download
managers, as they'll be obsolete when squid will do all the job.

second advantage is that if squid will do all the job it'll also _cache_
downloaded file, while when users download using download managers result is
-not- cached .

also - about anti-download managers - it would be enough to set 'range offset
limit' to -1 to prevent use of client-side download managers - this way even
if client will use download manager squid will download file from scratch,
using own multiple-path download mechanisms, with netiquette , and each
extra connection will only result of marking object 'more often used' - thus
more favored for caching.

On Monday 11 August 2003 02:15, Bob Arctor wrote:
> at first - not for every file it is worth doing (only for very large
> files) second - if you don't have more than one connection to internet
> (like multiple dialup lines) it is pointless - and thus it obviously
> shouldn't be _default_ option .
>
> the main thing it is usefull for is listening to mp3's live from server
> without need of downloading them - while still having ablity to cache them
> .
>
> in this case even if squid will open i.e. 10 connections via all gateways
> avaliable in LAN/MAN/WAN, overall bandwitch will be limited with bitrate
> of an mp3 file played live .
>
> i don't see this is 'antisocial' as otherwise users will just go for
> _faster_ download lines, NAT'ed like ADSL services which are common, and
> thus their web servers will not have multiple IP adresses to load balance
> (ADSL doesn't usually have static IP number, or - if it is shared ADSL -
> user will not get any IP number at all, just ablity to download via NAT)
>
> why having 10 slow static IP lines via different ISP's is better than one
> fast ADSL with no ip number ?
>
> because when user will establish own webserver it can be aliased on those
> 10 connections, and set up load balancing rule on his DNS server.
>
> also - it would provide better failover mechanism, when one of servers is
> down/overloaded, connection will continue over remaining ones
> same is with ISP lines users have.
> if one line is down/overloaded, and there is one not used at all, traffic
> could go via it.
>
> On Monday 11 August 2003 01:31, Antony Stone wrote:
> > On Sunday 10 August 2003 11:38 pm, Bob Arctor wrote:
> > > it accelerates in two ways :
> > > 1)if you have more than one connection to the internet, and your proxy
> > > does load balance, or you have multiple interfaces in your machine,
> > > multiple parts of file are downloaded via multiple connections
> > >
> > > 2) if server is load balanced, and it's domain have many aliases,
> > > chosen round robin as you connect, each part of file is downloaded
> > > from different server
> >
> > Both of these methods seem to assume that your connection to the
> > Internet is faster than the rest of the path to the remote server -
> > which I frankly feel is unlikely.
> >
> > I really don't see that opening up multiple connections for downloading
> > from a remote server is going to improve your network's performance,
> > unless there is a deliverate throttle being placed in your path in order
> > to share available bandwidth with other users, in which case trying to
> > bypass it is almost certainly against the Acceptable Use Policy of the
> > system you are connected through.
> >
> > If the remote servers are on a round robin DNS, then they're already
> > going to be nicely load balanced for different users each downloading
> > complete files, so there's no point in creating additional connections
> > for each server by only downloading part of the file from it.
> >
> > I certainly can't see favourable support for this sort of thing getting
> > included in Squid.
> >
> > Regards,
> >
> > Antony.
>
> --

-- 
--