Re: [cobalt-users] Failover and the Linux High Availabilty project

[Kris Dahl <krislists@xxxxxxxxxxxxx>]

on 6/28/00 4:32 PM, Chip at chip@xxxxxxxxxx wrote:

> Hi folks.
> 
> Has anyone actually implemented the Linux-HA code on a Raq?  From a brief
> glance, it appears that it would work, but it would be nice to know if
> someone out there has already done it  and could share their experience
> with it :)

I haven't done any clustering with Cobalt Machines.  Honestly I think that
its not really that cost effective... In most of the applications that
require that sort of availability, I would rather go with machines that have
RAID, etc.

> I looked into the redundant solution offered by Cobalt, but it apparently
> involves having the two Raqs close enough to each other that a serial cable
> can be run between them... which defeats my intended purpose of having the
> redundant Raqs in different NOCs.

Honestly having servers in different NOCs is *completely* overrated in my
opinion.  What are you protecting yourself against?  Natural Disaster?  Even
then your machine would likely remain online (depending on the nature of the
disaster.

If you get hosted in a decent data-center that has BGP routing with
redundant carriers and routes, you essentially are in two NOCs at the same
time.

Having them in different location puts a pretty big strain on the bandwidth,
and there is no real easy (and effective) way to replicate database servers
over (realatively) low speed connections.  Replication works much better on
ethernet.

For the best bang for your buck, I would just get a single server with
decent availibility and reliability features--something like a Sun Netra, HP
LPr, or IBM Netfinity.  Get it with RAID 5 or 50.  And I'd put it in a
high-end datacenter (like InterNAP, or perhaps Exodus).  If you need to, buy
a couple of the servers and put them behind like a F5 load balancer.  This
is the way it is done... I just haven't seen that many people trying to put
low end servers in multiple locations and trying to load balance them.
Although I suppose you could do it with like an DNS based load balancer (F5
makes one).  But you still have the replication problems, and with most
sites being heavliy database driven, you'll need that replication to be in
real time.

-k