Implementing a Linux Virtual Server [30 mn de lecture - paru le 9/19/2005 11:38:59 AM - Public : Expert]

Auteur

Vincent BEDNAREK Elève-Ingénieur Supinfo Paris Promotion SUPINFO 2005    Lui écrire    Tous les projets de cet auteur    Le mini-CV de cet auteur

2. Various algorithms of LVS

Various algorithms of load balancing are implemented in LVS. They define the order in which the load balancer will assign requests of the client to the various real servers proposing their service. Each service lodged in a farm of LVS servers (HTTP, FTP...) can have its own algorithm of distribution of load. Here algorithms proposed by LVS:

• Round-Robin Scheduling: In this configuration, the distribution functions in a cyclic way, without being concerned with load of the servers. The first request will be assigned to the 1st server, the second with the second server, so on in a loop.
• Weighted Round-Robin Scheduling: Same technique, but the real-servers can be affected by weights, to take account of the various processing capacities.
• Least-Connection Scheduling: The load balancer has a table of active connections. It will return all new requests to the server having less active connections, dynamically.
• Weighted Least-Connection Scheduling: Same idea that the preceding algorithm, by having the possibility of allocate weights to the servers.
• Locality-Based Least-Connection Scheduling: The load balancer chooses a real-server in a group according to the destination IP address.
• Locality-Based Least-Connection with Replication scheduling: Same that the precedent, with an additional functionality: if all the servers of a group are overloaded or inalienable, it chooses a server in another group and assign it to the 1st group of servers.
• Destination Hashing Scheduling : Affect the request arriving at a server of a fixed group in a hashing table, according to destination IP address.
• Source Hashing Scheduling: Assign the request to a real-server according to the source address.