Network Computingwww.networkcomputing.com

High-Availability Options

One of the most important metrics for measuring the ability of any network component, including network operating systems, is uptime. The NOS must not only provide access to data but also ensure that the data is available. Third-party market support for HA (high-availability) applications for a variety of NOSes has been strong. For example, both Octopus Technologies and Vinca Corp. offer solutions for Windows NT and NetWare, and a host of HA solutions for Unix exist; Veritas Software Corp. makes a Windows NT version. But a significant trend is the integration of high-availability a pplications into the NOS by NOS vendors themselves. Why? High availability and scalability are natural tasks for the operating system and should be integrated into it.

Both Hewlett-Packard Co. and Sun Microsystems offer somewhat comparable HA solutions. HP's Service Guard works with up to eight nodes, and offers N+1 failover, real-time addition and subtraction of nodes and a full-featured GUI via HP OpenView, which not only manages HP Unix clusters but NT clusters as well. A separate product technology that works with Service Guard is HP's PRM (Process Resource Management), which manages resource allocation to various processes running on the server.

PRM lets a system administrator allocate resources to different applications. If Server A were running a database application and Server B were running an HTTP/CGI application, and Server B's application failed over to Server A, PRM on Server A would be set up to prevent the HTTP/CGI application from consuming too many resources on Server A and adversely a ffecting performance of Server A's database application.

Sun's HA product, Sun Cluster 2.0, scales to four nodes, as opposed to HP's eight. One of the drawbacks of Unix HA solutions is the need to manually create scripts that manage the failure process. Sun has come up with a set of scripts already designed for applications like Netscape Communications Corp. servers and database servers from Informix Software, Oracle Corp. and Sybase. Sun says it plans to extend its clustering solution by offering a global cluster file system, devices, networking and process management by late 1998 or early 1999.

For its part, Microsoft Corp. wants to make NT HA and clustering as common as RAID systems. Microsoft's first iteration of Clustering Services (MCS 1.0) offers only two-way failover but in our tests, did work reasonably well. A significant issue with MCS is the lack of MCS-developed applications. The second phase of MCS, due by either late 1998 or 1999, is likely to offer clustering for scalability.

Novell h as had an HA product, SFT III, on the market for years. It was not a very robust product because it was limited to two-node failover, required a specialized interserver connect and had limited application support. But this will change with the arrival of Wolf Mountain, which Novell plans to release in a product code-named Orion. We saw a demo of a basic Java application running on a 16-node cluster at the BrainShare conference in Salt Lake City last year. Wolf Mountain offers both high availability and scalability. The 16-node cluster was a single system image on the network, allowing users to log into the cluster. Orion will use a shared disk farm configuration. The disk farm will connect to each server in the cluster via Fibre Channel or SSA. Orion will support all of IntranetWare's core services, such as file and print and NDS. Orion will require NetWare 5 as the base operating system. For now, though, IntranetWare users will have to rely on third parties, such as Octopus or Vinca, for HA solutions. As wit h Microsoft and Sun, an API for independent software vendors will be available.