> Are you saying that if we are using NLB, there is no need to
> cluster CFMX as well?
>
> We will have two separate servers that will use NLB in single
> affinity (sticky session) mode. I am assuming that, as long
> as both CFMX installations are configured correctly, there
> won't be a need to cluster CFMX. Does that sound right?
Well, I won't go so far as to say that there's no need. The way I've set up
NLB has been at a web server level - that is, the NLB IP address listens on
TCP/80 and TCP/443 and redirects that traffic to the cluster members.
JRun clustering generally takes place at the application server level. For
example, for a single virtual web server, you may have two or more CFMX
instances clustered together. You'd connect that cluster to the single web
server.
There are plenty of ways you can mix and match this together. For example,
if you have two physical machines with sufficient resources on each, you
might use NLB to figure out which machine's web server handles an incoming
request. Once the request is sent to that machine, you might use JRun
clustering to have multiple instances bound to that web server, which will
make it less likely that a specific machine will fail and which potentially
may allow you to better use the resources of that machine (at an obvious
price of greater complexity). In that case, you may want to configure
session sharing within the JRun cluster, which has its own issues.
One problem with using NLB alone is that, if CFMX fails, NLB won't
automatically stop sending requests to the machine, since your web server
itself will still be working. You'll have to figure out some way to down the
NLB host yourself. The way I've usually done this is by writing a CF page,
and using a server monitoring tool or script to call that page periodically
and down the NLB host.
Dave Watts, CTO, Fig Leaf Software
http://www.figleaf.com/
phone: 202-797-5496
fax: 202-797-5444
[Todays Threads] [This Message] [Subscription] [Fast Unsubscribe] [User Settings] [Donations and Support]
