Sources of Instability in Data Center Multicast
When replicating data in a cloud computing setting, it is common to send updates using reliable dissemination mechanisms such as network overlay trees. The authors show that as data center scale up, such multicast schemes manifest various performance and stability problems. Data centers, and particularly massive ones that support cloud computing, e-commerce, and social networking, necessarily replicate data. Replication is done for many reasons: to provision financial servers that respond to read-mostly requests, to parallelize computation, to cache important data items, for fault-tolerance, and the list goes on. Since updates to replicated data can be thought of as reliable multicasts, data center multicast is an important technology.