Community Research and Development Information Service - CORDIS

H2020

ENDEAVOUR Report Summary

Project ID: 644960
Funded under: H2020-EU.2.1.1.3.

Periodic Reporting for period 1 - ENDEAVOUR (Towards a flexible software-defined network ecosystem)

Reporting period: 2015-01-01 to 2016-06-30

Summary of the context and overall objectives of the project

"Objectives

ENDEAVOUR: Towards a flexible software-defined network ecosystem

"Bringing SDN to the Inter-domain settings"

The focus of the project is to enable added-value services to be provided thanks to Software-Defined Networking (SDN), on top of Internet Exchange Points and other network interconnnection fabrics. The services would relate not only to the flexibility of the interconnection fabric, but most importantly to enable the content and data center ecosystem that is present at the interconnection fabric to collaborate. The ultimate goal is to create a service marketplace on top of the ecosystem composed of Cloud/data centers, networked applications, and the interconnection fabric.

The objective of ENDEAVOUR is to address current limitations of the Internet interconnection model, as well as to open the opportunity for novel services, creating the possibility for new economic models around the created ecosystems

Context & Problem

The rise of the IXP

Internet eXchange Points (IXPs) originate at the time the Internet transitioned from an academic/research network into today’s commercial infrastructure. Indeed, the basic role of IXPs dates back to the establishment of Network Access Points (NAPs). Over the past 20 years, as the Internet grew by leaps and bounds by any imaginable metric, the original four NAPs were replaced by a steadily increasing number of modern IXPs. The critical role they have played in the Internet ecosystem has until recently gone largely unnoticed by the Internet community at large.
Cloud & datacenters in the Internet

To cope with the increasing demand for content, content delivery infrastructures (including Content Delivery Networks (CDNs), datacenters, and Cloud providers) deploy massively distributed server infrastructures. IXPs already play a key role in inter-connecting the infrastructures that host content and applications with the ecosystem that consumes this content or relies on these applications. IXPs will become even more critical when tighter delay constraints are necessary and more support from the network is required by applications running in the Cloud.
Software-Defined Networking (SDN)

Software-defined networking has attracted a lot of attention in the networking community over the last few years. With attention also comes myths and misunderstanding, which we will attempt to clarify below by defining the main characteristics of SDN.

Objectives

Interdomain SDN control plane

ENDEAVOUR will research, develop, and evaluate an SDN architecture for the network ecosystem of a large IXP and its members. It includes:
- Distributed SDN Control Plane
- SDN Programming Abstractions

Scalable fine-grained monitoring

ENDEAVOUR will develop a Monitoring Platform for the SDN-enabled IXP. To enable new networking services and technology, monitoring tools need to be flexible in what/how much information is collected, while being able to operate at the high-speed data rates and massive data volumes happening at current and future IXPs. This objective can be broken down into the following specific objectives:
- High Speed Data Plane
- Monitoring System Open and Extensible Monitoring Toolbox

Flexible IXP management for operators

The operation of an IXP is complex, given the volume of data involved, and the number and diversity of members that peer through its infrastructure. Currently, IXPs lack the awareness and manageability to make their operations more efficient and resilient. We will develop specific use cases that demonstrate how the ENDEAVOUR SDN and monitoring tools can greatly facilitate the internal management of IXPs:
- IXP Protection against Member’s Equipment Misconfiguration
- Internal Traffic Visibility

Network services offered to the IXP ecosystem

ENDEAVOUR will implement a set of use cases that showcase the types of novel interdomain network services that can be offered to the IXP ecosystem. We will explore multi-homing, application-level traffic engineering, and flow anomaly detection."

Work performed from the beginning of the project to the end of the period covered by the report and main results achieved so far

During the first reporting period, the consortium has been working in all the working packages and has met all the milestones. All the deliverables have been submitted to date, although some reports regarding the “Public project presentation” and the “Project Web site” were a bit late in time due to a deliverable scope misunderstanding. However, the website and the presentation were published and available in time online.

1. Work Package 1: Management

Queen Mary University of London (QMUL) has been leading work package 1 (WP1) and, as a project coordinator, organised and contributed to the achievement of all the required milestones. Such milestones set the ground for the ulterior developments of the project as well as helped the project to gain visibility through its involvement in multiple workshops and other publicity activities.

In addition, DE-CIX Management GmbH (DE-CIX) has already established a shared testbed environment. The testbed is located at DE-CIX data center facility, which provides an environment with the highest standards in terms of security and reliability. By now, the first set of computing hardware is installed by one partner. Additional partners are already committed to let their hardware be be installed by DE-CIX in the upcoming weeks.

