lynx-dev
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: LYNX-DEV Optimizing relative URLs


From: Klaus Weide
Subject: Re: LYNX-DEV Optimizing relative URLs
Date: Wed, 23 Oct 1996 19:29:36 -0500 (CDT)

On Wed, 23 Oct 1996, Al Gilman wrote:

> Lately I have been following the lynx-dev discussion by browsing
> the archive.  When you spend that much time wandering around one
> weakly-served site, you notice things.  Like the time spent
> repeating name lookup each time you move between pages at the
> same site.
> 
> I have been able to speed up my own meandering around the archive
> by perhaps 30-40% by starting with an URL which has an IP number
> for the host path.  Because I am navigating with relative URLs
> around the site, I pick up appreciable speed by the time not
> spent repeating name lookup.
> 
> This suggest two possible avenues of optimization:
> 
>       Rob could use IP-number URLs in the root page at
>       lynx.browser.org for at least the Lynx-dev
>       Archive and the Lynx Enhanced Pages.  This would 
>       accelerate the navigation at those sites for people
>       finding their way thence from lynx.browser.org [or
>       caching a bookmark from a visit started thus].
> 
>       Lynx itself could cache the IP number form of host-path
>       as part of BASE when known, and use it in the construction
>       of synthesized URLs relative to this BASE.  This 
>       introduces failures on the odd chance that the IP node
>       serving a domain name changes during the lifetime of
>       a BASE memory in Lynx [perhaps not our most prevalent
>       failure mode].  This would offer some speedup for all
>       navigation via relative URLs.

Thanks for giving me an opportunity to plug in my daily piece on
libwww5 :).  
 
The new version of the Reference Library has a hostname cache that
does what you propose - including automatically expiring the
cached entried after a configurable time, and automatic selection 
of the "best" address (fastest response) if a name happens to have
more than one IP addresses (lie for example www.w3.org. 

It doesn't have anything to do with BASE tags though, or with synthesising
URLs.  It is just a shortcut avoiding to actually do repeated lookups
of the same names.

You also point out correctly a problem with that:  that remembered lookup
results may become stale.  This would be a problem with sites which
for some reson change their addresses rapidly.  The Library does't
follow the requirement that such a cache must take the TTL (time to live) 
of DNS records into account.

> Both of these are armchair punditry because I am not in a
> position to Just Do It.  But I at least wanted to pass on the IP
> number acceleration idea for anyone who wants to juice up their
> most-used bookmarks.

Certainly at the moment a more practical approach.

Although, I think, if repeated (successful) lookups of the same name
are taking too long, it indicates that something is wrong (or at least 
not-optimal) with your setup.  You could consider running your own
BIND daemon (just for caching) on your machine...

(I assume your repeated lookups are taking to long because you are 
accessing a nameserver over a dialup connection.  This wouldn't
affect shell users on multiuser systems, because those systes will
presumably have a fast connection to the nearest nameserver.)

  Klaus

;
; To UNSUBSCRIBE:  Send a mail message to address@hidden
;                  with "unsubscribe lynx-dev" (without the
;                  quotation marks) on a line by itself.
;



reply via email to

[Prev in Thread] Current Thread [Next in Thread]