IBM is out with a new mainframe this week, which naturally led to claims that it is either the greatest thing since sliced bread or the final death gasp of a technology giving way to commodity systems.
First, the facts. The system is called the zEnterprise and it runs on the new z196 core consisting of 96 5.2 GHz microprocessors, a configuration that produces some 50 billion instructions per second. It also sports a new software stack optimized for heavy workloads and can be supplemented with more than 80 specialty engines, namely the System z Application Assist Processor (zAAP) for handling Java workloads, the System z Integrated Information Processor (zIIP) to handle general processing needs, and the Integrated Facility for Linux (IFL) to take on Linux workloads.
As can be expected, longtime IBM partners were quick to back the system. BMC Software, for one, came on board quickly with support for its suite of business service management solutions designed to simplify the provisioning and management of physical and virtual resources. As a cross-platform management stack, BMC said, the package should easily accommodate zEnterprise users regardless of which operating system(s) are deployed.
For those running IBM's Information Management System (IMS), a new offload package from NEON makes it even easier, and cheaper, to port those workloads with the zAAP or zIIP processors. The company is making its zPrime for IMS software available for only $1 per year with a two-year commitment. The company's zPrime 2.1, meanwhile, is needed for other workloads, such as DB2, CICS, TSO/ISPF and batch processing.
But does mainframe processing fly in the face of industry trends, both past and future? That's the take from industry.bnet.com's Erik Sherman, who argues that the reason IBM's mainframe sales are slipping is that enterprises are not buying the argument that big iron is the right approach for the cloud. After all, the cloud is all about scalability and matching resources to data loads. When processing is light, you can consolidate the load onto fewer servers and shut the rest off. With a mainframe, you can consolidate on cores, but the whole machine keeps running. And if peak loads start to press against available capacity, it's a much better value proposition to add a few servers than to power up another mainframe.
The problem with that argument is that it assumes the cloud will be rife with extreme fluctuations of data, calling for constant hardware reconfigurations. The fact is that major multinationals are likely to see a constant minimum data requirement that will more than satisfy the capacity of a mainframe or two, even when broken down on regional bases. For that core data load, buying and running a mainframe is still cheaper than a warehouse full of commodity servers.
Does that mean the mainframe market will continue to decline? Probably. But for a core group of Fortune X companies, it will still be the best deal around.