The testbed is designed as a shared computing and testing environment. It enables all partners to get hands-on experience with SDN hardware.

DE-CIX also made huge efforts in providing relevant data sets to be analyzed by the ENDEAVOUR partners. The data gives valuable insights in the technical implementation and daily business of IXPs. It covers both the control and data plane of the DE-CIX infrastructure.
The already established testbed will help ENDEAVOUR partners to first understand what the technical requirements of an IXP are. As a second step it will also enable the rapid development of SDN-based prototypes. Those prototypes will show the potential and proof the concept of SDN in an IXP context.

2. Work Package 2: Software-Defined Networking

Work package 2 (WP2) has been led by the Universite Catholique de Louvain (UCL), coordinating the joint work of the QMUL and the Centre National de la Recherche Scientifique (CNRS), in order to gather the initial requirements of the ENDEAVOUR SDN control platform by conducting an extensive research of the desired system goals.

UCLO contributed to Work Package 2 with several pieces of technical work that advance the state of the art of SDN in general, and in particular represent the building blocks for the creation of the ENDEAVOUR SDN control platform. Articles presenting some of these results have already been published at top venues (INFOCOM'15 and SOSR'15).

IBM has gathered the requirements of its SoftLayer-operated datacenters/Cloud, centered on SDN overlays, virtualization and monitoring - corroborated with the SoA documented in the deliverables. We have performed an in-depth analysis of the leading switch ASICs from Broadcom and Mellanox wrt. their SDN (and monitoring) capabilities vs. impairments, as related to Endeavour. The main gaps identified in the current multi-Tbps switches relevant to our project are (i) insufficient OpenFlow support, and, (ii) limited access to the existing h/w resources (lack of s/w interfaces). These are being characterized and prioritized, with constructive proposals/discussions ongoing with the respective vendors.
Takeaway: Endeavour is actively contributing to the design req’ts for the upcoming switching products from the market leaders. The results of our weekly technical sessions with the vendors, while currently under strict NDAs, will be --at least partially-- shared with the consortium in the near future.

Although not being part of the deliverable, CNRS proposed a new architecture and mechanisms called UMBRELLA that aimed at correcting scalability issues in SDN based IXP. It especially solve the problem of ARP messages flooding that dramatically impacts large scale networks. This work lead to the writing of two papers: one is under process review of a major international conference. The second still needs to be finalized. This work was also described in deliverables of the project, and contributes to the project achievements and milestones.

3. Work Package 3: Measurements & Monitoring

QMUL has been leading a joint effort with the University of Cambridge (UCAM), IBM and CNRS in order to identify the measurements and monitoring requirements for the project monitoring platform so as to accomplish the milestone for the work package 3 (WP3). In this line, IBM actively helped with gathering use cases from the literature, and their classification wrt. Endeavour’s targets.

IBM has evaluated the pros and cons of using COTS switches and SwitchOS (e.g., Cumulus, AristaOS) vs. FPGA-based OpenFLow switches vs. designing a new switching fabric. The latter intended specifically for datacenter and IXP apps, with the required SDN and monitoring capabilities. We concluded that the current ODM switches --while still lacking in OF support as compared to the current FPGA switches (e.g., Novi, Corsa)-- are the best compromise between cost, aggregate capacity, performance, Ethernet 802 compliance and monitoring features.
For load monitoring, IBM has started the design of a sub-ms/us aggregate buffer sampler --no per flow state, nor packet-level privacy concerns-- that relies on the QCN standard, in the Endeavour context re-used strictly for instrumentation and SDN interfacing.

Alongside this activity, UCAM carried out an extensive dissemination activity presenting new ways to assess (and monitor) systems performance beyond 10Gbps. These activities (i.e., talk, tutorial, demo) helped to drive useful discussions on high performance measurement artifact (the building block for WP3).

4. Work Package 4: Use-cases

Work package 4 (WP4) was led by DE-CIX and also contributed by QMUL, UCAM, CRNS, UCL, and IBM in order to produce the deliverable 4.1 (D4.1 Final use cases from related work), which is a comprehensive survey of use cases summarizing the relevant work from related publications. As the Cloud and SDN-virtualized datacenter operators are becoming an increasingly prevalent IXP customer/member, we have identified 7 related use cases centered on the future SDN-enabled IXP as a Cloud Broker. Together with IBM we’re gathering the respective req’ts and key features of these use cases collectively grouped under Cloud Broker.

