Availability Guide for Application Design

Overview of Server and Network Fault Tolerance
Availability Guide for Application Design—525637-004
2-7
Parallel Hardware Components
The number of channels per K-series processor depends on which HP NonStop
system you are using. Refer to the appropriate server description manual for details on
the model of HP NonStop system that you have.
Fault-tolerant management of devices on an S-series or NS-series server makes use
of the inherent parallelism of the dual ServerNet fabrics to provide two data paths
between a processor and an I/O device. A dual-ported ServerNet adapter or ServerNet
addressable controller can provide access to both ServerNet fabrics for any I/O device
and more than one I/O port for dual-ported devices.
Fault-tolerant management of devices on a K-series server makes use of dual-ported
controllers as follows:
•
Individual devices are connected to channels by device controllers.
•
Each controller has two ports that allow it to connect to two I/O channels. Each of
these I/O channels must connect to a different processor.
At any given time, one of the two processors manages the controller and is said to be
its owner. If the managing processor fails, the other processor takes ownership, and
management of the controller continues.
Dual-Ported Devices
Critical devices have two ports that enable them to be connected to two different
S-series or NS-series ServerNet adapters or ServerNet addressable controllers, or to
two different K-series device controllers. If an adapter or controller fails, then access to
the device can still be made using the other adapter or controller.
On an S-series or NS-series system, dual-ported devices protect against loss of
access if a ServerNet fabric fails. In such instances, the device has access through the
second port to the alternate ServerNet fabric, thereby ensuring its availability.
Mirrored Disks
Mirrored disks are a specific example of a dual-ported device. A mirrored disk is a pair
of physically independent disk drives containing a mirrored volume. Each disk drive is
usually attached to two separate controllers. The mirrored volume is accessed as a
single device and managed by a single I/O process.
Reading and writing works as follows on mirrored disks. All data that an application
directs to the volume is written to both the primary disk and the backup mirror disk.
Since both disks contain identical information, all data that an application reads from a
file can be read from either disk. The throughput of read operations against a mirrored
disk is much higher than for a single disk since separate reads can occur
simultaneously on each disk. Given that most applications result in many more disk
reads than disk writes, mirrored disks result in a significant improvement in overall I/O
throughput.
Operation is interrupted neither by disk failure, nor by repair. If one adapter, controller,
or disk fails, the other remains in service, and processing continues without