Mobility Management in Cellular Networks and MAP - Mobile Application Part Mobility Management in GSM GSM (2+ …) services Short Message Service Support of GPRS Raimo Kantola – S- 2015 Signaling Protocols 9 - 1 Summary of course scope H.323 or SIP or SIP ISUP IP IP HLR/ CAS, R2 Control Part of an Exchange HSS PABX Or Call Processing MAP ISDN Server CCS7 ISUP V5 INAP AN Megaco/MGCP/… Media Gateway SCP circuit or Switching Fabric packets Raimo Kantola – S- 2015 Signaling Protocols 9 - 2 Mobility Management in General Comparison of solutions for CS and PS networks Raimo Kantola – S- 2015 Signaling Protocols 9 - 3 Requirements for MM Business • Scalability – this requirement drives requirements the design – scaling to Billions of users – scaling to frequent moves of users Reliability Performance – scaling to battery powered devices • Performance – Seamless moves from cell to another system !duration of break< N ms – power saving of mobiles • Reliability – Success rate of handovers Scalability Flexibility • Business – Trust: Mobility can not be implemented without authentication and security – Roaming has been a killer! Raimo Kantola – S- 2015 Signaling Protocols 9 - 4 Analysis tree links signaling to routing From signaling: ABC – destination A ABCd – shortest subscriber number ABCdefgh – longest subscriber nr Buckets B C We assume that d the analysis is done e using a tree structure similar to this. f g In this mobility design: Last node points to a bucket h file that contains the physical location of the user (routing address) i.e. if logical numbers are also used as routing numbers ! we will show that this does not scale Raimo Kantola – S- 2015 Signaling Protocols 9 - 5 Mobility requires logical subscriber numbers - are mapped dynamically to network topology bound routing numbers • For most nodes it is enough to understand only the prefix of the routing number (topological proximity = proximity in number space). • Example: 6*109 subscribers, number length = 13 digits Rough memory estimate for the analysis tree based on dialed digits (no separate routing numbers). Tree is made of nodes of 64 octets. One node is used to analyse one dialed digit Use of numbering space: on average 5…6 values in each position are used m13 = 6 *109 13 lg m = 9 + lg 6 m = 5.65 Nrof nodes in the tree is (m is also the branching factor!) m13 - 1 1 + m + m2 + … m12 = ≈ 1.3 * 109 m - 1 Raimo Kantola – S- 2015 Signaling Protocols 9 - 6 Analysis tree calculus cont ... Memory requirement is 64 bytes * 1.3*109 = 83 GB • Need to be available for any calls in all nodes: replication will be expensive! • A single read with full number requires 13 memory references, is not a problem • Maintaining replicas is the problem: 4000 Mbit/s Assumptions: 3500 - an update takes a 50 bytes msg 3000 - all updates in 6 hours (of 24h) 2500 NB: 2000 5B - updates/subscriber may 6B need to be done significantly 1500 more often. 1000 500 Problem needs to be 0 0 5 10 15 20 25 30 35 partitioned! Nrof updates per day Raimo Kantola – S- 2015 Signaling Protocols 9 - 7 What if subscriber numbers are binary? • Example: 109 subs, sub nr length is 128 bits Rough memory estimate for analysis: Analysis tree is made of nodes of 64 octets, each for analysing 4 bits. Usage of hexa code points: mN = 109 N lg m = 9 m = 13.34 …1.9, when N goes from 8 to 32 Nrof nodes in the tree is mN - 1 1 + m + m2 + … mN-1 = = 114 to 4290 million m - 1 Result is of the same order of magnitude as for decadic numbers! NB: the branch factor is rounded up to the next integer Raimo Kantola – S- 2015 Signaling Protocols 9 - 8 Other problems in the design • When the number of subscribers grows, memory allocated to the Number Analysis needs to be upgraded by all operators. • For any single operator, most of the entries in the database are practically useless while a small portion is in active use – many national calls vs. few international calls – Data mining is possible: Every operator can learn something about every subscriber of every other operator ! business secret? • Update traffic (e.g. 100…1500 Mbit/s) per operator takes quite a bit of network capacity (expensive in PCM environment although in the times of Broadband 1Gbit/s is no big deal). Raimo Kantola – S- 2015 Signaling Protocols 9 - 9 In GSM the DB is partitioned by Operator and by Prefix of MSISDN nr • An HLR per a few 100 000…1M subscribers – Operator code + prefix map to HLR • MS-ISDN = directory number = what you dial is mapped to Mobile Subscriber Routing Number (MSRN) by VLR per call or per visit to another network – MSRNs are topology bound numbers ! any exchange (ISDN, GSM or PSTN) can easily use MSRNs to route calls. – MSRNs are managed and allocated by VLRs