DE-CIX focused on collecting applicable use cases for deploying SDN at IXPs and additionally take into consideration work that one can envision to be beneficial at IXPs. The identified fields comprise traffic engineering, peering, security, new enabling services, monitoring applications, and new opportunities to optimize IXP management.

UCAM performed an extensive research on monitoring features to be enabled in order to achieve the proposed use cases. CNRS also added value to the use cases by defining new proposals regarding anomaly detection in IXP’s network traffic.

In contrast to today’s peering environment, which is solely destination-oriented, we envision a much more fine-grained peering (e.g., application specific peering). This may also affect inbound traffic engineering by allowing forwarding rules that match any IP header field and thus can handle the traffic on the desired granularity. Moreover, we identified a multitude of security related use cases. Among other benefits, SDN-filtering mechanisms allow IXPs to be more robust against accidental and intentional misconfiguration. While ENDEAVOUR strives to provide the technology for this new services, it makes use of SDN’s separation of control and data planes by proposing services such as centralized routing or traffic steering.
The provided use cases will be revisited and extended by ENDEAVOUR in future deliverables to determine whether they are to be considered in our SDN architecture or not.
To date, a valuable feedback from a variety of different network operators has been collected. However, large CDNs were not addressed due to the venues we chose for our workshops. Therefore, we are aiming to have a discussion during the RIPE 71 meeting with such large CDNs (e.g., Akamai, Google, and Limelight). This feedback directly feeds our selection process to identify the most valuable use cases as extracted from the related work in D4.1. The work will be made public in D4.3 (Design of use cases for members of IXPs).

ENDEAVOUR also seeks to take the IXP operators into account. As one of the largest IXPs in the world, DE-CIX understands the needs of operators. They have collected the requirements suggested by the different departments of DE-CIX and combine them with use cases gathered by the related work survey (D4.1). This work will be comprised in D4.2 (Design of use-cases for operators of IXPs) and eventually used to design the SDN architecture.

To maximise the impact we want to get our testbed equipped with SDN switching hardware as soon as possible. Hence, we need to carefully evaluate the current SDN hardware market and synchronize our purchase intentions across the consortium. As soon as the decision making process is finalized we need to deploy the switches within the testbed and install the software.

We also evaluate the current SDN controller software capabilities and limitations. The SDN controller envisioned by ENDEAVOUR aims at using and extending the existing solutions to implement the selected use cases.

One of those existing solutions addresses the increasing broadcasting traffic in large layer-2 IXP networks. DE-CIX contributes to the recently proposed IETF Internet draft “Operational Aspects of Proxy-ARP/ND in EVPN Networks”, which eases such problem by enabling central ARP/ND resolution.

5. Work Package 5: Dissemination & exploitation

CNRS is leading work package 5 (WP5 Dissemination & exploitation). Consequently, CNRS was the main contributor for the production and dissemination of documents, such as a poster and a brochure. CNRS also designed, implemented and is hosting the project web site. CNRS has been disseminating the project activities by participating in major SDN-related events.

The ENDEAVOUR consortium, with substantial contributions from DE-CIX and QMUL, organized two workshops to present the goals of the project and explore the interest of the fundamental stakeholders of the IXP ecosystem, namely IXP members (i.e., ISPs, content delivery networks, and access providers) as well as IXP operators.

The first ENDEAVOUR workshop, “The Next Generation of Interconnection”, presented the project to a large audience of medium and small ISPs that are members of the DE-CIX IXP and surveyed their interest and expectations. The findings were unexpected and showed the slim understanding of SDN and its potential and thus the need to pursuit further dissemination efforts. This first workshop also revealed that the interest of medium and small sized ISPs on new solutions is accompanied by a limited perceived ability to invest the required resources for its implementation. In particular, IXP members showed special interest in having better information on potential peers, improved traffic monitoring, application specific peering capabilities, and IP prefix hijacking detection. Those findings were supported by face-to-face discussions with selected members during RIPE 70.

At the Euro-IX industry meeting, the second ENDEAVOUR workshop addressed the other cornerstone of the IXP ecosystem, namely, the IXP operators. The workshop succeeded in helping the consortium to gain a deeper understanding of the operational challenges and limitations of today’s IXP infrastructures. In particular, it showed that while there is interest in a breakthrough of the current interconnection paradigm, the ultimate constraint for operators is that their platform must remain highly available and neutral.

