In my Canute-style determination to hold back the approaching tide of middle-age, I've reluctantly signed on at the local gym, joining a large cross-section of the local business community in the ritual self-humiliation of lunch-hour workouts.
OK, it's not too bad once you get used to it, and at least my gym spares no expense in entertaining the hapless machine-jogger. As I pound away at the treadmill, trying desperately to ignore the dispiriting analysis of my vital signs on the display panel, I'm offered a choice of no fewer than four cable TV screens on which to fix my gaze, and a set of headphones that allows me to skip effortlessly between these four silent worlds at the touch of a button. One of the TV displays generally offers up-to-the-minute news; another provides regular visits to the popular soap operas; another will be disgorging pop videos brimming with energy; and the fourth will probably be showing a black and white 'B' movie from the post-war era.
It struck me the other day - as I was avidly absorbing six and a half minutes of a monochrome Cary Grant epic - how closely that row of TV screens mirrors the application mix of the average data centre. There are the highly graphical pop-video-style apps, 95 per cent presentation, delivering very little, but doing so with great aplomb, and to the sheer delight of the end user. There are the soap opera apps of the e-business world, in a constant state of flux, leaping dramatically between extremes of performance and capacity. There are the news-style ERP systems, crunching through sales and logistical data, if not always in the most exciting manner. And there are the black and white (or black and green!) legacy systems, solid, reliable, and reassuringly familiar. Each type of application displays distinct characteristics, each imposes different pressures on the resources available, and each - just like those blinking TV displays in front of me - fights desperately for attention.
In today's enterprise IT environment, it's the complexity of this mixed workload, its unpredictable demands, and its blurring boundaries, that are creating the biggest problems for system managers and capacity planners. You can scale horizontally, vertically, and probably diagonally; you can throw Mips and bandwidth at the problem, and implement a vast range of I/O performance and database management techniques to speed up the mix. But the real secret of success is effective workload management, and this is where the mainframe still stands head and shoulders above its competitors.
Looking at the recent announcements of IBM's zArchitecture (the z900 new-generation mainframe with its 64-bit z/OS operating system), workload management is key. The feeds and speeds are all very impressive - 2500 Mips on 16 processors, 9 billion transactions per day in a clustered configuration, 24 GB/sec of I/O bandwidth, 2000 SSL transactions per second, etc, etc - and few mainframe users will be running out of steam for the foreseeable future. But the real emphasis is on facilities for distributing workloads more efficiently across the hundreds, or even thousands, of virtual servers running within the system, via dynamic partitioning.
The key technology here is the Intelligent Resource Director (IRD), which pools the combined strengths of the Workload Manager, Logical Partitioning, and the Parallel Sysplex. The new Director manages resource allocation and I/O paths dynamically, controlling priorities according to user-defined rules. The system will also be able to dial out to IBM control centres, report potential failures or capacity restrictions, and effectively upgrade itself on the fly - 'non-disruptive growth', as IBM calls it. The combined effect is an infinitely scaleable pool of system resource, limited only by IBM's imagination and the customer's budget. More importantly, it provides the best solution yet for coping with the workload spikes that characterise so many e-business applications.
The other essential component of the dynamic workload management environment is the new usage based pricing scheme, directed by the License Manager. Again, the emphasis is on flexibility: IBM isn't promising any software bargains here, and at first sight of the options available, the scheme won't be much less complex than its predecessors either. But usage based pricing does finally give users the chance to implement new test-bed applications on the z900, and to allocate resources realistically, without the punitive disparities of processor based charges. As with the IRD, the License Manager monitors peaks in throughput and network traffic, and allows thresholds to float up or down, according to average usage.
Workload management is where IBM aims to differentiate the new mainframe range clearly from the other top-end server offerings. Mips for Mips, there are some devastatingly fast Unix systems in the market, and the competition gets hotter by the day. But the z900 is the latest incarnation of an architecture that has always been uniquely positioned for the management of complex workloads, and never has that been more urgently required than in today's e-business world.
This was first published in November 2000