Re: Efficiency of FTP URLs

Henrik Frystyk Nielsen (
Mon, 25 Jul 1994 20:04:15 +0200

> > However, NCSA Mosaic for Windows and its progeny DO leave the connection
> > open (permanently, as a matter of fact, until the server times it out.) I
> > thought this was true of 2a2, but it certainly should be of later releases
> > (unless this was changed since I left.) AIR Mosaic leaves it open.
> Hmm, oops. I think I realize the problem now. WinMosaic was setup to use a
> proxy server. So, that was CERN HTTPd 2.0pre5's fault, not WinMosaic. (The
> same actually for all the Mosaic's I tried. So, some of them might already
> have dealt with it...)

Oh - I thought it was CERN HTTPd version 3.0pre5 ;-)

This is no fault - it is the intended way the FTP client works in the
World-Wide Web Library of common code. The problem is typical when
trying to fit a state protocol like FTP into a state less framework
like HTTP. The arguments for the current implementation are:

1) The "standard" time-out period for a FTP-server when connected as
anonymous user is 3-6 minutes. In that time a user can access quite a
lot of FTP-servers and the probability that the FTP-servers simply will
block due to overload is substantial.

2) The FTP protocol doesn't define a hierarchical file structure like
HTTP using URI's. That is, you have no guarantee that you can get from
one place in the file hierarchy on a FTP server to another using CD
and/or CDUP (you can't in general assume that you can use one CD to jump
to the new location). The only way to be sure is to start a new login

Actually the Library _does_ support a FTP session where the control
connection can be reused. The client asks to set up a session and tells
when to take it down again. This is intended to use when/if the client
supports multiple selections from a FTP listing so that the user can
request, e.g., 10 files, the whole directory etc. etc. This doesn't
require any CD's or CDUP and can be done in a "safe" manner.


One final comment - why use FTP for retrieving HTTP documents at all
(with or without inlined images???)

-- cheers --

Henrik Frystyk
+ 41 22 767 8265
World-Wide Web Project,
CERN, CH-1211 Geneva 23,