180 likes | 340 Views
Welcome SMS. Welcome SMS in Erlang. Experiences of Rapid Deployment in a GSM network. The Requirement. Send out SMS messages to One 2 One customers when they arrive in a foreign network. Don’t send them any more messages for at least two weeks.
E N D
Welcome SMS Welcome SMS in Erlang. Experiences of Rapid Deployment in a GSM network
The Requirement • Send out SMS messages to One 2 One customers when they arrive in a foreign network. • Don’t send them any more messages for at least two weeks. • Different messages depending on destination Country. • “Marketing” usable GUI configuration. • Option to send messages to roaming customers when they arrive in the UK. • Ultra reliability not a hard requirement • Except NOT sending multiple messages
The Timescale • 8 (yes, 8) weeks from start of design to launch of service.
Access to Customer Behaviour? • Monitor all international c7 signalling links. • Extract contents of MAP messages: • Where is the customer? (VLR Address) • Whose customer is it (IMSI) • Who should receive the SMS? (MSISDN) • VLR address and IMSI are in Location Update, MSISDN is in Insert Subscriber Data. These messages can even take different physical routes...
Business Logic • Decide which message to send • Different depending on class of customer (prepay, postpay, wholesale) • Different depending on destination Country • Different for inbound roamers • Have we sent them one already in this country? • Have they elected not to receive these messages
Operational Requirements • Integrated with existing alarm management system (SNMP Traps) • Generate statistic reports of system activity into existing stats database. • Load balancing between SMS service centres • Ability to on the fly reconfigure which SMS service centres are active and the load balancing between them.
Design Options • Commercial c7 probes to do automatic correlation of all messages for a particular activity. • Best supplier delivery time longer than entire project duration. • Time taken to integrate.. Who knows. • Option to exercise Synergy across T-Mobile International group. • Several useful design ideas shared. Again, timescales not lined up • Standard commercial systems - expensive, don’t meet all requirements, integration time huge
Design Options 2 • Use Erlang? • But what about c7 probes? • Possibility to use standard test set (MPA 7400) to extract all c7 messages as raw data and send out as UDP • Two spare (but old) 4 CPU Sun machines about to be de-commissioned • Lead times of all elements OK. Now to work...
Design Overview SMS Service Centres Customer Database Also Erlang! TCP (Internal Protocol) TCP (GIP Protocol) Erlang - A Erlang - B UDP c7 UDP c7 C7 Probe C7 Probe
Erlang Nodes SMS Service Centre wsms_logic@bruce wsms_logic@burns sms sms db_read db_read inets mnesia stats snmp inets mnesia stats snmp Customer dB Failover logic logic correlator@bruce correlator@burns Failover correlator correlator gn_agent1@bruce gn_agent2@bruce gn_agent1@burns gn_agent2@burns C7 decode C7 decode C7 decode C7 decode Bruce Burns Active Standby UDP c7 messages
Problems encountered • Bug with restart of one process starting up too many sms sending processes. • LAN Latency - both for UDP and dB reads. • Fixed by tuning concurrency and timeouts. • One module not permanently loaded in the customer Database. • Scanning an Mnesia table to delete some rows is pretty heavy..
What else have we been up to? • Oracle Call Interface Binding to Erlang • Multithreaded Driver • Pools of Sessions • Decent Performance (> 1000 selects per second) • 1500 lines of C (so far!) • Garbage collection of stale handles
What else have we been up to? • Oracle Call Interface Binding to Erlang N = “447956123456”, F=fun() -> case oci:exec(“select name, balance from cust where telno = :e”, [N]) of [{Name, Amount}] -> A1 = Amount - 10.0, oci:exec(“update cust set balance = :e where telno = :e”,[A1, N]) [] -> oci:abort(“Customer not found”) end, oci:transaction(Connection, F).
What else have we been up to? • Generic “behaviour” for tcp/ip clients • link management (using heartbeats) • simple callbacks for protocol implementation • model is all transactions multiplexed down 1 socket • time taken to implement new protocol down to a few days • Call Data Record - record and playback
What else have we been up to? • The Original Project used to justify the use of Erlang never got into service. All hardware has been re-deployed for other Erlang applications!