Skip to content

What is PATIO?

PATIO aspires to be a research content delivery network (CDN) and autonomous system (AS) for Internet-scale research experimentation in Africa. PATIO is designed to be a well-provisioned and federated network measurement and systems research testbed. This work is funded by the Internet Society Foundation. The project, from University of Washington, and in collaboration with Makerere Univeristy, aims to foster Internet measurement, protocol design, and distributed systems research in Africa. Our design as both a CDN and an AS has two benefits:

  • 1 As a CDN, we can simulate the setting (anycast) of a local African cloud which can support experimentation around models for African cloud computing
  • 2 As an AS, we can support the ability of research to run BGP experiments which can aid out understanding interdomaing routing and provide a mechanism to testing innovations in this area.

Why PATIO?

Today the growing reliance on anycast by CDNs and the increasing use of remote peering in African IXPs makes it unclear how these new technical and economic arrangements are impacting the markets and infrastructure for interconnection in Africa. Recent studies on peering and content delivery infrastructure have shown us evidence that in western markets these practices are changing the economic and technical aspects of interconnection and content delivery. However these works say little about the impact of these practices on Africa networks and end-users. We highlight four problems facing networking and systems researchers that care about the future of the Internet in Africa: lack of visibility, lack of topology-aware test-beds, unsustainability of research infrastructure, and better data on Internet economics.

Problem 1: Lack of Visibility

Internet measurement researchers are cognisant of how recent trends in internet networking are reducing visibility, from the physical layer to the application layer, impeding the ability to perform independent research 1. Specifically, networking trends like remote peering 2 and anycast 6 are influencing the network engineering practices of ISPs, IXPs, and CDNs. The increasing research interest in these trends has led to the creation of new testbeds like \textit{PEERING}3 and \textit{TANGLED}4 to support researchers aiming to understand the impacts of these new practices. Current research has documented that despite their prevalence and continued uptake, both remote peering and anycast create some major unintended and unwanted consequences: increased centralization of the Internet 5, route inflation and sub-optimal surrogate selection 6, and increased latency due to BGP selection of remote peering route 7. We believe these issues are even more stark in the context of African networking. However, the current concentration of existing research testbeds and their vantage points outside of Africa (See Table \ref{table:vantage-points}) makes it difficult and in many cases impossible to truly understand what these trends mean for African networking 8.

Problem 2 : Lack of topology-aware systems and networking testbeds

Recent trends in networking, the lack of transparency in BGP, and proprietary routing policies and private peering arrangements, significant limits the ability of existing measurement and monitoring tools and platforms to provide a correct account of interconnection. To date, significant networking and systems research has been supported by investment in and build out of internet-scale research testbeds such as Cloudlab 9, Chameleon 10, emulab 11, Planetlab 12, Tangled 13, M-Lab 14, and EdgeNet 15. More recently, these testbeds and platforms have been used to study and understand many aspects of modern networking. However, these studies provide little visibility into modern networking patterns in Africa. Given the well-documented presence of triangular or \textit{circuitous} routing 16 and 17, it is clear that the conclusions of existing testbeds can mislead researchers about the impact of recent trends in networking and distributed systems design in Africa. Given the complexity of inter-domain traffic behavior in Africa 17,8,16,18, there is a need for testbeds that are \textit{topology-aware} allowing better control and visibility of Internet-scale experiments.

Problem 3: Unsustainable research infrastructure.

The software and hardware sustainability of research testbeds has been well documented in academic literature. Since our primary goal is to design and deploy a testbed infrastructure in low-resource and economic-aware settings, we aim to constructed a testbed that will be easy to use, less costly to operates, and simple to deploy, maintain, upgrade, and extend.

Problem 4 : Better data on Internet economics.

From the economic sustainability of a CDN cache/PoP or the demand for open or private peering agreements, it is important to understand how the logical and physical properties of networks influence the economics of networking in Africa. Optimal designs of edge content delivery networks or peering arrangements (remote or local) are dependent on how both physical and virtual properties of networks. For example, Africa is dominated by siloed data centers that suffer from inherent technological challenges such as intermittent connectivity, power outages, and infrastructure access/setup costs. Outside South Africa, the big cloud providers have no data center presence to allay the above fears. Some governments and institutions have set up in-country mini-private clouds to run most of their services. As the cloud market evolves and cloud economics in Africa changes, there is a need for better mechanisms to understand these market dynamics and their demands on infrastructure, particular as these clouds move closer and closer to the edge.

Infrastructure and Connectivity

PATIO will operate a research cloud leveraging and improving on the open-source Openstack platform. We design and implement an EVPN/VXLAN network19 fabric which allows us to support multi-tenancy and policy driven federation across our sites. We will also use braids, a BGP controller we are building from the ground up, to support not only management of our infrastructure but also virtualization of BGP and anycast experiments. PATIOs plane to colocate our infrastructure at IXPs, NRENs, and the provider edge. Our design require we peer directly with collaborating networks (like NRENs) at IXPs. We will run compute and storage service for the platform out of our core network. Our peering relationships with NRENs will serve at the infrastructural backbone to support our infrastrucutre federation goals.

