Linux Virtual Server

The major work of the LVS project is now to develop advanced IP load balancing software (IPVS), application-level load balancing software (KTCPVS), and cluster management components.

• IPVS: an advanced IP load balancing software implemented inside the Linux kernel. The IP Virtual Server code is merged into versions 2.4.x and newer of the Linux kernel mainline.
• KTCPVS: implements application-level load balancing inside the Linux kernel, as of February 2011 still under development.

LVS can be used for building highly scalable and highly available network services, such as web, email, media and VoIP services, and integrating scalable network services into large-scale reliable e-commerce or e-government applications. LVS-based solutions already have been deployed in many real applications throughout the world, including Wikipedia.

The LVS components depend upon the Linux Netfilter framework, and its source code is available in the net/netfilter/ipvs subdirectory within the Linux kernel source. LVS is able to handle UDP, TCP layer-4 protocols as well as FTP passive connection by inspecting layer-7 packets. It provides a hierarchy of counters in the /proc directory.

The userland utility program used to configure LVS is called ipvsadm, which requires superuser privileges to run.

Schedulers

LVS implements several balancing schedulers, listed below with the relevant source files:

• Round-robin (ip_vs_rr.c)
• Weighted round-robin (ip_vs_wrr.c)
• Least-connection (ip_vs_lc.c)
• Weighted least-connection (ip_vs_wlc.c)
• Locality-based least-connection (ip_vs_lblc.c)
• Locality-based least-connection with replication (ip_vs_lblcr.c)
• Destination hashing (ip_vs_dh.c)
• Source hashing (ip_vs_sh.c)
• Shortest expected delay (ip_vs_sed.c)
• Never queue (ip_vs_nq.c)
• Maglev hashing (ip_vs_mh.c)

Commonly used terms include the following:

• LVS director: load balancer that receives all incoming client requests for services and directs them to a specific "real server" to handle the request
• Real servers: nodes that make up an LVS cluster which are used to provide services on the behalf of the cluster
• Client computers: computers requesting services from the virtual server
• VIP (Virtual IP address): the IP address used by the director to provide services to client computers
• RIP (Real IP address): the IP address used to connect to the cluster nodes
• DIP (Directors IP address): the IP address used by the director to connect to network of real IP addresses
• CIP (Client IP address): the IP address assigned to a client computer, that it uses as the source IP address for requests being sent to the cluster

Setting up a virtual HTTP server with two real servers:

ipvsadm -A -t 192.168.0.1:80 -s rr ipvsadm -a -t 192.168.0.1:80 -r 172.16.0.1:80 -m ipvsadm -a -t 192.168.0.1:80 -r 172.16.0.2:80 -m

The first command assigns TCP port 80 on IP address 192.168.0.1 to the virtual server. The chosen scheduling algorithm for load balancing is round-robin (-s rr). The second and third commands are adding IP addresses of real servers to the LVS setup. The forwarded network packets shall be masked (-m).

Querying the status of the above configured LVS setup:

# ipvsadm -L -n IP Virtual Server version 1.0.8 (size=65536) Prot LocalAddress:Port Scheduler Flags -> RemoteAddress:Port Forward Weight ActiveConn InActConn TCP 192.168.0.1:80 rr -> 172.16.0.2:80 Masq 1 3 1 -> 172.16.0.1:80 Masq 1 4 0