[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: apt and multiple connections



On Sun, 24 Sep 2000, Hamish Moffatt wrote:

> > Yick.  The current method is much cleaner, if not quite as efficent.
> > I don't like the idea of getting _parts_ of a file from different places,
> > thank you.
> 
> Why not? 

Because when you get the file from a multiple places, you add the
requirement of checking the file for consistancy and errors.  A crc or
md5sum check is needed, both before the transfer (is it the _SAME_ file,
or just one the same size and date?) and after (did we put it back
together correctly?)

Yes, it's probably as secure as a single ftp or http transfer, but you've
upped the complexity of the transfer a magnitude.  It only works with ftp
or http servers that will do 'resume' (yes, many do).

> There's a non-free Windows program called FlashGet which
> does this and it's very effective. Given a URL, it will set up multiple
> connections to the same place to download part of the file each. If
> it can, it'll locate mirrors of the file and download parts from
> those servers (after determining the fastest with ping) instead.
> Very effective, if anti-social.

I have done stuff like this.  I downloaded the Storm Linux 2.0 iso in the
10+ parts they have broken it up into in order to take advantage of a
cable modem (and achieved 300Kbps download speed this way from a server
that was only giving me 50Kbps for a single stream)  When I had a DirecPC
satellite dish, multiple streams was the only way to get 800Kbps downloads
over the claimed bandwidth they sold me.  Any single stream topped off at
about 80K.

If someone wrote an apt routine that cloned Flashget, I'd try it out, but
I really don't see that the benefit outways the work needed to implement.
Adding multiple sources is _far_ easier and will yield almost the exact
same results speedwise in 95% of the cases out there. (getting just 1 file
is about the _only_ time it will be slower, since the more files, the less
it matters)





Reply to: