T-110.6120 – Special Course on Data Communications Software: Publish/Subscribe Internetworking Evolution vs. Revolution Arto Karila Aalto-HIIT arto.karila@hiit.fi 2011-09-13 AK T-110.6120: Publish/Subscribe Internetworking 1 Evolution vs. revolution • The Internet has evolved from the 1970’s and with no big changes since 1993 • This has led into a stagnated situation where it is very hard to make changes to the core protocols • It the Internet was redesigned from scratch, it would probably be very different from what the current Internet has evolved to • Various clean-slate solutions are current research topics and some of them may lead into a new Internet • It is possible that all the protocol layers, including the Internet Protocol, will change • However, any new solution will have to operate as overlay on the existing IP infrastructure to succeed • The publish/subscribe paradigm (pub/sub) is one of the most promising new paradigms 2011-09-13 AK T-110.6120: Publish/Subscribe Internetworking 2 Some evolutionary approaches A brief look at some evolutionary solutions proposed to the Internet’s shortcomings: • IPv6 • IPSEC • Mobile IP (v4 and v6) • HIP • DiffServ • DHT 2011-09-13 AK T-110.6120: Publish/Subscribe Internetworking 3 IPv6 • IPv6 was born in 1995 after long work • There are over 30 IPv6-related RFCs • The claimed improvements in IPv6 are: – Large 128-bit address space – Stateless address auto-configuration – Multicast support – Mandatory network layer security (IPSEC) – Simplified header processing by routers – Efficient mobility (no triangular routing) – Extensibility (extension headers) – Jumbo packets (up to 4 GB) 2011-09-13 AK T-110.6120: Publish/Subscribe Internetworking 4 IPv6 • Major operating systems and many ISPs support IPv6 • The use of IPv6 is slowly increasing in Europe and North America but more rapidly in Asia • In China, CERNET 2 runs IPv6, interconnecting 25 points of presence in 20 cities with 2.5 and 10 Gbps links, with each PoP providing 1 to 10 Gbps speeds to an access network (http://www.cernet2.edu.cn) • IPv6 really only solves the exhaustion of Internet address space 2011-09-13 AK T-110.6120: Publish/Subscribe Internetworking 5 IPSEC • IPSEC is the IP-layer security solution of the Internet to be used with IPv4 and IPv6 • Authentication Header (AH) only protects the integrity of an IP packet • Encapsulating Security Payload (ESP) also ensures confidentiality of the data • IPSEC works within a Security Association (SA) set up between two IP addresses • ISAKMP (Internet Security Association and Key Management Protocol) is a very complicated framework for SA mgmt 2011-09-13 AK T-110.6120: Publish/Subscribe Internetworking 6 Encapsulating Security Payload (IPv4) Original IPv4 Header Security Parameter Index (SPI) Sequence Number Coverage of Authentication UDP/TCP Header Coverage of Confidentiality ESP Payload Data Padding Pad Len Next Hdr Authentication Data 2011-09-13 AK ESP Header T-110.6120: Publish/Subscribe Internetworking ESP Trailer 7 Encapsulating Security Payload (IPv6) Original IPv6 Header Hop-by-Hop Extensions Security Parameter Index (SPI) Sequence Number Coverage of Authentication End-to-End Extensions UDP/TCP Header Coverage of Confidentiality ESP Payload Data Padding Authentication Data 2011-09-13 AK ESP Header T-110.6120: Publish/Subscribe Internetworking ESP Trailer 8 Mobile IPv4 • Basic concepts: – Mobile Node (MN) – Correspondent Node (CN) – Home Agent (HA) – Foreign Agent (FA) – Care-of-Address (CoA) • Problems: – Firewalls and ingress filtering – Triangular routing 2011-09-13 AK T-110.6120: Publish/Subscribe Internetworking 9 Mobile IP Triangular Routing Ingress filtering causes problems for IPv4 (home address as source), IPv6 uses CoA so not a problem . Solutions: Correspondent (reverse tunnelling) or Host route optimization Foreign agent left out of MIPv6. No special support needed with IPv6 autoconfiguration DELAY! Foreign Agent Home Agent Care-of-Address (CoA) Mobile Host 2011-09-13 AK Source: Professor Sasu Tarkoma T-110.6120: Publish/Subscribe Internetworking 10 Ingress Filtering Packet from mobile host is deemed "topologically incorrect“ (as in source address spoofing) Correspondent Host Home Agent With ingress filtering, routers drop source addresses that are not consistent with the observed source of the packet Source: Professor Sasu Tarkoma 2011-09-13 AK T-110.6120: Publish/Subscribe Internetworking 11 HIP • Host Identity Protocol (HIP, RFC4423 & others) defines