Hardware/Network Requirements

An LVS cluster consists of one or two LVS routers and a collection of real servers providing Web, FTP, or other services. The connection and hardware requirements are described below.

A Linux server that will be the primary LVS router is required. If NAT routing is used, this machine needs two network adapters, one connecting it to the public network and the other to a private network (where the real servers are located). The private network is masqueraded from requesting clients. The servers on the private network may be any kind of computing platform running any operating system or Web server.

With IP encapsulation, the primary LVS router and real servers may be on the same LAN, or the real servers may be geographically dispersed (on a WAN). Real servers process requests directed to them and return responses directly to requesting clients.

With direct routing, the primary LVS router and real servers are on the same LAN segment. As with IP encapsulation, real servers return processed requests directly to requesting clients.

If you want failover capability, a second Linux server that will be the backup LVS router is required. This machine, a hot standby, needs to be configured exactly like the primary LVS router. For example, with NAT routing, the backup LVS router also needs two network adapters connecting it to the public network and to the private network of real servers. The adapter devices must match on the two LVS routers. Thus, if devices eth0 and eth1 on the primary LVS router connect it to public and private network, respectively, then these same devices on the backup LVS router must connect it to the public and private network.

During configuration, you assign a weight to each real server. This is an integer reflecting each server's processing capacity (based on memory, processor speed, number of processors, etc.) relative to that of the others. It is the ratio between each server's weights that are significant. For example, if the weights of two servers are 2/1, 20/10, or 200/100, the result is the same — the first server has twice the capacity of the second. However, note that finer adjustments are possible if larger weights are used. The assigned weight, which may be dynamically adjusted based on load information, is used by two of the available job scheduling algorithms (described in Table 8-1). You should be prepared to assign accurate weights.