Cloud-Based Network Performance Modeling for Contingency and Long-Term Planning

Due to enormous growth of Internet traffic, the scale and responsiveness required for delivering applications to end users can be cost-effectively achieved by the increased adoption of cloud-based, virtualized services built by widespread use of compute, storage, and network virtualization. Therefore, IT infrastructures of enterprises, R&E networks (e.g., DOE ESnet), and commercial network operators are evolving from tradi- tional, centralized, on-premise-only infrastructures to distributed and heterogeneous ones that incorporate resources from private and public clouds, where multiple tenants share the same physical resources. This introduces new challenges in maintaining end-to-end visibility and resource planning to provide predictable infrastructure performance. Legacy planning solutions typically depend only on aggregated traffic volumes, and run the planning process in quarterly/annual cycles. As networks become heterogeneous and virtualized, resource planning needs to be a continuous process, by accounting for application performance and traffic data, tenant Service- Level Agreements (SLA), etc. Network administrators need to predictively manage performance of the network infrastructures by planning ahead in case of failures and congestion issues as well as take into account the application/service performance trends along with historical traffic data for future capacity planning. Therefore, for the next generation of distributed, multi-tenant networks, a new approach for resource planning is required to satisfy application and end-user performance. In this Phase I SBIR project, Ennetix intends to develop NetM, a new, cloud-based approach for resource planning for distributed, multi-tenant networks by leveraging comprehensive network performance models. NetM modeling and planning solution will be applicable to two important planning problems – (1) Contin- gency Planning and (2) Long-Term (i.e., Capacity) Planning. In Contingency Planning, NetM will focus on issues such as detecting critical elements, impact of critical element failures, emulating failure/“what-if” sce- narios, recommendations during failures, etc. In Capacity Planning, issues such as classifying infrastructure based on performance, SLAs (of end-users/applications) impacted by underperforming elements, projected measures to improve performance, scenario-based planning, upgrade timeline, etc. will be addressed. NetM will be built on performance measurements (collected by perfSONAR services), traffic and topology data, and other network metadata. During Phase I, Ennetix will develop essential components of the NetM solu- tion, and conduct feasibility and scalability studies to evaluate the effectiveness and performance of NetM in resource planning for large, distributed network infrastructures. The proposed solution will greatly benefit network administrators and managers at DOE and other gov- ernment organizations through a new approach for network resource planning which considers application and end-user performance along with traffic data and significantly reduces capital expenditures. The wider benefits of this effort will extend well beyond the immediate DOE scientific community, and on to other enterprises, network operators, and cloud-service providers. In particular, many commercial cloud-service providers and enterprises can leverage the proposed cloud-based service to proactively plan for their dis- tributed network infrastructures.