IP switching and flow classification

Transcription

1 1 IP switching and flow classification S Lecture Mika Ilvesmäki, Laboratory of Telecommunications Technology 2 Contents Problem statement IP router structure Advances in router architectures IP on ATM - the solutions IP flows IP switching Traffic based IP switching Topology based IP switching Flow classification Levels of aggregation Methods Distributing the criteria Conclusions

2 3 Problem statement Internet is growing Both the number of users and the amount of traffic per router Increasing number of routers increases the number of hops s are heavily loaded Several algorithms and methods developed to ease the router workload Routing is one way to manage resources in the Internet 4 IP on ATM How to map IP on ATM? IP over ATM RFC 1577, simple ATM Forum LAN emulation ATM Forum Multiprotocol over ATM ATM Forum solutions are usually heavy resource users IP switching Requires IP functionality AND connection-oriented technology, such as ATM

3 5 Basic Internet architecture (Edge) routers connected to (core) routers Edge routers /SW /SW /SW Core network 6 IP router architecture The route is looked up for every packet This is robust but slow TCP flow control Routing Timeouts, RTT tables Route look-up Forwarding load ICMP source - action performed for every packet Taking Forwarding the packet the packet quench (theory) Routing load - action performed based on routing protocol implementation OSI layer 3 OSI layer 2

4 7 Improvements to IP routers Caching the routing decisions Gigabit-solution Implementing functionality on the hardware Distributed architectures Packet classification algorithms Speeding up the routing table search 8 IP flow - definition Flow (packet train): Packets traversing from the same source to the same destination. within a time interval (flow timeout) Granularity levels in IP networks What constitutes as a source or destination TCP ports & IP addresses, essentially IP address pair HTTP, 80 applications Telnet, 23 IP addresses FTP-data, 20 SSH, 22

5 9 IP switching a.k.a multi-layer routing Reduce the workload by switching the flow on a lower level, i.e. before it reaches the forwarding processor Internet model Map flows to (ATM) connections or to VPs Topology Traffic Routing switching switching and VCs Switching is quicker than routing Slow changes Routing PVC SVC Handover Telephony Two approaches: Traffic and Topology model Datagrams Fast changes 10 Traffic based IP switching Usually but not necessarily high level granularity(tcp+ip) Traffic induced decisions -> Reactive Near-optimal usage of connections and QoS resources Depends on the definition of flow and flow timeout Might overload connection space if too high a level of granularity Global decisions on flow classification -> signalling, RSVP Local decisions -> no signalling and we have...

6 11 Case: Ipsilon IP switching Nokia IP routing, production line terminated RFC 1953: Ipsilon Flow Management Protocol (IFMP) Notifies adjacent routers about classified flows Requests cut-through connections RFC 1987: General Switch Management Protocol Controlling the switching layer Realizing local decision making New GSMPv2 in planning phase QoS capable 12 Ipsilon IP switching in action Normal routing (1), cached routing(2) and cut-through routing(5) Soft state -> no signalling, timers tear down connections Upstream IP switch 2b flow classification IFMP redirectconventional router 4b IFMP redirect Downstream OSI L3 ATM-switch GSMP 1 1 default VC 2a 3 flow specific VC 5 4a L2 L1

7 13 Verification of the concept Traffic measurements IP address pair granularity 10% of the longest flows carry 90% of the packets 70% of the shortest flows carry only 5% of the packets Switch the longest flows The problem of flow classification Reduces router workload 14 Traffic measurements Measurements made in HUT Detailed analysis may be found The cumulative proportions of packets and flows compared to flow duration 100,00 % 90,00 % Basic assumption of IP switching: Map 80% of packets to 20% of flows 80,00 % 70,00 % 60,00 % 50,00 % 40,00 % 30,00 % %-Pkts %-Flows 20,00 % 10,00 % 0,00 % Flow duration in seconds

8 15 Ipsilon IP switching problems Only prioritization No absolute QoS Edge devices heavily loaded Local decisions -> routed hops possible and other problems related to local decision making. 16 Lecture break Things to follow Topology based IP switching Flow classification Welcome back

9 17 Topology based IP switching Usually but not necessarily low level granularity (IP address prefixes) Routing protocol based Proactive, fully connected VC mesh Might keep up empty connections Multiple connections must exist if QoS is to be supported Only global criteria on flow classification make sense 18 Problems in the Topology approach Routing loops How to spread the knowledge of the topology The workload on the edge routers Sensitivity to router malfunctions (in the core)

10 19 Case: Cisco Tag Switching RFC 2105: Tag Switching Architecture Edge routers, Core routers, Tag Information Base (TIB) Tag Distribution Protocol (TDP) TIB Edge routers TDP /SW /SW TIB Core network /SW 20 Tags might be ATM VP/VC fields Cisco Tag Switching in action OSI 3 OSI 2 Edge router Switch TIB Destination ntw link tag DATA DATA OSI 3 OSI 2 Core router Switch TIB InTag link Out Tag DATA DATA

11 21 Technological solutions Nokia IP routing a.k.a Ipsilon IP switching CANCELLED!! Cisco Tag switching, IBM ARIS Toshiba CSR, Bay Networks IP navigator etc. MPLS-standards IETF WG, Work still in progress 22 Flow classification But, after a flow is classified what kind of QoS/CoS parameters are used? This is not easy, let alone relatively easy Source characterization Division to routed and prioritized (switched) traffic is relatively easy Objectives Ease the router workload, Help the user, provide QoS/CoS

12 23 Levels of flow aggregation IP address prefix IP address prefixes AND service identifiers (i.e. TCP ports) Multicast trees IP address prefix HTTP, 80 FTP-data, 20 Virtual Path, VP Telnet, 23 SSH, 22 IP address prefix HTTP, 80 X-win, Virtual Channels, VCs 24 Methods of flow classification I All flow All flows classified, fine granularity flows 100% level for performance evaluation Packetcount & Adaptive packetcount Wait for X pkts on a flow candidate before prioritizing Change X according to resource usage

13 25 Methods of flow classification Static Service set A set of services is prioritized Dynamic Service set Service set changes according to pkt/flowbehavior dynamism achieved applying the LVQ algorithm 26 Restrictions for the use of flow Connection oriented VC space, VC setup capacity Connectionless environment Route lookups, Packets prioritized, services Bandwidth Delay to setup flows Connection oriented technologies Services 50% Connectionless technologies Flows 50% 50% Prioritization 50% Flow space

14 27 Flow Classification - router performance Resource oriented Use the resources of the network as effectively as possible Flows containing a lot of packets should be switched not to burden the routing process Limiting factors: VC space, processor capacity, bandwidth, flow setup delay Flow Granularity changes Packet count, Adaptive packetcount, Ambiguous effect on users 28 Flow Classification - user satisfaction User based Flow classification should benefit the user first and the network second Rational choice of flows -> Flows important to users should be switched How to define important? Your Internet connection request can not be completed at the moment. Please try again as soon as some resources are made available somewhere in the net. Ambiguous effect on resource usage effects on users too Static and the dynamic service set classifiers Combination of user and resource

15 29 Spreading information - local approach How to spread the info of the flow mappings to the switching layer Local decisions: Let each router decide whether to switch or not Internet solution, connection state may be limited to two or three routers (reduced functionality and benefits). Hop by hop decisions, policy based prioritization. 30 Spreading information - global approach Global decisions: Negotiate the flow classification criteria with other entities Distribute tables of labels (tags) to a group of routers (core routers) hop by hop or explicit routing QoS connections for QoS differentiation The edge routers remove IP switching encapsulation (if any).

16 31 Problem issues with IP switching How much of the functionality existing in the lower layer(s) is taken into the concept Signaling, QoS features, traffic management, VC -space usage, flow granularity, flow setup capacity Maintaining state in the network (RSVP) Does it make any sense Is it Internet? 32 Summary workload deduction methodology Offers a possibility to Quality of Service Several unresolved issues exist Introduces the concept of an IP flow Makes or could make use of the connection oriented layer 2 technologies ATM, IP over SDH, ISDN Standardization is on the way Competing solutions on the edge of release

17 33 Information sources on the WWW RFC s and Internet drafts General info on IP switching and related research