Re: FTP or HTTP consumes more resources?
Sat, 21 Jan 1995 22:22:20 +0100

Skip Montanaro writes:
> Dan Connolly asked about the relative weight of the FTP and HTTP protocols.
> I have no first-hand experience to share, but will make the observation that
> most Web browsers (until Netscape was released?) treated FTP as stateless,
> so it was still one connection per file with the added overhead of user
> authentication for each file transmitted. It's hard to see how FTP could be
> "lighter weight" than HTTP under those conditions.

The CERN libWWW code has _always_ cached ftp connections. Therefore, it
takes a conscious decision of a browser author to disable it. I know for
certain that at least: AIR Mosaic, Emacs-w3, LineMode Browser, Cello,
WinMosaic, OmniWeb, ad nauseum, all did this correctly. I think that for a
time Mosaic/X was broken in this regards, but am not 100% sure. Not sure
about MacWeb, WinWeb, or MacMosaic. I'm pretty sure that all the SpyGlass
mosaics have always worked correctly in this regard, but could not swear to

> I think Netscape caches FTP connections, however. I can't otherwise
> explain the speed of succesive file retrieves from ftp servers and the
> output of netstat that shows ESTABLISHED connections long after I've left
> the server. Their downfall at one point seemed to be that they
> maintained ftp connections far too long, so that they were consuming
> valuable slots on heavily used ftp servers that restrict the number of
> connections.

This is a bit of a problem. Browsers should have a user-specifiable
timeout for each ftp connection.

> In short, if you're worried about serving FTP-based information to people
> with Web browsers you'll have to try and understand the mix of clients
> (traditional ftp programs vs. inefficient first generation Web browsers vs.
> more efficient tools like Netscape) to see which is the better protocol.

I'll argue against the 'inefficiency' of the first generation web
browsers. :)

-Bill P.