a new global Internet name space • The Host Identity name space decouples the name and locator roles, both of which are currently served by IP addresses • The transport layer now operates on Host Identities instead of IP addresses • The network layer uses IP addresses as pure locators (not as names or identifiers) 2011-09-13 AK T-110.6120: Publish/Subscribe Internetworking 12 HIP Architecture Source: http://infrahip.hiit.fi 2011-09-13 AK T-110.6120: Publish/Subscribe Internetworking 13 HIP • HIs are self-certifying (public keys) • HIP is a fairly simple technique based on IPSEC ESP and HITs (128-bit HI hashes) • It addresses several major issues: – Security – Mobility – Multi-homing – IPv4/IPv6 interoperation • HIP is ready for large-scale deployment • See http://infrahip.hiit.fi for more info 2011-09-13 AK T-110.6120: Publish/Subscribe Internetworking 14 Base exchange • Based on the SIGMA family of key exchange protocols Source: Dr. Pekka Nikander Select precomputed R1. Prevent DoS. Minimal state kept at responder! Does notstandard protect against replay Diffieattacks. authenticated Initiator solve puzzle Responder Hellman key exchange for session key generation I1 HIT , HIT or NULL R1 HIT , [HIT , puzzle, DH , HI ] I2 [HIT , HIT , solution, DH ,{HI }] R2 I R I R I R R R sig I I sig [HIT , HIT , authenticator] I R sig verify, authenticate, replay protection User data messages ESP protected TCP/UDP, no explicit HIP header 2011-09-13 AK T-110.6120: Publish/Subscribe Internetworking 15 HIP Mobility • Mobility is easy – retaining the SA for ESP 2011-09-13 AK T-110.6120: Publish/Subscribe Internetworking 16 HIP in Combining IPv4 and IPv6 • An early demo seen at L.M. Ericsson Finland (source: Petri Jokela, LMF) IPv4 access network WWW Proxy HIP CN Internet HIP MN IPv6 access network 2011-09-13 AK T-110.6120: Publish/Subscribe Internetworking Music Server 17 DiffServ • Differentiated Services (DiffServ, RFC 2474) redefines the ToS octet of the IPv4 packet or Traffic Class octet of IPv6 as DS • The first 6 bits of the DS field are used as Differentiated Services Code Point (DSCP) defining the Per-Hop Behavior of the packet • DiffServ is stateless (like IP) and scales • Service Profiles can be defined by ISP for customers and by transit providers for ISPs • DiffServ is very easily deployable and could enable well working VoIP and real-time video • Unfortunately, it is not used between operators 2011-09-13 AK T-110.6120: Publish/Subscribe Internetworking 18 Distributed Hash Table (DHT) • Distributed Hash Table (DHT) is a service for storing and retrieving key-value pairs • There is a large number of peer machines • Single machines leaving or joining the network have little effect on its operation • DHTs can be used to build e.g. databases (new DNS), or content delivery systems • BitTorrent is using a DHT • The real scalability of DHT is still unproven • All of the participating hosts need to be trusted (at least to some extent) 2011-09-13 AK T-110.6120: Publish/Subscribe Internetworking 19 DHT The principle of Distributed Hash Table (source: Wikipedia) 2011-09-13 AK T-110.6120: Publish/Subscribe Internetworking 20 Some More Revolutionary Approaches 1. ROFL M. Caesar, T. Condie, J. Kannan, K. Lakshminarayanan, I. Stoica, and S. Shenker, ROFL: Routing on Flat Labels, In ACM SIGCOMM, Sep. 2006, pp. 363–374 2. DONA T. Koponen, M. Chawla, B.-G. Chun, A. Ermolinskiy, K. H. Kim, S. Shenker, and I. Stoica, A Data-Oriented (and Beyond) Network Architecture, In SIGCOMM ’07: Proceedings of the 2007 conference on Applications, technologies, architectures, and protocols for computer communications, New York, NY, USA, 2007, pp. 181-192 2011-09-13 AK T-110.6120: Publish/Subscribe Internetworking 21 ROFL • ROFL routes directly on host identities, leaving aside the locations of the hosts • Self-certifying identifiers (tied to public keys) • Create a network layer with no locations • Advantages: – No new infrastructure (no name resolution) – Packet delivery only depends on the data path – Simpler allocation of identifiers (just need to ensure uniqueness) – Access control based on identifiers 2011-09-13 AK T-110.6120: Publish/Subscribe Internetworking 22 ROFL • Three classes of hosts: – Routers – Stable hosts – Ephemeral hosts • Each ID is resident to its Hosting Router (the host’s first-hop router) • The hosts form a two-way ring – each with pointers to its successor and predecessor • There can be shorter routes cached • OSPF-like routing