Predictable Data Centers

Опубликовано 8 августа 2016, 22:09

The Predictable Data Centers (PDC) project tackles the issue of unpredictable application performance in the cloud. A key contributor to such unpredictability is shared resources like network and storage. The bandwidth for a VM, when communicating with other VMs or with shared services (like Azure storage), can vary significantly; recent studies have observed bandwidth variability up to a factor of five. To address this, we have designed a predictable data center architecture whereby users get aggregate performance guarantees across shared resources. This talk provides a project overview with a focus on how predictable network and storage performance can be enabled. Through key scenarios of interest, I will illustrate how PDC can- (i) enable end-to-end performance guarantees in shared settings, (ii) provide performance benefits to both tenants and providers, and (iii) enable new pricing models that lead to a symbiotic tenant-provider relationship.