Objectives

The long-term goals of PATIO system are:

  1. Collaborative research infrastructure. PATIO aims to provide an Anycast and BGP testbed to provide more fine-grained experimentation that reflects that nature of interconnection in Africa and also simulate new protocols and peering configurations. %The leveraging of a community-AS that can be used to power production-like experiments. These ideas are similar to those use in Peering \cite{}: the ability to multiplex and delegate control of an AS to experiments.

  2. Experimental Flexibility (Slice-based). Support longitudinal, time-delimited, fine- and coarse-grained measurements and experiments. A PATIO experiment can involve an interconnected set of resources from different vantage points. Researcher can defined how these distributed resources are used in their experiments.

  3. Ease of Use. Simple and intuitive interfaces for researchers to define and managed experiments. Simplify the process of adding new PATIO physical resources.

  4. Extensibility and Composability. Decouple the experiment from the infrastructure to support evolvability and interoperability.
  5. Multiuser and on-demand Isolation. Platform capacity grows as a function of contributed nodes by accounts. Each contributed node can be added a part of shared experimentation infrastructure or isolated for research-specific experiments. Additional probes can be added on the fly for specific short-term or long-run experiments.

Prefix origination.

PATIO will announce only prefixes allocated to the platform. This condition ensures that we can only affect traffic destined to PATIO IP addresses.

Autonomous System origination.

The PATIO system will only announce BGP messages with AS-PATHs that start with the AS numbers allocated to the testbed.

  1. KC Claffy, David Clark, John Heidemann, Fabian Bustamante, Mattijs Jonker, Aaron Schulman, and Ellen Zegura. Workshop on overcoming measurement barriers to internet research (wombir 2021) final report. SIGCOMM Comput. Commun.Rev., 51(3):33–40, jul 2021 ↩

  2. Ignacio Castro, Juan Camilo Cardona, Sergey Gorinsky, and Pierre Fran¸cois. Remote peering: More peering without internet flattening. Proceedings of the 10th ACM International on Conference on emerging Networking Experiments and Technologies, 2014. ↩

  3. Brandon Schlinker, Todd Arnold, Italo F. S. Cunha, and Ethan Katz-Bassett. Peering: virtualizing bgp at the edge for research. Proceedings of the 15th International Conference on Emerging Networking Experiments And Technologies, 2019. ↩

  4. Leandro Marcio Bertholdo, Jo˜ao Marcelo Ceron, Wouter B. de Vries, Ricardo de Oliveira Schmidt, Lisandro Zambenedetti Granville, Roland van Rijswijk-Deij,and Aiko Pras. Tangled: A cooperative anycast testbed. 2021 IFIP/IEEE International Symposium on Integrated Network Management (IM), pages 766–771,2020 ↩

  5. Kevin Vermeulen, Loqman Salamatian, Sang Hoon Kim, Matt Calder, and Ethan Katz-Bassett. The central problem with distributed content: Common cdn deployments centralize traffic in a risky way. Proceedings of the 22nd ACM Workshop on Hot Topics in Networks, 2023. ↩

  6. Thomas Koch, Ethan Katz-Bassett, John Heidemann, Matt Calder, Calvin Ardi, and Ke Li. Anycast in context: A tale of two systems. In Proceedings of the 2021 ACM SIGCOMM 2021 Conference, SIGCOMM ’21, page 398–417, New York, NY, USA, 2021. Association for Computing Machinery ↩↩

  7. Fabrıcio M. Mazzola, Pedro de B. Marcos, Ignacio Castro, Matthew J. Luckie, and Marinho P. Barcellos. On the latency impact of remote peering. In Passive and Active Network Measurement Conference, 2022 ↩

  8. Roderick Fanou, Pierre Francois, and Emile Aben. On the diversity of interdomain routing in africa. In Passive and Active Network Measurement Conference, 2015 ↩↩

  9. https://www.cloudlab.us/ ↩

  10. https://www.chameleoncloud.org/ ↩

  11. https://www.emulab.net/portal/frontpage.php ↩

  12. https://planetlab.cs.princeton.edu/ ↩

  13. https://anycast-testbed.nl/ ↩

  14. https://www.measurementlab.net/ ↩

  15. https://www.edge-net.org/ ↩

  16. Arpit Gupta, Matt Calder, Nick Feamster, Marshini Chetty, Enrico Calandro, and Ethan Katz-Bassett. Peering at the internet’s frontier: A first look at isp interconnectivity in africa. In Passive and Active Network Measurement Conference,2014 ↩↩

  17. Roderick Fanou, Pierre Fran¸cois, Emile Aben, Michuki Mwangi, N. Goburdhan, and Francisco Valera. Four years tracking unrevealed topological changes in the african interdomain. Comput. Commun., 106:117–135, 2017. ↩↩

  18. Roderick Fanou, Francisco Valera, and Amogh Dhamdhere. Investigating the causes of congestion on the african ixp substrate. Proceedings of the 2017 Internet Measurement Conference, 2017 ↩

  19. See Cisco https://www.cisco.com/c/dam/en/us/td/docs/switches/datacenter/nexus9000/sw/vxlan_evpn/VXLAN_EVPN.pdf for a overview of this design ↩