The Santa Cruz Operation is introducing a version of Non Stop Clusters for Intel boxes that doesn't need Compaq's proprietary ServerNet interconnects.
Needing only two 100Mbps Ethernet adaptors to enable heartbeat communications, and able to run on bog standard white box PCs, this latest version of NSC is probably the cheapest real datacentre class clustering that we've seen on the hunble PC.
SCO is asking $7799 for a two-node bundle of UnixWare NSC 7.1.1+IP. UnixWare NSC is currently sold by Compaq and licensee Fujitsu Siemens for clusters of up to six nodes (that's 48 CPUs using standard Intel Profusion eight-way boxes) but ServerNet is mandatory. No more, though.
SCO does recommend that larger configurations than two nodes ought to use ServerNet switches. As before, NSC provides a single system image for administrative purposes, and requires no changes to applications. For example, Oracle 8i need only be kicked off from a custom script to become cluster aware. This doesn't apply to Oracle Parallel Server, which isn't supported by UnixWare NSC.
It marks something of a change of heart for Compaq, too. We wondered if Q hadn't let one of its crown jewels slip into the hands of box shifters? No, SCO says, there'll be joint promotional activity and it widens the potential base for NSC.
By the end of the year UnixWare NSC will support up to 16 nodes (128 CPUs) and Gigabit Ethernet. The new two node, IP-based NSC is due to ship on July.
Despite some slips in its roadmap, UnixWare remains way ahead of other Intel clusters. Microsoft looks like leaving Wolfpack, aka Microsoft Cluster Server, where it is on the mortuary slab, and joining Sun in an SMP scaling race. Windows 2000 DataCenter Edition will scale to two or four nodes. This isn't the heady stuff it was promising four or five years ago.
What's worse is that Oracle has no intention of producing a cluster-aware database for Windows 2000, and Microsoft's own applications are only just getting round to be cluster aware, and only in parts.
Open source Linux clustering for real database and TP jobs abounds, with SGI and SuSE co-ordinating work on porting SGI's FailSafe to Linux. And work on a shared-everything, VAX-like distributed lock manager should bear fruit towards the end of the year. This will eventually be the lynchpin for cluster file systems, unless FailSafe - which is also based on a DLM - hasn't scooped up the mindshare by then. ®