NonStop NS14000 Series Planning Guide (H06.13+)

D NonStop NS14000 Series System Architecture
Integrity NonStop NS14000 series systems employ a unique method for achieving fault tolerance
in a clustered processor environment, utilizing Intel Itanium microprocessors without cycle-by-cycle
lock-stepping. Instead, two or three microprocessors run the same instruction stream concurrently
in a loose lockstep process. In loose lockstep:
• Each microprocessor runs at its own clock rate.
• Results of each command execution are compared on processor output to the ServerNet fabric.
• Error recovery and minor indeterminate processing results from one microprocessor do not
cause output comparison errors.
If the output of one microprocessor is incorrect, it is discarded, and its source microprocessor is
taken offline for error handling and correction. The remaining functional hardware continues normal
operation until the errant microprocessor is again operational. At that time, all microprocessors
are synchronized and then proceed with executing instructions.
NonStop NS14000 series systems use modular hardware implemented in enclosures with fiber-optic
cabling between these enclosures. These enclosures reside in 19-inch modular cabinets.
NonStop Blade Complex
The basic building block of NS14000 series systems compute engine is the NonStop Blade Complex,
which consists of two or three CPU modules called NonStop Blade Elements. Each NonStop Blade
Element houses two or four microprocessors called processor elements (PEs). A logical processor
consists of one processor element from each NonStop Blade Element plus a logical synchronization
unit (LSU). Although a logical processor physically consists of multiple processor elements, with
each element contained in a separate enclosure, it is convenient to think of a logical processor as
a single entity within the system. Each logical processor has its own memory and its own copy of
the operating system, and it processes a single self-checked instruction stream. NS14000 series
logical processors are usually referred to simply as processors.
All input and output to and from each NonStop Blade Element goes through a logical
synchronization unit (LSU). The LSU interfaces with the ServerNet fabrics and contains logic that
compares all output operations of the PEs in a logical processor, ensuring that all NonStop Blade
Elements agree on the result before the data is passed to the ServerNet fabrics.
A processor with two NonStop Blade Elements (NSBEs) and their associated LSUs makes up the
dual modular redundant (DMR) NonStop Blade Complex, which is also referred to as a duplex
processor. This duplex processor provides data integrity and system availability that is comparable
to NonStop S-series systems but at considerably faster processing speeds.
Three NonStop Blade Elements plus their associated LSUs make up the triple modular redundant
(TMR) NonStop Blade Complex, which is referred to as a triplex system. The triplex processor
provides the same processing speeds as the duplex processor but also enables hardware fault
recovery that is transparent to all but the lowest level of the HP NonStop operating system (OS).
This diagram provides an overview of the modular NS14000 series system and shows one NonStop
Blade Complex with four processors, two VIO enclosures (the I/O hardware), and links to external
I/O and storage:
NonStop Blade Complex 103