[CentOS] High Availability using 2 sites

Thu Jan 5 21:03:50 UTC 2006
Bryan J. Smith <thebs413 at earthlink.net>

Benjamin Smith <lists at benjamindsmith.com> wrote:
> I've been exploring high-uptime availability solutions for
> our own, database-driven ASP. We have two sites, much as
> original poster describes, and 5-minute DNS, but many
larger
> providers (EG: SBC, AOL) have DNS servers that seem to
ignore 
> TTL.

Apparently many others here serve content to users on
networks other than those on AOL, SBC, etc...  Or at least
their comments seem to repeat that.  ;->

It's gotta be either that, or it's the reality that they keep
testing when their servers are using DNS that talk directly
to (or are the) authority for the domain.

I suspect the latter.  ;->

> So, I've been at a quandary on this very same issue. We had
> a problem about 2 years ago where we had to switch to the
> failover in an emergency. From our end, we were "back up"
in
> < 3 hours, (far less than the 6 hours allowed by contract)
but
> it took over 48 hours for availability to approach 100%,
due to 
> the aforementioned DNS issues. (I hate you, SBC!) 

Which is why you need 1-to-1 NAT for near-immediate uptime.

Of course, that doesn't help you if the provider of those IPs
can't reach your 1-to-1 NAT equipment.  That's why it's not a
true failover.

An ideal, although bandwidth using, solution is to keep your
router/1-to-1 NAT equipment at different locations.  E.g.,
you have 4 sites -- 2 router/NAT, 2 servers.  Only if and
when you lost 3 sites would you go down.

But that gets mighty expensive.  Which brings us to the next
concept ...

> So, do you know of a "getting started" for how to get an
> autonomous system number and run BGP? My skills as a
network
> admin are a distant second to my primary skills... 

In the US, start with the authority, ARIN --
http://www.arin.net/  They will give you the "do it yourself"
cost.

I suspect you're below that, so you need to talk to your
provider(s) about a solution.  Unless you have just 1
provider (which means you're putting all your eggs in their
backbone basket), it's pretty tough to do without ARIN at
some point.

That's why it's typically better to rely on a partner who
already has their own AS, and ties into 3+ providers.  Again,
this isn't something that you can do on your own, unless you
have a lot of dough.

Again, this is where the small-to-medium ASP finds
him/herself at the point where they either have to make a
major investment to go bigger, partner to go bigger (although
they will always be smaller than the partner), etc...  It is
*NEVER* something you can do in software, and that's the
chronically dead wrong assumption.


-- 
Bryan J. Smith     Professional, Technical Annoyance                      b.j.smith at ieee.org      http://thebs413.blogspot.com
----------------------------------------------------
*** Speed doesn't kill, difference in speed does ***