On 14 Aug 2006, at 14:10, Matthew Toseland wrote:

> On Sat, Aug 12, 2006 at 03:55:06PM -0700, Ian Clarke wrote:
>> On 12 Aug 2006, at 12:36, Matthew Toseland wrote:
>>> Whatever happened to maximizing usability?
>>
>> Imposing a compression format that may be ill suited to what is being
>> inserted doesn't maximize usability, it simply means that the node is
>> meddling in a client issue in which it lacks the competence to
>> meddle.  I can't think of any other data transmission software that
>> takes it upon itself to compress data by default, about which it
>> knows nothing, by default (BitTorrent doesn't, Secure Copy doesn't,
>> LimeWire doesn't, Kazaa didn't, even Apache doesn't).
>
> Transfer-Encoding: gzip / Content-Transfer-Encoding: gzip is pretty
> standard. Very few web pages are put up on the web as .html.gz,  
> which is
> the appproach you are advocating. What actually happens is that the
> server and the browser transparently compress it. Which is what I
> propose for Freenet.

Firstly, last time I checked, no web server compresses files by  
default.  Furthermore, your proposal is to transparently compress all  
file formats, how many non-HTML file formats do web servers  
transparently compress?

>> I'm not arguing against the value of compression, I'm just arguing
>> against the node doing it.  Different types of compression are
>> appropriate for different file types, it doesn't make sense for us to
>> impose one type of compression, that happens to be optimized for text
>> and some binary formats, on all data types.  The client has the
>> expertise to decide on the appropriate format in which to insert
>> content, we don't.
>
> Files which should be compressed lossily already are. We are  
> required to
> transport the data intact; we do so.

I never claimed that compressing files by default violated any kind  
of obligation, except the obligation to make good design decisions.

> When I have tried to insert video etc files, they have usually been
> shrunk by the node. As I said, 1% is worth it.

If so, then let the client or inserter make that determination, it  
isn't the node's place to make it for them.

>>
>>> since FCPv2 is a deliberately high level API
>>> designed to do as much as possible of the work for the client  
>>> author.
>>
>> Yes, but not such that it imposes a uniform solution across all
>> clients where different solutions are needed for different clients.
>
> Different mutually incompatible clients? Lossy compression is done at
> the file format level, but *lossless* compression can be done anywhere
> along the chain. It is for example often used in PPP on low level
> internet links. It is supported transparently in HTTP as well.

It is not done transparently in HTTP *BY DEFAULT*, and it is not done  
in HTTP on all file types.  It is doing it by default that I object to.

>>
>>> Doing it on the client will just lead to a mass of incompatible
>>> standards, and more (duplicated) work for client authors.
>>
>> If by "incompatible" you mean using jpeg to compress an image rather
>> than gzip, or mpeg to compress a video rather than bzip2, then
>> sometimes incompatibility is justified.
>
> JPEG and MPEG are lossy. This is why they can never be applied
> transparently. Another recent poster gave an example where he  
> compressed
> an MPEG file 23% with gzip (27% with lzma).

If people want to further compress these file types, then that is  
their decision to make, not ours to make for them.

>>
>> Now I have presented an argument for why we shouldn't support in-node
>> compression of data at all, but I'm willing to compromise and agree
>> to it provided that it isn't enabled by default.  This way, client
>> authors can use it if they deem it appropriate, but are much less
>> likely to use it where it isn't.
>
> I don't understand your objection. It is usually beneficial even with
> AVIs, and transparent compression is widely used in other protocols.

No protocols that I am familiar with do it regardless of file type as  
you are proposing, further most I am familiar with don't do it by  
default (eg. Apache doesn't, SCP doesn't).

Ian.

Ian Clarke: Co-Founder & Chief Scientist Revver, Inc.
phone: 323.871.2828 | personal blog - http://locut.us/blog

-------------- next part --------------
An HTML attachment was scrubbed...
URL: 
<https://emu.freenetproject.org/pipermail/tech/attachments/20060814/7ab6c1e1/attachment.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: PGP.sig
Type: application/pgp-signature
Size: 186 bytes
Desc: This is a digitally signed message part
URL: 
<https://emu.freenetproject.org/pipermail/tech/attachments/20060814/7ab6c1e1/attachment.pgp>

Reply via email to