Availability modeling and analysis of a data center for disaster tolerance

摘要

Availability assessment of a data center with disaster tolerance (DT) is demanding for cloud computing based businesses. Previous work attempted to model and analyze the computing systems without a good consideration on disaster occurrence, unexpected failure of network connection and proper dependencies between subsystems in a data center. This paper presents a comprehensive availability model of a data center for DT using stochastic reward nets (SRN). The model incorporates (ⅰ) a typical two-level high availability (HA) configuration (i.e., active/active between sites and active/passive within a site), (ⅱ) various fault and disaster tolerant techniques; (ⅲ) dependencies between subsystems (e.g. between a host and virtual machines (VMs), between a network area storage (NAS) and VMs) and dependency between disastrous events and physical subsystems; and (ⅳ) unexpected failures during data transmission between data centers. The constructed SRN model is analyzed on the basis of steady state analysis, downtime cost analysis, and sensitivity analysis with regard to major impacting parameters. The analysis results show the availability improvement of the disaster tolerant data center (DTDC) and featured system responses corresponding to the selected variables. The modeling and analysis of the DTDC in this paper provide a selection basis of designing for disasters in consideration of the trade-off between system availability and downtime cost with infrastructure construction cost.