On Tuesday 25 September 2007 11:49, Rick Litton wrote: > My first concern is to ensure reliability by avoiding an SPF > (single point of failure). Also, I'm not so > sure that a single HttpService service bundle can do the job, i.e. it > doesn't become a bottleneck.
First, I would use a DNS trick. Set the up a a zone, where each redundant GW machine is the NS, and that the record for the GW to the client points back to the same machine. I would probably embed Apache Directory Server to serve that DNS zone (see below). Let's say; app.ops4j.org is the entry point of the system. The machine app1, app2, app3 are the redundant machine. The IP numbers of those are the NameServers for the app.ops4j.org zone, and the @ entry in each machine is listing itself. This brings you; 1. If a machine dies completely, the DNS client are required to try the next NS. So on DNS lookup, you will always get to a working machine. 2. Since Apache Directory Server sits in the same JVM, if the JVM dies the DNS goes down with it, and point 1 will apply. 3. Automatic load balancing of the clients to each of these machines. DNS spec states that the client must choose a NS at random. Secondly, I would try to cut out these machines from the loop of normal requests if possible, so that long running sessions are going to fail due to crashes in this machines. The actual approach is of course very dependent on the requirements, the application type and other factors. Thirdly, I would hire my team to do the job ;o) Cheers -- Niclas Hedhman, Software Developer I live here; http://tinyurl.com/2qq9er I work here; http://tinyurl.com/2ymelc I relax here; http://tinyurl.com/2cgsug _______________________________________________ general mailing list general@lists.ops4j.org http://lists.ops4j.org/mailman/listinfo/general
