Direct Routing

Direct routing provides increased performance benefits compared to NAT routing. Direct routing allows the real servers to process and route packets directly to a requesting user rather than passing outgoing packets through the LVS router. Direct routing reduces the possibility of network performance issues by relegating the job of the LVS router to processing incoming packets only.
LVS Implemented with Direct Routing

Figure 1.24. LVS Implemented with Direct Routing

In a typical direct-routing LVS configuration, an LVS router receives incoming server requests through a virtual IP (VIP) and uses a scheduling algorithm to route the request to real servers. Each real server processes requests and sends responses directly to clients, bypassing the LVS routers. Direct routing allows for scalability in that real servers can be added without the added burden on the LVS router to route outgoing packets from the real server to the client, which can become a bottleneck under heavy network load.
While there are many advantages to using direct routing in LVS, there are limitations. The most common issue with direct routing and LVS is with Address Resolution Protocol (ARP).
In typical situations, a client on the Internet sends a request to an IP address. Network routers typically send requests to their destination by relating IP addresses to a machine's MAC address with ARP. ARP requests are broadcast to all connected machines on a network, and the machine with the correct IP/MAC address combination receives the packet. The IP/MAC associations are stored in an ARP cache, which is cleared periodically (usually every 15 minutes) and refilled with IP/MAC associations.
The issue with ARP requests in a direct-routing LVS configuration is that because a client request to an IP address must be associated with a MAC address for the request to be handled, the virtual IP address of the LVS router must also be associated to a MAC. However, because both the LVS router and the real servers have the same VIP, the ARP request is broadcast to all the nodes associated with the VIP. This can cause several problems, such as the VIP being associated directly to one of the real servers and processing requests directly, bypassing the LVS router completely and defeating the purpose of the LVS configuration. Using an LVS router with a powerful CPU that can respond quickly to client requests does not necessarily remedy this issue. If the LVS router is under heavy load, it may respond to the ARP request more slowly than an underutilized real server, which responds more quickly and is assigned the VIP in the ARP cache of the requesting client.
To solve this issue, the incoming requests should only associate the VIP to the LVS router, which will properly process the requests and send them to the real server pool. This can be done by using the arptables packet-filtering tool.