Control Exchange Points: Providing QoS-enabled End-to-End Services via SDN-based Inter-domain Routing Orchestration
Vasileios Kotronis, Xenofontas Dimitropoulos, Rowan Kloti, Bernhard Ager, Panagiotis Georgopoulos, Stefan Schmid
CControl Exchange Points: Providing QoS-enabled End-to-End Servicesvia SDN-based Inter-domain Routing Orchestration
Vasileios Kotronis , Xenofontas Dimitropoulos , , Rowan Klöti , Bernhard Ager , Panagiotis Georgopoulos , Stefan Schmid ETH Zurich, Switzerland Foundation of Research and Technology Hellas (FORTH) T-Labs & TU Berlin, Germany
Introduction.
This paper presents the vision of the Control Exchange Point (CXP) architectural model. The model is moti-vated by the inflexibility and ossification of today’s inter-domain routing system, which renders critical QoS-constrained end-to-end (e2e) network services difficult or simply impossible to provide. CXPs operate on slices of ISP networks and are built onbasic Software Defined Networking (SDN) principles, such as the clean decoupling of the routing control plane from the dataplane and the consequent logical centralization of control. The main goal of the architectural model is to provide e2e serviceswith QoS constraints across domains. This is achieved through defining a new type of business relationship between ISPs, whichadvertise partial paths (so-called pathlets [7]) with specific properties, and the orchestrating role of the CXPs, which dynamicallystitch them together and provision e2e QoS. Revenue from value-added services flows from the clients of the CXP to the ISPsparticipating in the service. The novelty of the approach is the combination of SDN programmability and dynamic path stitchingtechniques for inter-domain routing, which extends the value proposition of SDN over multiple domains. We first describe thechallenges related to e2e service provision with the current inter-domain routing and peering model, and then continue with thebenefits of our approach. Subsequently, we describe the CXP model in detail and report on an initial feasibility analysis.
Motivation and Challenges.
Complexity and ossification:
The notorious complexity of the inter-domain routing systemrenders its management difficult and error-prone, leading to various inefficiencies such as suboptimal inter-domain paths. Indica-tively, 60% of all Internet paths today are suffering from triangle inequality violations [9]. The current ossification of the system,hindering the introduction of new solutions, aggravates the problem further. Highly popular inter-domain services, such as high-definition e2e real-time video streaming, already test the limits of the status quo, or are simply impossible. This is because suchservices require tight coordination along entire chains of ISPs demanding QoS provisioning. More advanced and mission-criticalservices, such as telemedical applications, are usually out of the question.
Participating IXPs R e a c h a b l e I P v a dd r e ss e s IXP adjacent addressesAdjacent with one hop customersTotal announced addresses
Fig. 1:
Cumulative coverage of IP address space origi-nated by IXP members and their 1-hop customer coneversus number of IXPs
Challenges associated with ISP peering:
ISPs today peer with eachother either directly over private peerings, or via the rich IXP ecosystemthat interconnects thousands of ISPs and is currently morphing the Inter-net landscape into a dense mesh, greatly increasing path diversity. Here wehighlight the limitations of the current peering ecosystem. Firstly, we arguethat the economic model of bilateral business agreements is not suitable forservices offered by multiple ISPs together. Secondly, peer ISPs exchangeall their clients’ IP prefixes coarsely, without any differentiation specific tothe cross-domain service that they want to provide to their clients. Fine-grained service-specific peering is a potential “nice-to-have” for ISPs, butBGP does not provide the needed mechanisms for implementing it withthe appropriate granularity (e.g., flow-level). Thirdly, peering practices to-day disfavor small (and potentially innovative) ISPs, since tier-1 ISPs formrestricted peering groups and have little incentive to offer peering, whenthey could simply charge for transit. This could change if, for example, anew type of business relationship for cross-domain services would jointlybenefit all chained ISPs, despite their differences in size or their original business association (customer-provider, peer-peer).
Benefits of the CXP model:
With this work, we are investigating how SDN principles can help deal with the inflexibility andsuboptimality of the inter-domain routing system and the issues related to classic ISP peering. The CXP model enables dynamicservice-specific relationships between ISPs, to provide cross-domain e2e services that can be guaranteed over the Internet. Exam-ple services that can be provided via peering under the supervision of the CXP entities include high quality video conferencing, telemusic [3] (teaching music or performing with remote participants over long distances by guaranteeing low-delay HD videostreaming), or mission-critical real time information streaming for telemedicine purposes [8]: the bandwidth and latency sensitivecontent in this case, could be the live video between an operating room in a hospital in Moscow and a doctor performing a remoterobot-assisted surgery, located in Zurich. Such applications demand a user-transparent, QoS-enabled, multi-domain WAN.
Control Exchange Points: A new notion of ISP peering based on Software Defined Networking.
