[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: www load banlance
- To: misc_(_at_)_openbsd_(_dot_)_org
- Subject: Re: www load banlance
- From: Hal Snyder <hal_(_at_)_vailsys_(_dot_)_com>
- Date: 05 Sep 2000 17:55:01 -0500
David Terrell <dbt_(_at_)_meat_(_dot_)_net> writes:
> On Tue, Sep 05, 2000 at 10:36:50AM -0400, Adam A. Newman wrote:
> > moreover than that, if the nslookup gets cached in more commonly used
> > nameservers such as a that of a large dial (or dsl, or cable,
> > etc.) provider, then all http requests will go to the down or up
> > ip depending on which one the NS gave out.
>
> That's simply not true.
>
> 1. Multiple A records ("round robin DNS") are sent to caching
> nameservers and on to clients.
> 2. Clients _should_ make a best effort to try multiple A records
> if the first is unreachable, though clearly this is implementation
> dependant.
If you control DNS, you can set a short TTL to improve failover
response. I believe Eddieware defaults to DNS TTL of 120 sec for this
reason. This assumes you have some mechanism to delete failed systems
from DNS as they are detected.
More info on this sort of thing, including link to ISC's statement, at
http://www.stanford.edu/~riepel/lbnamed/
Visit your host, monkey.org