Mr Brett Rosolen1, Mr David Wilde1, Mr Chris Myers1, Mr Warrick Mitchell1
1Aarnet, North Ryde, Australia, firstname.lastname@example.org
Data intensive research needs high capacity frictionless networks that can reliably and consistently deliver very large research data transfers without detrimental impacts on other uses of the network.
Much has already been done to date to enhance Australia’s National Research and Education Network, AARNet, to make this frictionless networking a reality. The national backbone now operates at 100Gbps, multiple 100Gbps services are in place across the Pacific, and the Science DMZ architecture has been implemented at 10Gbps rates at a small number of campus networks. The latter has created the potential for very large data flows to consume all the available bandwidth for a campus, so separating this bandwidth from the campus capacity allows data transfers at line speed, ensuring business continuity and unfettered research use.
However, the game changes completely when individual sites are connected at 100Gbps. Testing to date has demonstrated that single flows can consume a very significant portion of this capacity, creating the opportunity for extremely large data flows (dubbed “elephant flows”) between research infrastructure services, and greatly increasing the likelihood that research flows may impact business continuity.
The possible solution for this dilemma is to provide network capacity specifically for data intensive science, by enhancing the network to allow business as usual traffic to traverse paths that are separated from research flows.
AARNet’s new pathfinder network infrastructure, AARNet-X (AX) is designed to address this challenge, and to support extreme, unique and evolving customer requirements. It will also enable AARNet to develop expertise with new platforms and technologies.
This talk will identify the science drivers and subsequent design approach of the AARNet X network and how our community can use it to freely move data for better science outcomes.
Brett Rosolen is the Data Program Manager, eResearch at AARNet