and not visible to users. • Location area hierarchy decreases nrof updates – Not all location changes need be told to HLR Raimo Kantola – S- 2015 Signaling Protocols 9 - 10 Rough calculus of location update traffic in an HLR with 200 000 subs • 200 000 subscribers • 1 update/5min/subscriber • Rough upper bound estimate: – let one update = 100 octets Traffic = 200 000 * 100 * 8/(5*60) = 0,53Mbit/s. Can be transported on a single PCM-line (2 Mbit/s)! ! Makes sense, is clearly feasible. NB: Existence of HLR and Request to HLR to map MSISDN to MSRN means that this mapping for a subscriber needs to be maintained in two places only: the VLR and the HLR. Raimo Kantola – S- 2015 Signaling Protocols 9 - 11 Location Area Hierarchy in GSM reduces the need for HLR updates MSC/VLR-area HLR knows MSC/VLR Visitor Location Register (VLR) knows SijaintialueLocation Area - = a set of cells - update once/6 min….24h and when CellRouting area power switched on/off etc. - updates need to be authenticated SoluCell In case of GPRS, the SGSN knows the Routing Area = a subset of Location area. Final location is found by paging: - call is sent to all cells in LA - MS receives in favorite cell - cell with best connectivity is chosen Raimo Kantola – S- 2015 Signaling Protocols 9 - 12 Mobility from IP network point of view (1) • Packet forwarding/packet is based RT on routing tables (RT). • Routers maintain RTs by routing protocols. destination-IP addr • Feasible size of the RT is 100 000 … Outgoing I/f/ Next hop IP-addr 500 000 entries = rows. – Bigger RT ! more expensive Routers • Longest match search/packet takes many memory reads (<32). - 1-4 B users !provider addressing (i.e. addresses are aligned with network topology) results feasible RT size - search is based on address prefix not a full 32 bit address When an MS moves, its (topology bound) IP address must change. If not, the RT would need to store host addresses of all MSs! These MS address entries would need to be maintained in all core routers ! not feasible! Raimo Kantola – S- 2015 Signaling Protocols 9 - 13 Mobility from IP network point of view (2) • TCP sessions are identified by – Source-IP-add, Source-port, dest-IP add and dest port • Internet follows provider addressing for scalability reasons – i.e. IP addresses allocation follows network topology closely ! route aggregation in the network core. – Exception that violates topology aligned addressing is multi-homing: e.g. one corporate network connected to at least two ISP networks results that the IP- address aggregate of this corporation becomes visible in the non-default routing core of the Internet – A movement of a mobile node from one network to another means that a new IP address is allocated to the Mobile node. The result is that an ongoing TCP session will fail. Possible solutions to this are: Mobile-IP, tunneling, a new transport protocol instead of TCP etc. – GPRS and 3G WCDMA use tunneling, US CDMA uses Mobile IP, an example of a new transport protocol is SCTP, etc. – Tunneling = carry IP-packets(with non-changing IP addr) inside IP-packets with IP-address that changes with movements from cell to cell • The problem with a large number of entries in a core routing table is not the size itself, rather the problem is maintaining all these route entries up to date at all times in all nodes in the non-default core. Raimo Kantola – S- 2015 Signaling Protocols 9 - 14 Mobile IP • Variants for IPv4 and IPv6 • Is a tunneling solution on layer 3 = within IP protocol itself • Uses IP addresses as Identifiers. • Leads to triangle routing: – Correspondent Node! Home Agent ! Mobile Node!Correspondent Node !! Route optimization + Hierarchical Solutions Raimo Kantola – S- 2015 Signaling Protocols 9 - 15 Nrof probable handovers from cell to cell during a telephone call NrofKanavanvaihtojen hops from cell määrä to cell Call duration 3 min 100 10 1 Speed 150 km/h 0 1 2 3 4 5 6 7 8 Speed 100 km/h 0,1 Speed 50 km/h Speed 15 km/h Speed 5 km/h 0,01 radiusSolun of the säde cell km in km An architecture with less than one handover on average makes sense! Raimo Kantola – S- 2015 Signaling Protocols 9 - 16 Power saving is important for mobiles • Sending and complex processing consume most power. • The more the mobile can sleep the better. • Small cells lead to frequent location updates ! power consumption increases • If main processor of the Mobile needs to wake up each 30s ! battery will be flat in <<12h (different brands behave differently in terms of power consumption) Conclusion from the slide with nrof hops from cell to cell as a function of cell size: Systems that do not allow building cells with the radius in kilometers can not succeed for voice services.