[lug] Clustering for Load-Balancing and Fault-Tolerance??

[Sean Reifschneider]

On Wed, Jan 30, 2002 at 07:18:57AM -0700, Nate Duehr wrote:
>DNS is perfectly fault-tolerant when built correctly.  Your first comment
>says that clients are left high and dry when a server goes down... yes, but

In theory DNS is perfectly fault-tolerant...  In practice, it is not,
because of clients that don't do what they should.  Until most clients get
resolvers that are smarter, having DNS servers that are more robust is just
a fact of life for most ISP-type clients.

>Not 100% true.  Load balancing and IP failover can't rescue a half-finished
>(sent but not ACK'ed) DNS query at all.  If your clients do a lot of DNS

Not 100% true...  The load-balancer could watch the DNS requests and
re-submit a request which has not been responded to in an appropirate
time...  Usually, the client resolver will do that for you, though...

Load balancers and failover tend to work quite nicely for DNS, because it's
not the 60 seconds of unavailability that hurt...  Most users won't make a
call if it clears up in that period of time.  Considering the general
reliability of DNS servers, that 60 seconds happens pretty infrequently.

The problems that the users notice are when the name server is freaking out
for an hour or two that it takes for somone to notice the name server is
down (usually because of an unusually high number of calls from users about
it), and to get it corrected...

DNS is a great service to fail over...  It's relatively easy and you don't
have to worry about already open TCP connections -- most requests are UDP
and the TCP requests usually aren't urgent...  While you can easily
load-balance them, ideally you want a shared cache between all
load-balancers...

Sean
-- 
 Peppermint Patty gets a DSL line in "YOU'D TELL ME IF YOU WERE IN A GERMAN
 SCHEISSE VIDEO WOULDN'T YOU, CHARLIE BROWN"
Sean Reifschneider, Inimitably Superfluous <jafo at tummy.com>
tummy.com - Linux Consulting since 1995. Qmail, KRUD, Firewalls, Python