网站高可用架构tech.pdfVIP

High Availability Introduction Businesses rely heavily on email for central business processes like customer transactions, vendor relationships, and internal communications. Our industry-leading 99.99% uptime guarantee is a reflection of both our understanding of emails central importance to many businesses and a reflection of the high availability email system engineered by our experts. Webmail.us has selected an approach to high availability popularized by Google: large numbers of computer servers built on commodity hardware. The basic theory is that server failures are common, so arrange servers in pairs or clusters, then build failure detection into the network or the software. In the past, we have tried the more traditional, high availability design approach: high-end servers with high-end disk subsystems. The mean time between failure (MTBF) was only somewhat better than commodity hardware, although the high-end server hardware was many times more expensive. For the same cost, key services are now spread across large numbers of servers in single-purpose clusters. Our experience has shown this greatly reduces the risk that any service might become unavailable to our customers. Service Architecture Front-end Application Servers There are four core email-related services offered by Webmail.us: inbound email (SMTP), outbound email (SMTP), remote email client access (POP IMAP), and web-based email access (webmail). Each of these four services operates from their own high performance cluster of servers. Each cluster is load-balanced by a pair of network load balancers. These load-balancing devices are capable of sensing failure in cluster nodes and seamlessly route traffic to the other functioning nodes. When a server fails, engineers are alerted and issues are diagnosed and resolved before the server is made available again to the server cluster. Excess cluster capacity is implemented, such that multiple server failures could occur without a n