A CXP is an externalto the ISP entity that orchestrates the e2e stitching of slices that the ISPs provide, for the benefits of e2e service revenue. ACXP manages slices of multiple ISPs and provides inter-domain routing coordination based on SDN APIs. A slice is defined bya flowspace (associated with a specific service) and a virtual topology (e.g., pathlets). An ISP abstracts its network as a set ofpathlets connecting the network edges and then advertises these to the CXP. More specifically, this abstraction could be realizedwith tunnels instantiated with e.g., OpenFlow or MPLS. Slices are connected via inter-domain links e.g., over IXPs, to other ISPdomains to form an inter-domain virtual topology. The pathlet abstraction is bundled with properties that the ISP provides. For a r X i v : . [ c s . N I] N ov xample, a pathlet can be annotated with minimum bandwidth and/or maximum delay guarantees, the number of routers it iscomposed of, the presence of middleboxes associated with a network service, or disjointness properties related to other offeredpathlets. The ISP may choose to advertise multiple pathlets that connect the same domain edges (e.g., for backup functionality oreven to diversify the provided guarantees as a form of market segmentation), thus enriching the set of offered pathlets.The task of the CXP is to admit requests for QoS-guaranteed e2e paths, embed paths in the inter-domain virtual topology [4],and monitor the provided QoS guarantees. The CXP can operate in parallel with BGP, as long as the control over the service-specific IP flows on the ISP domain edges is outsourced to the CXP, so that the rest of the BGP prefixes are isolated from theservice prefixes. The intra-domain control of the flows lies, of course, at the ISP’s hands. CXPs are the main entities forminga framework that is based on partial routing outsourcing, in contrast to full control plane outsourcing [10]. In both approaches,inter-domain routing is treated “as a service”, but the CXP approach requires much less outsourced control from the IXP’s side,and is more appealing for deployment and market adoption. Related work [5] in routing as a service focuses on similar notionsbut neither explores incremental deployment strategies nor investigates scalable solutions based on SDN mechanisms.CXPs are elements that operate solely within the control plane. They interface with ISPs in order to receive pathlet adver-tisements and be able to switch (i.e., route) between the ISPs’ pathlets, so as to establish e2e paths dynamically. Dynamic e2epath control allows to adjust to changing network conditions and to migrate embedded paths to admit new ones. We proposethat inter-domain connection between ISP slices takes place over IXPs. The data plane anchors that are used for switching arenetwork elements deployed within the layer-2 infrastructure of the IXPs. The main reason for selecting IXPs as the CXP dataplane anchors is that ISPs typically peer with many different IXPs in parallel for traffic offloading and transit cost reduction. Thisdense mesh of links provides path diversity which our model can exploit. All pathlets are attached to IXPs. The CXP anchors inthe IXPs are controlled by the logically centralized OS platforms [10] of the CXPs, based on SDN APIs. The introduction of theSDX [6] concept and the potential deployment of SDN-enabled infrastructure within IXPs, complement our model nicely. TheCXP could for example interface with the IXP through the SDX controller APIs and use SDX-defined slices of the current IXPfabric, according to the service and IXP clientele with which the slices are associated. LINX DE-CIX Terremark AMS-IX Equinix AshburnLINX - 2429 1093 2443 1427DE-CIX 2429 - 1093 2429 1427Terremark 1093 1093 - 1093 1093AMS-IX 2443 2429 1093 - 1427Equinix Ashburn 1427 1427 1093 1427 -
Table 1:
Pairwise path diversity between the 5largest IXPs
Regarding pathlet provision, we propose two models depending on whetherISPs are willing to provide hard guarantees or not. In the first model, an ISP pro-vides tunnels across its domain without any strict guarantees. The CXP cleverlystitches these pathlets together in order to satisfy the e2e requirements of the end-clients. To do that, active rerouting across unreliable ISP chains is required, basedon real-time measurements taken from the substrate network. IXPs could ideallyserve as monitoring anchors for such measurements. In the second model, an ISP provides tunnels with guaranteed performanceparameters across its domain. The CXP is still responsible for stitching these advertised, locally guaranteed pathlets so as to pro-vide e2e guarantees. In this case, monitoring is required for verifying that advertised guarantees are adhered to in practice. Theadoption of the most suitable model for current markets relates heavily to political and financial factors. We believe though, thatthe simple fact that an ISP is proficient in providing intra-domain tunnels and handling the traffic matrix for his own network [9],combined with the smart routing that the CXP can perform end-to-end, is a strong basis for the deployment of such models.
Preliminary Feasibility Analysis.
We claim that an architecture based on ISP-IXP-CXP collaboration can work in practice,considering the large number of IP addresses and the rich path diversity that even a small deployment of CXP anchors canprovide. We illustrate these two properties in Fig. 1 and Table 1, respectively. We use IXP membership data from [2] and builda map of possible pathlets connected to IXPs. Each pair of IXPs is connected via multiple pathlets traversing the joint memberASes (a single pathlet per joint AS per IXP pair). Fig. 1 depicts the IP address coverage ([1]) by IXP members (plus their 1-hopcustomer cone) versus the number of participating IXPs, assuming an optimal strategy maximizing IP address coverage. Weobserve that we can serve over 1 billion IP addresses via a small number ( ∼ Acknowledgements.
This work has received funding from the European Research Council Grant Agreement n. 338402.
References
1. CAIDA datasets. .2. Euro-IX: European Internet Exchange Association. .3. C. D
RIOLI ET AL . Networked Performances and Natural Interaction via LOLA. In
ECLAP (2013).4. G. E
VEN ET AL . Competitive and Deterministic Embeddings of Virtual Networks. In
ICDCN (2012).5. K. L
AKSHMINARAYANAN ET AL . Routing as a Service. Tech. Rep. UCB-CS-04-1327, (2004).6. N. F
EAMSTER ET AL . SDX: A Software Defined Internet Exchange. In
ONS (2013).7. P.B. G
ODFREY ET AL . Pathlet Routing. In
SIGCOMM (2009).8. R. R
AYMAN ET AL . Effects of Latency on Telesurgery: An Experimental Study. In
MICCAI (2005).9. T. A
NDERSON . Networking as a Service, HOTI-21 keynote. , (2013).10. V. K
OTRONIS ET AL . Outsourcing the routing control logic: better internet routing based on SDN principles. In