[squid-users] range_offset_limit and idm

Amos Jeffries squid3 at treenet.co.nz
Mon Oct 26 18:02:29 UTC 2015

On 27/10/2015 5:29 a.m., joe wrote:
>> Sounds right. idm is sending multiple parallel Range requests.
>> Essentially trying to fake faster downloads by forcing as many resources
> tks but in that case u say squid should fetch the start of the file range
> and continue sending one after the other right so idm see one only at the
> time
> ok why wen u click paus or stop then resume  idm sending multiple parallel
> Range requests ?

Because thats the way idm was designed. To be greedy. Trying to grab as
much as it can in parallel before the network becomes congested. Its own
behavirou makes the network congested, but only after its already got an
unfairly large share of the pipe.

But things have changed since 56K modems were the top of the line CPE
hardware idm and its type were pretending to optimize for users. In
modern hardware and Mbps+ networks a handful of parallel connections
going at full speed are not going to stress anything. Nor speed it up.

The design assumption that a request will be faster when split into
pieces falls flat when it hits HTTP. Which is about stateless messaging
with caches along the way multiplexing the traffic. Not an end-to-end
stateful single-connection protocol that can be abused.

> so squid only know the begining of the file start then user can abius 
> multiple parallel Range requests by just clicking stop resume lol

Yes. But only until idm has received all the bytes up to whatever
range_offset_limit is configured with. Or until quick_abort handling
finishes caching the first Range request ever received. Whichever comes


More information about the squid-users mailing list