Study And Improvement Of Load Dispatching In Virtual Service Cluster Environment

With the rapid development of Internet, virtual server based cluster technology is put forward to satisfy the needs for scalable, high available service. LVS provides an IP layer based cluster load balance scheduling method, but it is inefficient in the reliability of load dispatching.Based on the research into virtual server implementation mechanism, this thesis carefully analyzes the key scheduling technology of LVS, and provides an improved corporated dual-server load dispatching scheme. To detail the failure detection of the system, detecting module is integrated into the scheduling system and the availability judgment of scheduler is conducted at the task level. During the working process of the cluster, the linking information of clients and real servers is stored by the scheduler. When the main scheduler fails to work for some hardware or software reason, the backup scheduler does its best effort to make task handing over, based on the original linking information. The hash table where the linking information is stored is sent to the backup scheduler increasingly and timely by the main scheduler. When the main scheduler recovers from failure, the hash table is copied from the backup scheduler. The increasingly backup of hash table is implemented in the link layer by registering high priority protocol to shield the default ARP and TCP/IP protocol of the backup scheduler. This thesis constructed a runable virtual server cluster system, and implemented the high reliability virtual server load dispatching scheme on real-time embed operating system VxWorks.