Most Hypervisors such as VMware, Hyper-V and Xen work as a thin layer on the server, carving up system resources among multiple partitions or virtual machines, each running their own copy of an operating system. The typical small system operates at less that 10% utilization of the hardware, so combining multiple workloads on a single server allows for much better utilization.
If all the operating systems under the hypervisor are independent, this is an excellent way of increasing the utilization of the server as a whole, at the expense of some overhead for the hypervisor, and some increase in latency for shared resources such as I/O. There are some issues of performance and security management in this partitioning model, but where the virtual systems are small compared with the physical system, it performs well. Overall the utilization can be improved by 300% to 600%, as there is a cap on very high utilization because of overhead, locking and latency constraints. External software such as VMware’s control center can manage the overall environment.
However, what if the workloads are not independent, but part of an overall workload? Examples might be clustered database or specialty engines. In these cases the systems have to communicate with each other directly, and they have to determine the priorities and management themselves.
The traditional partitioning model described above does not work nearly as well in this case. The virtual operating systems and sub-systems understand far more about the workload, utilization, and prioritization than any external workload manager ever could.
The approach that Unisys has taken with the Unisys ClearPath Libra 4100 mainframe built on an Intel platform is much more suited to a large single systems environment. Figure 1 shows the architecture, which allows the MCP to run in a large partition, and control the execution of the specialty engines. These engines can be Java processors ( JProcessor), ePortal processors or encryption (crypto) processors. They run with in a very skinny Microsoft or Linux kernel, under the control of the MCP OS.
This approach allows the MCP to control full security and management of the virtualized engines and deliver the mainframe characteristics of very high utilization, availability and security.
Another example of this approach outside of mainframes is Volume Shadow Copy Service Snapshots (VSS), which works more efficiently in a Hyper-V environment using a "control domain" to take consistent snapshots for all applications on a virtual host.
Action Item: Many workloads will be well suited to the traditional virtualization approach. However, for mission critical workloads requiring strong cooperation between fewer partitions, the approach taken by Unisys is far superior. Infrastructure managers should expect similar approaches from Microsoft (e.g., Volume Shadow Copy Service Snapshots (VSS) allows consistent snapshots in a Hyper-V environment using a "control domain") and Oracle for mission critical workloads.
Footnotes: Link to Unisys sPAR Announcement