Is you're requirement just failover or also high throughput? I would
NOT use windows clustering for the other adapters. It will only slow
them down. If they want fault tolerance for HTTP receive then use a
load balancer. Except for the adapters you specified, all the in the
box adapters are made to be hosted in host instances on multiple
BizTalk servers. This creates a logical cluster. The reason this
works for BizTalk is because it's all message based, so everything is
going in and out of the message box databasse (which is the thing you
really need to make sure is fault tolerant and configured properly).
Even SSO doesn't need to be clustered unless your operational
requirements are that you can change configuration at any time, you
can just add another SSO server that is like a secondary server. I
know a lot of network / operations people are all crazy about MSCS,
but it really only keeps one server running at a time, so if you have
high throughput requirements it will not cut it for you.
You want every host to have two host instances at least (except the
clustered adapter ones) and you want SSO setup with a master and a
secondary. Basically partition your hosts based on your message and
processing profile. If you have four servers and 75% of your traffic
is HTTP receive and 25% HTTP send then you want the host for HTTP
receive running on at least three of those four instances and send on
two (because you do want the high availability so one wont do it).
Also take into account the host that will do the Orchestration
processing if you have any, run this instance on multiple servers
too. If you have a very even load distribution profile (like all
parts are equally likely to be used) then I would just put a host
instance on every server for each host. Please remember, I'm talking
about the BizTalk concepts of Host and Host Instance. If you need
more info you can send me a message.
Kind Regards,
-Dan