The Mainframe in a Virtualised World

The average utilisation of a Wintel server in a commercial organisation has been widely reported as 10 per cent or even less. Despite this, for a long time increasing capacity in a Wintel or Unix environment has simply meant adding more servers. Many large corporations lost control of the resulting server-sprawl, and the hidden-costs that came with it—people tinkering with the servers or worrying about LAN-backups instead of providing a customer-facing service.


Server consolidation attempts to address the problem of sever-proliferation. Consolidation can include any of the following (in order of increasing benefit): -

* Standardisation (reducing the number of server platforms and configurations used)
* Centralised Management (remote management of multiple systems)
* Co-location (physical migration of distributed servers into central datacentres)
* Rationalisation (collapsing numerous small servers into a larger system).

Co-location has largely consisted of migrating to racks of blade servers closely packed to save on data centre floor-space; however high power consumption and dense packing may require costly cooling systems, so promised TCO savings often do not materialise. Rationalisation gives the most benefit, but is hard to implement, except with homogeneous workloads.

Virtualisation enables differing workloads to be consolidated onto one machine, as each virtual server has its own operating system image, insulating it from other servers, and providing the following advantages: -

* Efficiency—increased system utilisation from running different workloads with differing profiles on the same server.
* Manageability—multiple systems located in the same place.
* Flexibility—make system changes without having to reboot/make hardware changes.
* Facilities and Power—the same levels of performance as physically separate servers can be achieved at a fraction of the space and electricity.

IBM introduced the VM (Virtual Machine) operating system back in the 1960's, but where does the mainframe stand in today's virtualised world? IBM would like to position the mainframe as the ultimate consolidator, but porting a Unix or Windows application to z/OS is not trivial—it is more like a re-write. However, porting to Linux is more straightforward, and it is here that IBM has seized an opportunity. Many organizations wishing to de-emphasize their mainframes cannot just decommission them because they continue to provide services that are simply unavailable in other environments. Moving over to Linux for those things that it does well by creating virtualised Linux partitions alongside existing z/OS partitions gives a smoother and cheaper migration path—especially so if you have excess mainframe capacity, as additional hardware cost will be zero and the software costs zero or very low.

The main strengths of the mainframe are well known: scalability (a mainframe with VM can provide 100s or 1,000s of virtual servers), availability (typical scheduled down time is measured in minutes per year), performance (the architecture was designed top-down with virtualisation in mind), security (RACF has been the “gold standard” for the past 30 years), manageability (there is only one actual machine to manage no matter how many virtual machines you have) and high IO bandwidth (mainframes are happiest handling large amounts of data with heavy I/O content).

These also happen to be the main attributes required in a big web server farm or large-scale e-commerce server. High volume, high transaction rate, large number of users, unpredictable web workloads—these are optimal application characteristics for mainframes.

Set against these is the initial outlay—mainframes tend to be expensive. But a reasonably sized mainframe capable of running 1000 simultaneous Linux machines does not cost much more than a 1000 low-end rack mount Linux system, and also requires less floor space and power. However overall cost is more about people than hardware and facilities; as a rule-of-thumb one person can support about 20 servers, so 1000 servers require 50 people. A mainframe can scale up—more memory, processors, disk capacity—without needing more people, making the cost for large systems quite attractive. Labour costs for setting up a new server is dramatically lower when you are simply creating a virtual machine on your box by pressing a button, and networking is also, when you can coordinate internal communication between your virtual machines via a virtual LAN rather than a physical topology of switches and fibre.

IBM claims hundreds of mainframe customers worldwide running Linux; those who purchased mainframes exclusively for the purpose of running Linux on them number in the scores. New workloads—Java, Linux, ERP and other non-traditional (COBOL/CICS) mainframe applications—now account for 60 percent of IBM's mainframe sales, fuelled by the availability of specialized, low-priced (relatively speaking) mainframe engines. IBM mainframes have had System Assist Processors to free up general-purpose engines from handling system I/O for years. Since 2001 IBM has added engines for Linux, then for WebSphere and Java, and more recently for database processing to offload these workloads from higher cost general-purpose mainframe engines.

The “spoiler” in this nice argument is the appearance of increasingly sophisticated virtualisation solutions in the Wintel market from the like of VMware and Xen, which are beginning to assume mainframe-like abilities. The competition to mainframe virtualisation is unlikely come from plain physical blade servers, but some kind of virtualisation solution that takes advantage of the performance to power-consumption ratios available on the x86 platform.

Without IBM's emphasis on new workloads and special-purpose engines to drive these workloads onto the mainframe, it is likely that its mainframe server business would have shrunk. Special purpose engines have made the mainframe competitive in the place that matters most to IBM: the glasshouses where mainframes still reign supreme.