Availability Guide for Application Design

Improving Availability on the Internet

Availability Guide for Application Design—525637-004

3-6

Standards and the Network

NonStop systems have a variety of adapters to support a variety of network standards.

Besides the very efficient Ethernet adapters that are able to operate at speeds up to 1

Gb per second, HP continues to support adapters that can handle asynchronous,

bisynchronous, synchronous data link control (SDLC), and several unusual protocols

and speeds.

The most common current standard is TCP/IP v6, although many people still use v4.

However, there are several aspects of TCP/IP implementation that are distinct on

NonStop servers based on its parallel processing architecture.

All NonStop systems feature a parallel processing architecture and the TCP/IP v6

stack takes advantage of that multiprocessing environment. The software architecture,

called Parallel Library TCP/IP, tunes TCP to high efficiency within the NonStop system

architecture.

Routing intelligence is built into the Ethernet communications adapters, so the

adapters can dynamically direct traffic directly to the CPU that is supposed to receive

it. This enables you to connect a ‘fat pipe’ to the system without totally overloading the

first processor that handles that pipe.

Ethernet adapters can be configured with backup adapters so that if one fails, the

"alternative adapter" automatically picks up the related traffic without dropping

sessions. In fact, the architectural design utilizes "active backup," that is, both adapters

carry their full normal bandwidth. In failover mode, the surviving adapter would have to

run degraded for a time. Adapters can be swapped out while the system continues to

run, and the system will recover to the original active-active mode without operator

intervention.

Finally, there are means for customers to define subnetworks within their overall

communications network. All the traffic with a particular set of IP addresses could be

designated Network A and thus be routed via a number of adapters directly to

Application A. Similarly, other traffic could be designated for Network B and routed to

Application B. Think of a system with several different applications on it. You'd like to

keep the finance data separate from the inventory data, which would be kept separate

from the test data, and so on. This capability, called Logical Network Partitioning

(NLP), has many applications in enterprise systems.

For example, in a popular healthcare application, the enterprise wants to isolate patient

care traffic from administrative traffic (for example, billing, insurance). With NLP, it can

create several logical networks for patient care (for example, doctor records, lab,

surgery, intensive care, and so on) and separate logical networks for admissions,

billing, logistics, and so on. All the while, the devices on such a system may be on one

or more physical local area networks and the applications do not have to be concerned

about the subnetworking. It is transparent to them, but it provides additional data

access isolation and protection (for example, logistics cannot access medical records

information).