Programmable,Extensible and Secure perfSONAR Appliance for Multi-domain Enterprise Monitoring

perfSONAR, a multi-domain measurement and monitoring framework has been widely deployed in high-performance computing and networking communities at hundreds of sites for over several years. These communities have been publishing their network performance data sets via web services to: (a) users, who can check health of network paths of interest, and (b) researchers, who analyze multi-domain network measurements to detect, diagnose and localize anomalies, as well as derive performance forecasts for network-aware applications. The successful adoption of perfSONAR can be attributed to the open, modular and distributed architecture. However, perfSONAR framework in its current form cannot be integrated into multi-domain enterprise monitoring due to lack of programmability and extensibility that causes significant challenges to: (i) add newer performance tools to collect data from unique devices or services, (ii) deterministically control sampling patterns of measurement points in a distributed manner, (iii) implement policy-awareness for enterprise perfSONAR federations with measurement-level agreements, and (iv) effectively analyze and visualize uncorrelated and correlated performance problems that can be diagnosed in real-time or offline. We propose to investigate and develop a set of perfSONAR extensions viz., E- perfSONAR that allow network operators within enterprises to more effectively customize perfSONAR instances for meeting monitoring objectives, enforcing measurement policies, and diagnosing performance bottlenecks. The extensions will be based on the DOE-funded R & amp;D efforts such as OnTimeSample and OnTimeDetect tools developed for perfSONAR communities. These tools use principles of web services, real-time scheduling, policy inference engines, estimation theory and signal processing to overcome existing programmability and extensibility limitations in perfSONAR. The E-perfSONAR will comprise of software packages that augment existing perfSONAR software distributions, and will be delivered to enterprise users in the form of low-cost, small form-factor appliances. DOE network operators, DOE labs and Universities will be the audience for Phase 1 trials of E-perfSONAR. Building upon the early trials of these users, we will develop a commercialization strategy that will target the appliances to be widely integrated into networks of any enterprise that monitors end-to-end network paths across multiple domains, and is concerned with Cloud-to-edges performance monitoring. Commercial Applications and Other Benefits: E-perfSONAR will be vital for network operators in performance bottleneck troubleshooting, as well as for network resource provisioning and will ultimately enable: (a) handling of wider-scale enterprise user requests for measurements (e.g., lab-to-lab, user-to-lab, user-to-user) with variety of tools that have not yet been integrated into the perfSONAR framework, (b) highly customizable measurement federations that satisfy monitoring objectives amongst multiple enterprise perfSONAR deployments, and (c) network-awareness in emerging autonomic network management applications that will adopt perfSONAR multi-domain measurement and monitoring capabilities.