protocol (w/ network map) is assumed for recovering from routing failures • Global ROFL-ring for inter-domain routing 2011-09-13 AK T-110.6120: Publish/Subscribe Internetworking 23 DONA • DONA replaces the hierarchical DNS namespace with a cryptographic, self-certifying namespace for naming data • This enables entirely distributed namespace control • The namespace is not totally flat but consists of two parts: the principal’s identifier and a label • This two-tier hierarchy helps make DONA scalable • Clean-slate naming and name resolution 2011-09-13 AK T-110.6120: Publish/Subscribe Internetworking 24 DONA • Strict separation between naming (persistence and authenticity) and name resolution (availability) • Each principal has a public-key pair • Each datum (or any other named entity) is associated with a principal • Names of the form P:L (Principal:Label), where P is a cryptographic hash of the principal’s public key and L is a locally unique label • Name resolution by Resolution Handlers, primitives: FIND(P:L), REGISTER(P:L) 2011-09-13 AK T-110.6120: Publish/Subscribe Internetworking 25 Networking Named Content • Based on: Van Jacobson, V.; Smetters, D. K.; Thornton, J. D.; Plass, M. F.; Briggs, N.; Braynard, R. Networking named content. Proceedings of the 5th ACM International Conference on Emerging Networking Experiments and Technologies (CoNEXT 2009); 2009 December 1-4; Rome, Italy. NY: ACM; 2009; 1-12. http://conferences.sigcomm.org/co-next/2009/papers/Jacobson.pdf Warm thanks to Van for providing the figures and allowing me to use them! 2011-09-13 AK T-110.6120: Publish/Subscribe Internetworking 26 Content-Centric Networking (CCN) • CCN – a communication architecture built on named data • “Address” named content – not location • Preserve the design decisions that make TCP/IP simple, robust and scalable • From IP to chunks of named content • Only layer 3 requires universal agreement 2011-09-13 AK T-110.6120: Publish/Subscribe Internetworking 27 TCP/IP and CCN Protocol Stacks Source: Van Jacobson, PARC, 2009 2011-09-13 AK T-110.6120: Publish/Subscribe Internetworking 28 Interest and Data packets • There are two types of CCN packets: – Interest packets – Data packets Source: Van Jacobson, PARC, 2009 2011-09-13 AK T-110.6120: Publish/Subscribe Internetworking 29 CCN Node Model • There are two types of CCN packets: – Interest packets – Data packets • Consumer broadcasts its Interest over all available connectivity • Data is transmitted only in response to and Interest and consumes that Interest • Data satisfies an Interest if ContentName in the Interest is a prefix of that in the Data 2011-09-13 AK T-110.6120: Publish/Subscribe Internetworking 30 CCN Node Model • Hierarchical name space (cmp w/ URI) • When a packet arrives on a face a longestmatch lookup is made • Forwarding engine with 3 data structures: – Forwarding Information Base (FIB) – Content Store (buffer memory) – Pending Interest Table (PIT) 2011-09-13 AK T-110.6120: Publish/Subscribe Internetworking 31 CCN Node Model • FIB allows a list of outgoing interfaces – multiple sources of data • Content Store w/ LRU or LFU replacement • PIT keeps track of Interest forwarded up-stream => Data can be sent downstream • Interest packets are routed upstream – Data packets follow the same path down • Each PIT entry is a “bread crumb” marking the path and is erased after it’s been used 2011-09-13 AK T-110.6120: Publish/Subscribe Internetworking 32 CCN Forwarding Engine Source: Van Jacobson, PARC, 2009 2011-09-13 AK T-110.6120: Publish/Subscribe Internetworking 33 CCN Node Model • When an Interest packet arrives, longest-match lookup is done on its ContentName • ContentStore match is preferred over a PIT match, preferred over a FIB match – Matching Data packet in ContentStore => send it out on the Interest arrival face – Else, if there is an exact-match PIT entry => add the arrival face to the PIT entry’s list – Else, if there is a matching FIB entry => send the Interest up-stream towards the data – Else => discard the Interest packet 2011-09-13 AK T-110.6120: Publish/Subscribe Internetworking 34 CCN Transport • CCN transport is designed to operate on unreliable packet delivery services • Senders are stateless • Receivers keep track of unsatisfied Interests and ask again after a time-out • The receiver’s strategy layer is responsible for retransmission, selecting faces, limiting the number of unsatisfied Interests, priority • One Interest retrieves at most one Data packet => flow