The same day that Computer Associates International Inc revealed its plan to get into the multiplatform clustering software market in conjunction with Tandem Computers Inc, (see other top story today), IBM Corp’s RS/6000 division began talking-up a next-generation clustering technology of its own, code-named Phoenix. IBM says its new high-availability software will eventually allow up to 128 nodes to be clustered to work together as a single logical system, claiming it will make availability scalable and transcend anything in the market today. Phoenix will begin life as a mechanism for tying up to 32 parallel RS/6000 SP or other AIX nodes together – up 128 in 1997 – but has already been demonstrated running on OS/2 and Windows NT and will be ported to other vendors’ Unix environments in addition. The company says it will disclose further details of its Phoenix porting and partnering plans in around three weeks time. It’s also currently evaluating how to extend the offering to all IBM server platforms, including S/390 mainframes from which some of the Phoenix technologies have been borrowed. In addition to providing traditional high-availability (HA) resources such as application failover to alternative systems in the event of node failure, IBM says Phoenix will allow managers to monitor operating system, middleware, communications, global storage resources, load balancing and other kinds of events it says current mechanisms do not address. Phoenix, a multi-year development of RS/6000 division’s SP parallel systems group and IBM’s research division, includes a new set of HA application programming interfaces which will extend, then subsume and eventually supersede the current High Availability Cluster Multiprocessing for AIX (HACMP) software the company offers for clustering up to eight SP or RS/6000 nodes. Phoenix technologies will extend HACMP for AIX to support up to 32 nodes before the two technologies are combined later next year. HACMP interfaces will be supported in Phoenix and therefore allow customers to run existing HACMP applications unchanged. The company’s currently looking at how external nodes attached to SP parallel switches can be viewed and managed from within an SP cluster using Phoenix.
