[ih] FTP RIP

[Joseph Touch]

> On Sep 27, 2020, at 6:54 PM, Grant Taylor via Internet-history <internet-history at elists.isoc.org> wrote:
> 
> On 9/27/20 7:29 PM, Joseph Touch via Internet-history wrote:
>> This thread begs the question of where we’d be had TBL had tried to leverage FTP rather than reinventing the wheel with HTTP.
> 
> I feel like I've dabbled with this in the past.  E.g. retrieved an HTML page via FTP and used ftp:// links in the src parameters for elements.
> 
> I want to say I did similar with Gopher.  Or rather, I created a local file that used a gopher:// link for the source of an image that I knew was on a Gopher server.
> 
> I seem to recall it working surprisingly well.  Though it wasn't practical.

For many years, FTP was always the metric of “how is my link speed”, where HTTP always lagged. The application side of FTP was much more mature (it might still be).

> 
>> I recall an early interview that claimed the rationale was that FTP opened two connections for every transfer and he only wanted one.
> 
> I don't know that that's /completely/ accurate per say.  Yes, there are two connections; control and data.  But I believe the control connection can be re-used for multiple subsequent data connections.

Yes, that’s true for FTP, for subsequent transfers to the same site. 

I was relaying what I recalled, which included the rationale that “since we’re only getting one file from a site, two connections result in too much delay”. The irony is indeed that HTTP rather quickly devolved into grabbing dozens of files from a single site (all the images, formats, etc.) and continues to suffer from multiplexing both control and multiple connections with different behaviors over a single channel.

> Firewalling and NATing are two of FTP's Achilles Heals.  Specifically FTPS.

Passive mode addresses both.

> Aside SFTP (SSH) is significantly different than FTPS (FTP over SSL / TLS).

Yes, FTP might have evolved differently...

>> It’s been unfortunate how many of FTP’s features had to be (or still remain to be) reinvented in HTTP.
> 
> I learned in the last few years that it's possible to establish FTP connections with two servers and instruct them to exchange data directly between themselves without traversing the common client.  Or at least the protocol supports it.  I'm not aware of it being a common implementation, much less execution.  --  I do think that FTPS may hinder this somewhat.

Yes - see https://en.wikipedia.org/wiki/File_eXchange_Protocol <https://en.wikipedia.org/wiki/File_eXchange_Protocol>
That’s in RFC 959.

And FTP included resume since 1972 at least as well.

Joe