All in all, the two workshops showed that at this point the interest in SDN is not matched by a sufficient understanding of this technology, and hence further dissemination efforts by ENDEAVOUR are necessary. On the other hand, the findings of both workshops exposed the complexity of the Internet and the limitations and intricacies faced by the agents that compose it. To realize the benefits promised by SDN, ENDEAVOUR must develop a comprehensive and in-depth understanding of the particular problems of all relevant agents of the interdomain ecosystem, especially regarding how they are affected by the adoption of SDN technology.

Additionally, UCL and IBM gave two presentations about the project at the SWITCH SDN Workshop held in Bern on 11th June 2015.

QMUL presented ENDEAVOUR and chaired the posterior discussion in both of the aforementioned workshops and was also one of the presenters of the project and its developments to the External Advisory Board (EAB). During the EAB meeting, the consortium received feedback on the project from leading researchers in the area, such as Prof. Jennifer Rexford, as well as relevant companies interested in the developments, like Google, Corsa Technology and Brocade. The EAB will also continue providing feedback in the future.

As the local organizer of SIGCOMM, QMUL helped to publicize the ENDEAVOUR project by delivering the aforementioned poster and brochure. These documents present the basic ENDEAVOUR information, and were delivered to more than 500 attendants of the conference together with presents depicting the ENDEAVOUR logo. Besides, a demo (Enabling Performance Evaluation beyond 10Gbps - Antichi, Rotsos, Moore. SIGCOMM demo 2015) and a tutorial (Open Hardware Networking - Zilberman, Antichi, Rotsos. SIGCOMM tutorial 2015) were also presented by UCAM during the conference.

Additionally, a paper presenting a novel interdomain routing mechanism which was co-authored by a QMUL member of the ENDEAVOUR project (Route Bazaar: Automatic Interdomain Contract Negotiation. Ignacio Castro, Aurojit Panda, Barath Raghavan, Scott Shenker, Sergey Gorinsky. HotOS 2015) was accepted and presented at the HotOS Workshop.

IBM has given two tutorials on Datacenter SDN and Layer 2 monitoring: (a) HPSR-16 (http://www.ieee-hpsr.org/2015/), and, (b) HOTI-23 (http://www.hoti.org/hoti23/tutorials/ with 100+ paid registrations). Both were selected as main tutorials and delivered in the plenary rooms to larger than usual audiences. Performance impairments of the current Cloud SDNs were presented alongside with solutions, leading to a productive dialog with the industry --Broadcom, Cisco, Intel, Brocade, Mellanox, HP, Juniper, telcos-- on how to build observable (instrumented) underlay and overlay/SDN fabrics at 40/50/100 Gbps.

Alongside these activities, UCAM carried out an extensive dissemination effort by presenting new ways to assess (and monitor) systems’ performance beyond the 10Gbps. All these actions (i.e., talk, tutorial, and demo) helped to drive useful discussions on the high performance measurement artifact (the building block for WP3).

Progress beyond the state of the art and expected potential impact (including the socio-economic impact and the wider societal implications of the project so far)

ENDEAVOUR will have direct benefits for IXPs, Cloud providers and ISPs, but given the fundamental role of the Internet in our society and economy, the benefits of research work, if successful, will likely extend to virtually any community of Internet users. As such, our potential beneficiaries include the general public, governments, industry, academia and the third sector. More concretely, the expected impact of ENDEAVOUR includes:
- Creating new economic models in the form of marketplaces to meet offer and demand between Cloud/application providers and ISPs.
- Enabling regional networks to be more competitive by exchanging traffic directly. This is crucial to keep small regional networks simple.
- Improving the quality and resiliency of the communication infrastructure, ultimately aiding to reduce the digital-divide.
- The way IXPs work will change completely from a basic best-effort switching fabric to a manageable and controllable network that provides enhanced traffic engineering services. This will change the architecture of the Internet, by enabling intelligent services within the network connectivity fabric.

ENDEAVOUR aims at addressing to major requirements for networking, that are especially critical when SDN is concerned: scalability and security.Scalability is a major requirement for the Internet that still increases in an exponential way, and for which the current management systems are not scalable. By offering new scalable management systems for IXP networks, we will provide IXP some tools they need for the Internet to survive the still ongoing exponential increase of user demands. Security is the second major issue in networking, and CNRS will provide a solution for autonomously detecting and discarding malicious traffic that significantly reduces network performances by uselessly consuming network resources.

Related information

Follow us on: RSS Facebook Twitter YouTube Managed by the EU Publications Office Top