NAD 3020 manual High Availability for the Blade Servers

Design Guide

link redundancy combined with a deterministic topology design to achieve application-availability requirements. Servers are typically configured with multiple NICs and dual homed to the access layer switches to provide backup connectivity to the business application.

High availability is an important design consideration in the data center. The Cisco Catalyst Blade Switch 3020 has numerous features and characteristics that contribute to a reliable, highly available network.

High Availability for the BladeSystem Switching Infrastructure

High availability between the Cisco Catalyst Blade Switch 3020s in the HP c-Class BladeSystem and the aggregation layer switches requires link redundancy. Each Cisco Catalyst Blade Switch 3020 in the HP c-Class BladeSystem uses four SFP uplinks for connectivity to the external network, allowing for redundant paths using two links each for more redundancy. Redundant paths implemented between the HP c-Class BladeSystem and each aggregation layer switch when each path uses two links provide a highly resilient design. However, this setup introduces the possibility of Layer 2 loops; therefore, a mechanism is required to manage the physical topology. The implementation of RSTP helps ensure a fast-converging, predictable Layer 2 domain between the aggregation layer and access switches (the Cisco Catalyst Blade Switch 3020s) when redundant paths are present.

The recommended design is a triangle topology (as shown in Figure 4 earlier), which delivers a highly available environment through redundant links and a spanning tree. It allows for multiple switch or link failures without compromising the availability of the data center applications.

These channels support the publicly available subnets in the data center and traffic between servers. The server-to-server traffic that uses these uplinks is logically segmented through VLANs and can use network services available in the aggregation layer. There is also a port channel defined between the two blade-enclosure switches. This path provides intraenclosure connectivity between the servers for VLANs defined locally on the blade-enclosure switches. Clustering applications that require Layer 2 communication can use this traffic path, as well as mirrored traffic. Each of these port channels is composed of two Gigabit Ethernet ports.

RPVST+ is recommended as the method for controlling the Layer 2 domain because of its predictable behavior and fast convergence. A meshed topology combined with RPVST+ allows only one active link from each blade switch to the root of the spanning-tree domain. This design creates a highly available server farm through controlled traffic paths and the rapid convergence of the spanning tree. The details of the recommended design are discussed in a later section.

High Availability for the Blade Servers

The HP c-Class BladeSystem provides high availability to blade servers by multihoming each server to the Cisco Catalyst Blade Switch 3020s. The two Cisco Catalyst Blade Switch 3020s housed in the interconnect bays are connected to the blade server over the backplane. Four backplane Gigabit Ethernet connections are available to every blade-server slot.

Multihoming the server blades allows the use of a NIC teaming driver, which provides another high- availability mechanism to fail over and load balance at the server level. Three modes of teaming are supported:

© 2008 Cisco Systems, Inc. All rights reserved. This document is Cisco Public Information. Page 13 of 28