----- Original Message -----From: Jim Frentress
Below I assume the application entry-point is a servlet, even though the
concept applies for any entry-point above the AppServer.
Am I correct to assume that vendor-specific clustering cannot handle session
fail-over when HTTP services are provided by anything but the clustering
AppServer? This seems obvious but I just want to clarify my understanding.
Not necessarily the case. It's possible for the app server to plug
into the front line HTTP servers in a way that allows it to provide
load-balancing and failover in the servlet engine. Consider two front line
HTTP servers, H1 and H2. Using DNS, you can map the address H1 and H2 to a
single host name H. When H is used in a URL, the browser may send the
request to either H1 or H2 (DNS round-robins between these addresses on
each lookup). The http server then forwards the request to
one of the app servers using the plug-in. The plug-in may have
smarts to load-balance and fail-over between the app servers. To
allow for fail-over, the app servers must ensure that session state is available
from more than one.
This is exactly what the most recent WebLogic cluster solution does
(4.5 beta 1). Plug-ins are provided for NSAPI and IIS.
Note that there are better solutions for fail-over and load-balancing
between the browser and front line HTTP servers like LocalDirector which
redirect one IP address to multiple servers. While DNS is sufficient, a
redirecting solution can provide more control over load-balancing and faster
fail-over.
Cheers, Eric
Eric Halpern
BEA/WebLogic
