When I was young, I thought scissors came with the tips broken off … I learned later in life that actually they had been broken off by my dad when he used them as screwdrivers … an early lesson in using the right tool for the job. When we started the process of defining the ClearPath NextGen architecture we knew we needed to preserve the partitioning capabilities from earlier systems and at the same time we wanted to exploit all the capabilities in Intel’s latest system designs.. While virtualization is all the rage, we chose not to use a general purpose hypervisor for partitioning … did we choose wisely?
Partitioning and virtualization is not the same thing. General purpose hypervisors like Xen and VMware are tools designed to facilitate consolidation, they are general purpose in nature and optimized to save cost by minimizing the number of physical servers in an enterprise. They are complicated because they are generalized, and designed to maximize the use of the hardware. They are feature rich because they try to address a large number of potential use cases, further they try to mitigate the uncertainty of the real world through flexibility. All of this comes at a cost … most of it in the management of the environment. Virtualization has created a whole new IT discipline … the management of virtual servers.
ClearPath has always been about simplicity in operation and efficiency in design. We spent a lot of time evaluating various hypervisors … ones you have heard of and several that you probably haven’t. After careful consideration we decided to design and develop our own hypervisor layer to embed in the ClearPath processing complex. This decision wasn’t taken lightly, but in the end it was clearly the best way to protect and extend the ClearPath architecture.
You have to remember, the requirement was to provide a way to partition a large server so that specialty engines and the ClearPath central processing complex could operate side by side, isolated from one another and tightly integrated at the same time to deliver predictable performance and fault isolation … something no commercial hypervisor can deliver. It was imperative that processing load in one partition could not impact the processing load in another. It was imperative that faults in hardware or software in one partition could not impact, in any way, the operation of another partition. These are guiding principles for ClearPath that we simply were not willing to compromise. This technology would come to be known as Unisys Secure Partitioning, or sPar for short.
sPar is a type 1 Hypervisor, one that runs on the bare metal, that provides us with a flexible “soft backplane” so to speak that allows us to create arbitrary numbers of partitions of differing and arbitrary sizes optimized for specific purposes. In the Libra 4100 we deploy four partitions … one for the MCP Operating System and three for specialty engines. To deliver the ClearPath computing experience sPar creates partitions with dedicated rather than shared hardware resources at the core, socket, IO port or slot level. In this way we can isolate faults and guarantee consistent and predictable performance, In today’s modern system designs it isn’t enough to just supply the right number of CPU cores, which cores are used and where they are located in the system in relation to each other can have a significant impact on performance … changing the mix each time the system is started, or dynamically as the system is running is unacceptable as performance will not be predictable.
We are frankly quite pleased with sPar. It performs well and allows us to deploy inter partition services designed for specifically ClearPath in ways we never could have with a commercial offering, in this way it advances the ClearPath Heterogeneous Multiprocessing (HMP) architecture to the next level. With sPar we can deploy entire HMP environments on a single processing complex and at the same time greatly simplify the management of the system. What started out as a way to simply partition a system has become the foundation technology for all ClearPath systems to come. So, did we get it right?