balance 2011-09-13 AK T-110.6120: Publish/Subscribe Internetworking 35 Reliability and Flow Control • Flow balance allows for efficient communication between machines with highly different speeds • It is possible to overlap data and requests • In CCN, all communication is local and flow balance is maintained over each hop • This leads into end-to-end flow control without any end-to-end mechanisms 2011-09-13 AK T-110.6120: Publish/Subscribe Internetworking 36 Naming • CCN is based on hierarchical, aggregatable names at least partly meaningful to humans • The name notation used is like URI Source: Van Jacobson, PARC, 2009 2011-09-13 AK T-110.6120: Publish/Subscribe Internetworking 37 Naming and Sequencing • An Interest can specify the content exactly • Content names can contain automatically generated endings used like sequence #s • The last part of the name is incremented for the next chunk (e.g. a video frame) • The names form a tree which is traversed in preorder • In this way, the receiver can ask for the next Data packet in his Interest packet 2011-09-13 AK T-110.6120: Publish/Subscribe Internetworking 38 Intra-Domain Routing • Like IPv4 and IPv6 addresses, CCN ContentNames are aggregateable and routed based on longest match • However, ContentNames are of varying length and longer than IP addresses • The TLV (Type Label Value) of OSPF or IS-IS can distribute CCN content prefixes • Therefore, CCN Interest/Data forwarding can be built on existing infrastructure without any modification to the routers 2011-09-13 AK T-110.6120: Publish/Subscribe Internetworking 39 Intra-Domain Routing An example of intra-domain routing Source: Van Jacobson, PARC, 2009 2011-09-13 AK T-110.6120: Publish/Subscribe Internetworking 40 Inter-Domain Routing • The current BGP version has the equivalent of the IGP TLV mechanism • Through this mechanism, it is possible to learn which domains serve Interests in some prefix and what is the closest CCN-capable domain on the paths towards those domains • Therefore, it is possible to deploy CCN in the existing BGP infrastructure 2011-09-13 AK T-110.6120: Publish/Subscribe Internetworking 41 Content-Based Security • In CCN, the content itself (rather than its path) is protected • One can retrieve the content from the closest source and validate it • All content is digitally signed • Signed info includes hash of the public key used for signing • We still need some kind of a Public Key Infrastructure (PKI) 2011-09-13 AK T-110.6120: Publish/Subscribe Internetworking 42 Trust Establishment Associating name spaces with public keys Source: Van Jacobson, PARC, 2009 2011-09-13 AK T-110.6120: Publish/Subscribe Internetworking 43 Evaluation • The CCN architecture described has been implemented and evaluated • Voice over CCN and Content Distribution were tested with small networks • The results are interesting but not alone convincing regarding the scalability of the design • There still are some fundamental questions that remain unanswered 2011-09-13 AK T-110.6120: Publish/Subscribe Internetworking 44 Voice over CCN • Secure Voice over CCN was implemented using Linphone 3.0 and its performance evaluated • Caller encodes SIP INVITE as CCN name and sends it as an interest • On receipt of the INVITE, the callee generates a signed Data packet with the INVITE name as its name and the SIP response as its payload • From the SIP messages, the parties derive paired name prefixes under which they write RTP packets • There is a separate paper on Voice over CCN 2011-09-13 AK T-110.6120: Publish/Subscribe Internetworking 45 Merits of CCN • Very simple and understandable scheme • Shown to work also with streamed media • Clever reuse of existing mechanisms • Easy to implement based on current routing software • Easy to deploy on existing routing protocols and IP networks • Easy, human-readable naming scheme 2011-09-13 AK T-110.6120: Publish/Subscribe Internetworking 46 Concerns about CCN • The simple hierarchical (URI-like) naming scheme is built deep into the design • Will it scale to hundreds of billions of nodes? – Flooding (send out through all available faces) – Flow balance – an Interest for every Data – How large can the FIB grow (soft state)? – Data takes the same (possibly non-optimal) path as Interest – assuming two-way links • Are the performance measurements made with only a couple of hosts convincing? • Security architecture looks quite conventional 2011-09-13 AK T-110.6120: Publish/Subscribe Internetworking 47