DNS Failover

Maria Iano maria at iano.org
Thu Oct 14 18:42:50 UTC 2004


My apologies for being lazy with my hits-per-day stats. I just pulled the actual info for that web site on last Monday, and it received 11.3 million hits that day. So it really isn't that the site gets a proportionately high number of unique hits. The low TTL method of failing over is working for us very well for a number of different web sites, but I think this one is a good example.

However, I wonder whether it is considered inconsiderate behavior to set very low TTL values, and thus increase the burden for resolvers? We have several records with a 30s TTL, and three with 10s. Those are the lowest.

Maria

On Thu, Oct 14, at 01:35%P so wrote Maria Iano (maria at iano.org):

> I think you may have misread my message. The stat for hits was per day, but the unique visitors stat was per month. The particular site I mentioned contains information and entertainment spread amongst different pages, and clients tend to stay on the site and travel around the different pages for quite a while.
> 
> Maria
> 
> > -----Original Message-----
> > From: bind-users-bounce at isc.org [mailto:bind-users-bounce at isc.org] On
> > Behalf Of David Botham
> > Sent: Thursday, October 14, 2004 1:19 PM
> > To: comp-protocols-dns-bind at isc.org
> > Subject: Re: DNS Failover
> > 
> > bind-users-bounce at isc.org wrote on 10/14/2004 12:40:24 PM:
> > > Pete,
> > > 
> > > It is still the case that our GSLBs work pretty well. I'm not just 
> > saying that
> > > from the point of view of using my own browser. When we use the GSLBs
> > to 
> > 
> > > redirect traffic for our sites, we can see that the traffic does
> > indeed 
> > get redirected. 
> > > 
> > > For example, one of our sites gets over a million hits per day from
> > all 
> > over 
> > > the world. It gets over one million monthly unique visitors. When the
> > IP 
> > 
> > 
> > To Pete's point:  As your hits are coming from a large number of unique 
> > hits, that implies unique brower sessions, and therefore implies that
> > that 
> > your users will not have cached information from your glsb's and this
> > type 
> > of traffic might not be problematic.
> > 
> > However, the ones you have to worry about are those people that are not 
> > unique and have a browser cached answer.  These people will have the 
> > problem that Pete describes. 
> > 
> > With the large number of unique connections your site takes you can
> > expect 
> > to see the results you have described.  As will all technical caveats,
> > it 
> > is important to evaluate your particular circumstances and determine the
> > 
> > potential impact to your situation.
> > 
> > 
> > Just my $.02.
> > 
> > 
> > Thank,
> > 
> > Dave...
> > 
> > 
> > > address handed out for that site by the GSLBs is changed, the change 
> > takes 
> > > effect in the traffic to the site extremely quickly and effectively.
> > So 
> > that 
> > > gives a good kind of average over the various different browsers and 
> > resolvers
> > > throughout the world.
> > > 
> > [clip...]
> > 
> > 


More information about the bind-users mailing list