ICN for Cloud Networking Lotfi Benmohamed Advanced Network Technologies Division NIST Information Technology Laboratory
Information-Access Dominates Today s Internet is focused on point-to-point communication Information access is the dominant use case in today s Internet More than 90% of Internet traffic is content retrieval Video accounts for 60% (both user-generated and VoD services) Solving content distribution problems via IP point-to-point communication as done today is inefficient Need an architecture that matches usage with design Evolution
Named Data Networking (NDN) Instead of addresses, packets have names users request content with an Interest packet network returns information in a Data packet Data packets are cached in NDN routers Interes t Forwarding Information Base (FIB) indicates output interfaces for every name received in Interest packets Pending Interest Table (PIT) records requesting interface for each Interest until Data packet is received Content Store (CS) temporarily caches returned Data Data PIT FIB CS
NDN Benefits Saves bandwidth through deferred and simultaneous multicast Data packet cached in the CS is returned directly on receipt of Interest Simultaneous multicast (eg, live video) realized using PIT entry, if name is already in PIT do not forward Interest Security built in rather than added on Every Data packet is signed Ease of application development In particular through proper naming I S P I S P I S P I S P
1. Do I have this data? 2. Is a request already pending? 3. Which next hop might lead to the source? NDN Forwarding
NDN Testbed NIST NDN team collaborates with the larger NDN community we have a NIST node on the NDN testbed Used for advancing the research agenda of NDN Managed out of University of Washington and includes 26 gateway nodes in the US (13), Europe (7), and Asia (6) Provides a means for real world testing that an emulation/simulation cannot provide
Related developments Standardization effort currently within IRTF Working Group on Information Centric Networking (ICNRG) Commercial products Cisco: for 5G backhaul in support of video distribution for mobile deployments
1. NDN for Data Intensive Science $1M, 2-year NSF funding starting July 2017 SANDIE: SDN-Assisted NDN for Data Intensive Experiments Northeastern U. (lead), Caltech, Colorado State Data-intensive science: Extracts knowledge from massive datasets with growing scale and complexity Uses global data distribution, processing, access, and analysis Needs coordinated use of computing, storage and network resources Project scope Explore NDN as a natural fit for data-intensive science Uses NDN principles to redesign the cloud infrastructure for the high energy physics (HEP) Large Hadron Collider (LHC) program NIST team engaged with the SANDIE team in a support role Exploring the role our NDN high-performance forwarding project can play for SANDIE
NDN for Data Intensive Science In support of Large Hadron Collider (LHC) program high energy physics program with world s largest data intensive applications LHC network connects CERN to 500 tiered sites worldwide LHC network traffic projected to grow by another 2 orders of magnitude by 2026 currently about 1 Exabyte (1 million terabytes)
2. Hadoop over NDN Project lead by one of the NDN PIs at U. Arizona that we're collaborating with (B. Zhang) Hadoop is a popular MapReduce framework for distributed storage and processing of large data sets Hadoop is a complex piece of software that requires nontrivial configuration and tuning for good performance NDN can improve the performance through Caching, multicast, multi-path and multisource data retrieval.
Hadoop over NDN Project goal: apply NDN to cloud network environment to improve the storage, access, and processing of large amount of data. Initial phase: modify Hadoop to run on top of NDN to establish performance baseline Next phase: design NDN-native distributed filesystem and network mechanisms to improve system performance Run various benchmarks comparing the performance of regular Hadoop versus Hadoop over NDN Writing 1GB data 128 node cluster utilizing Amazon s EC2 cloud service to explore the possibility of caching in a more realistic setup Conclusions: Opportunities for traffic reduction Caching and multicast.
3. NDN for Edge Computing/Networking Cloud computing consolidates computing resources into massive data centers Edge computing is an new computing paradigm that can provision storage and computing resources closer to the edge in small proximity to mobile devices / sensors / end users offer users low-latency and high-bandwidth access to cloud resources enables a new class of latency and bandwidth sensitive applications, not realizable with current cloud computing architectures (Iot at the edge) Other names used: fog / cloudlet / intelligent edge / micro-data centers Neighboring edge servers may need to cooperate for better user experience (cooperative edge systems) such as handoff of compute/storage resources under user mobility Mobile Edge Computing (MEC) in cloud-enabled vehicular networks DOI: 10.1109/RNDM.2016.7608300
A lot of interest from academic community: Topics of IEEE EDGE 2018 include: Architecture of Edge Computing systems Communication between Edges and Central Cloud IoT Edge Computing Mission Critical Edge Computing Edge Analytics Real-Time AI at the Edge
Intel is interested in ICN at the Edge Partnered with NSF to start a NSF-Intel ICN-WEN Program: Information Centric Networking in Wireless Edge Networks (ICN-WEN) $6.5M over 3 years One of the winners is the NDN team at UCLA ICE-AR: ICN-Enabled Secure Edge Networking with Augmented Reality Smart Campus use case with AR Apply NDN with domain-specific computing technologies to accelerate AR at the edge Just kicked off few months ago
https://www.nist.gov/news-events/events/2018/09/named-data-networking-community-meeting-2018