Can one mirror a web server ?

[Kevin S. Pickard]

Jim,

WRT your question:

> Question for Bind developers:  Wny couldn't a feature (defaulted to off)
> be added that allows for a simple ICMP test before a request is
> returned. Yes it would slow things down, but it could allow for
> rudimentary failover.  Just a thought.

I am not one, (a BIND dev), but I imagine the answer would go something like: "Fail over is not a function of the DNS".  It has been my observation that, due to it's solid design, people love to "bolt on" some kind of new functionality to the DNS.  These types of ideas are commonly rebuffed with the argument that, although well designed, the DNS was never meant to perform function X, and generally should only be relied on for it's primary functionality.

It's my belief that things work as well as they do because very smart people kept things simple and straightforward, and didn't try to design something that does everything.  That doesn't mean ideas like this aren't neat, or cool, or even a praiseworthy hack, just probably not something that the devs need to be spending significant amounts of time on. :-)

Good question,

ksp


Kevin S. Pickard
AT&T (SBC?) Managed Hosting Operations - Tier III
Infrastructure Engineering, Security and Support


Jim Popovitch wrote:

>On Wed, 2005-02-02 at 15:53 -0500, barrett bonden wrote:
>  
>
>>Can one mirror a web server ?
>>
>>The idea is to make a  failsafe  and transparent dual computer system at two
>>sites.
>>
>>This would involve (can it be done ?) one dns setting point to two ip
>>addresses ?
>>    
>>
>
>Yes/No.
>
>Yes: It can be done such that www.yourdomain.com resolves to multiple IP
>addresses.  Do a `host www.google.com` for an example.
>
>No: What you will get with multiple A records is... well... multiple A
>records.  One user will get the first, the next visitor will get the
>second, the third visitor will get the third.  This achieves
>load-distribution, but not failover.  In the event that one IP goes
>dead, that A record will still be distributed and the "lucky" visitor
>will get a page timeout.
>
>Question for Bind developers:  Wny couldn't a feature (defaulted to off)
>be added that allows for a simple ICMP test before a request is
>returned. Yes it would slow things down, but it could allow for
>rudimentary failover.  Just a thought.
>
>-Jim P.
>
>
>
>
>
>
>
>  
>