1.3. Requirements
Turbolinux Cluster LoadBalancer 10 is used to combine the resources of several computers. The requirements for each of these computers varies according to its function within the cluster. The two main functions are advanced traffic manager (ATM) and cluster node. Cluster nodes are simply systems that provide network services. The traffic manager is the machine that receives all incoming packets and forwards them to the cluster nodes. You will also have backup traffic managers, which will become active only if the primary ATM fails. A system may be configured to function as both a traffic manager and a cluster node at the same time.
1.3.1. Software
To install Turbolinux Cluster LoadBalancer 10 you must be running:
Turbolinux 10 Server
All traffic managers must have Turbolinux Cluster LoadBalancer 10 installed and running on a Linux server running the Turbolinux 10 Server (other operating systems are not supported). Cluster nodes that are not traffic managers are not required to run the Turbolinux Cluster LoadBalancer 10 product. However, it will simplify cluster management if all the systems are running the same operating system and the Turbolinux Cluster LoadBalancer 10.
In addition to the Turbolinux Cluster LoadBalancer 10, you will need to have software providing the services that are to be clustered. For example, if you are creating a web server cluster, each node in the cluster must be running its own web server. This software is not included with the Turbolinux Cluster LoadBalancer 10 product, but many network services are included with most operating systems. For example, Turbolinux Server and virtually every other Linux distribution comes with Apache web server.
1.3.2. Hardware
While Turbolinux Cluster LoadBalancer 10 can be run on modest hardware, such as a Pentium class with 128 MB of RAM, the product is designed to provide high performance. We suggest that you use hardware that fits these high performance needs. The hardware specifications for a traffic manager are similar to that of a network router. Choose hardware that is reliable and efficient. The important factors that you will want to focus on are network interface speed, memory, and CPU speed. Today that would mean at least a 100-Mbs Ethernet card, 256 MB of RAM, and a 700-MHz processor. (Turbolinux Cluster LoadBalancer 10 is only available for Intel-compatible architectures.)
Disk space is less critical, unless you are running other services on the machine as well. Be sure to factor in any other software that will be running on the machine. The Turbolinux Cluster LoadBalancer 10 software itself will take up approximately 40 MB of disk space. Additional space will be required for log files and other administrative tasks.
If an Advanced Traffic Manager is supporting NAT cluster nodes, then the ATM should have two network cards. One network card will be used to accept incoming client requests. The other will be used to connect to the NAT private network.
The hardware requirements for cluster nodes are the same as if the systems were running stand-alone. The primary concern will be what services are running on the node. There are no additional requirements beyond the hardware recommendations of the operating system and the applications that will be running on the node.
In order to provide the highest amount of uptime, you will want to employ as much hardware redundancy as possible. You should obviously use UPSes to ensure that the cluster will remain running in the event of a power failure. You may also want to consider redundant power supplies in each system. To ensure constant data access, you can use a RAID hard drive array. Drive mirroring and RAID 5 can provide redundancy, and hot-swappable hard drives will allow you to replace faulty components. Don't forget to perform routine system backups; redundant hardware can't prevent software catastrophes.
A CD-ROM drive is required to install the product. The CD-ROM does not necessarily need to be installed in the server; you may mount the CD-ROM on a different server and access it via NFS or some other method. You will also need a connection to the Internet to download updates and to register the product.
Table 1-1. Recommended Hardware Specifications.
| ATM (Advanced Traffic Manager) | |
|---|---|
| Computer Processor | 700 MHz or higher, Intel compatible |
| Memory (RAM) | 256 MB or greater |
| Network Card | 100 Mbps Ethernet cards |
| Hard Drive | RAID5 recommended |
| CD-ROM Drive | Required |
| Power Supply | UPS recommended |
 | If you plan to connect to a NAT network, you will need more than one Ethernet card. |
1.3.3. Infrastructure
To run a cluster of network services, you will obviously need to have a stable network. If possible, it is recommended that you have all the cluster nodes on a single subnet, and that this subnet be separate from the rest of the network. This allows the cluster to run at maximum performance, while isolating any problems from the rest of the network. For very high-traffic clusters, you may saturate the bandwidth of a single subnet; in that instance you might have to consider multiple subnets.
While putting all the nodes on a single subnet or LAN is recommended for maximum performance, it is by no means required. You have the flexibility to locate your nodes anywhere, especially when using the tunneling method. However, all the ATMs must be on the same subnet. This is because the ATMs will all need to be able to take on the virtual IP address of the cluster itself. This can only be done on the subnet that would normally contain that IP address.
If you are looking to create a high availability web site, you should consider redundant Internet routers on the network. If one of the routers goes down, you can still access the cluster from the outside. For maximum redundancy, the routers should go through separate Internet Service Providers. The high availability of your cluster won't matter much if you become disconnected from the Internet.
It is highly recommend that you have a DNS server running to map domain names into IP addresses. Reverse DNS lookups must be working properly as well, resolving IP addresses back into domain names. Like all servers, the systems within the cluster should have static IP addresses, not DHCP-assigned addresses.