flex-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From OmPrakash Muppirala <bigosma...@gmail.com>
Subject Re: [Intaller] Speed up download
Date Wed, 04 Sep 2013 22:43:12 GMT
On Wed, Sep 4, 2013 at 3:31 PM, Frédéric THOMAS <webdoublefx@hotmail.com>wrote:

> What I want to try  is chunks downloads with one file per worker and
> compute
> md5 in workers too to go as fast as possible to the next install step, I've
> got a slow bandwidth but when I use my IDM, I download twice faster.
>
> If we want to be smart, I guess we can improve the algorithm having one
> chunk per mirror, quickly check the speed, and retain the 4 fasters.
>

I see a couple of problems with this approach.

1.  Every mirror needs to support the "Transfer-Encoding" (chunked) header
vs. the "Content-Length" header.  I dont think there is any way to ensure
that all the mirrors support this.
2.  If you get one chunk per mirror and if one of the mirrors has a
corrupted artifact, then it will be almost impossible to figure out and
track where the MD5 checksum.
3.  In the past, I have run into a very rare problem with Flash runtime not
being able to receive a chunk received notification on Macs on Wireless.  I
know this is a specific scenario, but not that uncommon.  That bug might
have been fixed now, though.

I think computing MD5 in a worker thread might be a good investment of our
time since we know for sure that it would benefit us.  I am not so sure
about the chunked transfers approach.

Thanks,
Om


>
> Well, I'm not there yet, at the moment I'm playing with URLStream and
> URLRequestHeader to check is possible or not.
>
> Any thoughts or experiences using those 2 classes with Ranges headers ?
>
> -----Message d'origine-----
> De : Alex Harui [mailto:aharui@adobe.com]
> Envoyé : jeudi 5 septembre 2013 00:14
> À : dev@flex.apache.org
> Objet : Re: [Intaller] Speed up download
>
>
>
> On 9/4/13 2:54 PM, "Nicholas Kwiatkowski" <nicholas@spoon.as> wrote:
>
> >We can't.  We have to rely on the Apache system to determine the
> >download mirror.  We are locked into that method.
> I know we have to use mirrors, but are you sure we can't use a different
> algorithm?
>
> -Alex
>
>

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message