Using Distributed Diagnosis to Deploy Highly-Available Web Servers

This work presents SAPOTI (Dependable TCP/IP application servers - in portuguese: servidores de aPlicacoes cOnfiaveis Tcp/Ip), a distributed tool that guarantees the availability of TCP/IP application servers, in particular of Web servers. The tool is based on the SNMP framework and is executed on a group of Web servers running on a set of hosts that are monitored by a dependable distributed network management tool based on the hierarchical diagnosis algorithm Hi-ADSD with Timestamps. One server is elected to be responsible for the service. After this server becomes faulty and this event is diagnosed, the service is automatically recovered by electing another server among those that are fault-free. A priority scheme based on identifiers is defined. The service is available even if only one host/server is fault-free. Experiments are described obtained from a implementation of SAPOTI on a LAN with real Apache servers. Experiments that involved the injection of 210 faults distributed among a group of six servers were run, the measured availability was at least 97.3%. In another experiment with five servers, where 27 faults were injected, the availability was 99.5% during the whole experiment time.