----------------------------- A UNIX Server Is NO Mainframe ----------------------------- Source: Datamation magazine Copied from: http://www.datamation.com/servr/AUNIXServer.html (The above link is no longer working...) As enterprise servers, UNIX systems still fall short of mainframes in some areas. Measuring the gaps will help you put the right apps on the right platform. By David Simpson The mainframe mantra used to be RAS, for reliability, availability, and serviceability. Those are the areas where Big Iron held distinct advantages over smaller hardware platforms. More recently, mainframers have lengthened the mantra to something like RASSIM, adding scalability, integrity, and manageability to the mix. Client/server aficionados, however, are willing to trade some of those benefits for the flexibility of open systems. They think Big Iron belongs on the scrap heap. These IS managers have relied on UNIX servers as their dominant mainframe alternative, despite some drawbacks. UNIX enterprise servers fall short of mainframes when it comes to availability, manageability (including security), and possibly scalability. An increasing number of IS managers are now choosing Windows NT for their server solution (see Users Are Adding Windows and NT to Their Enterprise Client/Server Mix, December 1, 1995). But, all of the potential shortcomings of UNIX used for enterprise servers apply equally well--or more so--to NT. UNIX vendors are narrowing the gaps. But, in the meantime, understanding the differences between UNIX and MVS environments will help you decide which applications to leave on the corporate mainframe and which to off-load to distributed systems. AVAILABILITY UNIX vendors are using clustering to achieve more availability (see Can't Tolerate Server Downtime? Cluster 'Em! August 15, 1995). Mainframe vendors generally claim almost 100% availability (including scheduled and unscheduled downtime). IBM, for example, claims 99.9999% availability for its Parallel Sysplex architectures, which averages out to a few minutes per year. No UNIX system can claim as much, but adding RAID devices, clustering interconnects, and software can bring a UNIX system close to 99.99% availability. Even so, downtime for UNIX platforms can amount to as much as an hour or more per year. For many applications, that's not a big difference, but in 7x24 shops it can be significant. Studies have shown that downtime can cost on average $1,400 per minute and, in mission-critical 7x24 operations, downtime costs can soar much higher (see "Downtime Costs"). As a result, many IS managers simply opt to keep applications that require very high availability on their mainframes. Analysts advise IS managers intent on moving mission-critical apps over to open systems to take a close look at how their particular UNIX vendor tackles the availability issue. The key, they say, is to ignore reliability claims based on downtime due to hardware failure, which is not the most frequent cause of downtime (see table, " Downtime Causes"). Rather, focus on application availability by factoring in scheduled downtime. Scheduled downtime for software and hardware upgrades and maintenance accounts for up to 25% of all downtime, say analysts. Here, mainframes have the advantage. Via "concurrent maintenance," a mainframe can be shut down without bringing down the entire computing operation. Even with clustered UNIX systems, you may have to shut down systems for the duration of the downtime--or at least suffer noticeable performance degradation. UNIX vendors, however, are working on decreasing scheduled downtime by adding to their clustering schemes features such as rolling software upgrades (which enable software to be upgraded without shutting down the system) and journaled file systems (which allow faster reboots). Although you should consider a variety of factors, in particular don't overlook the switchover times (failover) of clusters. In best-case scenarios, failover will take from 30 seconds to five minutes in clusters with expensive high-speed interconnects. If your applications can't tolerate that much downtime, it may be best to keep them on the mainframe. Also bear in mind that actual switchover times depend on a number of factors, including the size of the database, how quickly the database (as opposed to the hardware) can recover, and the number of transactions that have to be recovered. To boost UNIX clusters up to near-mainframe availability, you'll have to add expensive equipment, including fully redundant components and high-speed, often proprietary, interconnects. Costs for this added equipment can add up. Eventually, it may erase the price advantages of an "open" system platform. Network downtime is another factor that affects availability. Here, again, the gap between clustered UNIX systems and mainframes is narrowing, but IS managers with experience in both environments say that network downtime is often a more significant factor than either system hardware or software failures. Mainframe networks approach the uptime of their connected systems, but in large UNIX-TCP/IP LANs (sometimes referred to as "spray-and-pray" by mainframe bigots), the network becomes an Achilles' heel. According to Charlie Burns, an analyst with the Gartner Group consulting firm in Stamford, Conn., large UNIX LANs often drop to 70% availability. The functionality built into MVS, along with the sophistication of its network-monitoring and management tools, makes it easier to detect and recover from network-based errors. "UNIX-based management tools are becoming better, but they're still rudimentary compared to mainframe management tools," says Burns. Scaling Up Scalability MANAGEABILITY Manageability is another often-cited mainframe advantage. And in fact, the high cost of system management in distributed environments is one of the largest contributors to the unexpectedly steep costs of client/server computing. In a Gartner Group survey, IS managers said system management is the second biggest challenge they face in moving to client/server (overall complexity is first). IS managers point to vendors such as Boole & Babbage (Command/Post), Computer Associates (CA-Unicenter), and IBM (SystemView) as having done a decent job of bringing the functionality and ease of use of their mainframe management products into the open system arena. But, management is still a big problem, especially for a far-flung distribution of clients and servers. Gartner Group studies show that the complexity of managing distributed client/server environments rises geometrically as components are added. In other words, as you double the number of managed devices, you quadruple the complexity--and, to a lesser degree, costs--of managing those devices. Again, the problem is related to the network: While mainframe environments are hierarchical, UNIX networks are more peer to peer, which is inherently more difficult to manage. The system management functions IS managers complain most about in client/server systems are performance monitoring and management, problem resolution, software distribution, job scheduling, capacity planning, and storage management. "For systems management in UNIX environments, we're turning the clocks back to the '70s, although the management software vendors are catching up pretty rapidly," says Dave Dubnick, manager of technology operations at CCH, a Lake Success, N.Y., software and service provider for the legal and tax markets. He says that capacity planning tools for UNIX environments are particularly weak compared with the tools available in the mainframe world. Dubnick advises buying management tools from as few vendors as possible to reduce complexity and the potential of noninteroperability between packages. CCH uses CA-Unicenter for almost all of its system management tasks. One reason why UNIX management tools suffer in comparison to mainframes is that vendors have to write for so many different versions of UNIX. "Systems management tools in the UNIX world are too generalized because they have to run on so many hardware platforms, UNIX variants, and networks. Generalization breeds compromise," says Richard Finkelstein, president of Performance Computing, a consulting firm in Chicago. The bottom-line advice on client/server system management? Don't expect the level of top-down management you had in the MVS world and be sure to factor in rising management costs and additional personnel as you budget for your migration to client/server. SECURITY Within the realm of system management, lack of security in UNIX environments is another big concern among IS managers. Security issues range from the operating system to third-party packages and network vulnerability. Gartner Group's Burns says, "Mainframe prices are dropping fast enough [approximately 30% per year] that a lot of companies are rethinking their planned move off of mainframes because it's just not worth the risks." At the layered software level, there are no products in the UNIX arena that rival the security of mainframe mainstays like IBM's RACF or CA's TopSecret and ACF2. However, some near-equivalent packages are starting to emerge. The UNIX package that comes closest to RACF-level security, according to Gartner Group analyst William Malick, is Memco's SeOS, which is resold by Platinum Software. And CCH's Dubnick thinks the security modules in CA-Unicenter are almost equivalent to ACF2 and TopSecret. A variety of other vendors supply beefed-up security packages for client/server environments, including Bull, Dynasoft, 4th Dimension Software, IBM, ICL, and Raxco's Axent subsidiary. Robert Melford, a Mission Viejo, Calif.-based consultant who specializes in security, says that "you can make UNIX basically as secure as an MVS-RACF environment. It's just a lot more work for systems administrators." Third-party security packages, says Melford, may not actually boost the security of UNIX, but they make it easier to administer, with features such as enhanced audit capabilities and improved account management. Although the lack of security in client/server environments is a problem, bear in mind analysts' estimates that typically only 10 to 20% of corporate data requires restricted access. Then again, sacred cows make the best hamburgers. //