Concurrent programmin. ng xp. The industrial experience

Size: px

Start display at page:

Download "Concurrent programmin. ng xp. The industrial experience"

Imogen Shepherd
5 years ago
Views:

1 Concurrent programmin ng xp The industrial experience

2 Karol Ostrovský M.Sc. Comenius University, Bratislava Ph.D. Chalmers Post-doc Chalmers System Designer Dfind, on assignment at Ericsson Operations & Maintenance Subsystem Guest Lecture PPxT VT2011 Karol Ostrovsky Ericsson AB Page 2

3 Karol Ostrovský Language skills Basic Pascal C/C++ Scheme SmallTalk Java JR (MPD) Haskell Erlang Ocaml LaTeX VAX assembler Trilogy Ada Agdag Guest Lecture PPxT VT2011 Karol Ostrovsky Ericsson AB Page 3

4 The chalmers years Research in static analysis of concurrent programming languages Type systems Protocol analysis Main course responsible PPxT Developed the course between 2005 and 2010 Guest Lecture PPxT VT2011 Karol Ostrovsky Ericsson AB Page 4

5 The case of the missing slide The Message from this Course Should you forget everything from this course, please, remember at least this saying: Use the right tool for the job. PPVT10 Introduction 3 Guest Lecture PPxT VT2011 Karol Ostrovsky Ericsson AB Page 5

6 What does this buy you? Course Goals General Introduction to the problems common to many computing disciplines: Operating systems Distributed systems Real time systems Appreciation of the problems of concurrent programming g Classic synchronisationn problems PPVT10 Introduction 9 Guest Lecture PPxT VT2011 Karol Ostrovsky Ericsson AB Page 6

7 What does this buy you? Course Goals Practical Understanding of a range constructs for concurrent of programming language programming Ability to apply these in practice to synchronisation problems in concurrent p rogramming g Practical knowledge of the programming techniques of modern concurrent pro ogramming languages PPVT10 Introduction 10 Guest Lecture PPxT VT2011 Karol Ostrovsky Ericsson AB Page 7

8 what does this really buy you? Everything this course teaches is used in industrial practice As is Erlang lectures almost production code Applied or customised Resource allocation Master slave Timeouts, alarms Guest Lecture PPxT VT2011 Karol Ostrovsky Ericsson AB Page 8

9 The pthreads experience SGSN shell in C Problem: Unix signal handling was insufficient Solution: create a separate thread for signal handling My yproblem I never practically used pthreads before My solution general solution based on co oncurrency principles - synchronisation, mutual exclusion, message passing translate to pthreads Guest Lecture PPxT VT2011 Karol Ostrovsky Ericsson AB Page 9

10 The general solution Shared data Main Handler Guest Lecture PPxT VT2011 Karol Ostrovsky Ericsson AB Page 10

11 Pthreads translation init lock Mutex variable unlock destroy Main create join unlock Handler Guest Lecture PPxT VT2011 Karol Ostrovsky Ericsson AB Page 11

12 Ericsson where am i? Business Unit Networks Development Unit IP and Broad band Product Development Unit Packet Core - SGSN-MME O&M sub-system 2G sub-system 3G sub-system... - GGSN-MPG - CPG -... Guest Lecture PPxT VT2011 Karol Ostrovsky Ericsson AB Page 12

13 Mobile telecom network Guest Lecture PPxT VT2011 Karol Ostrovsky Ericsson AB Page 13

14 packet core network 3GPP Defines standards (mostly proto ocols) Interoperability is essential SGSN-MME Servicing GPRS Support Node (2G/3G) Mobility Management Entity (4G G) Control signalling Admission control, Authentication Mobility, roaming Payload transport (not in 4G) Guest Lecture PPxT VT2011 Karol Ostrovsky Ericsson AB Page 14

15 SGSN-MME mkvi 3 sub-racks 21 blades (2+19) 2 core PowerPC ~ 114 simultaneously running processes Backplane: 1Gbps Up to 3 MSAU Guest Lecture PPxT VT2011 Karol Ostrovsky Ericsson AB Page 15

16 SGSN-MME MkViii 3 sub-racks 14 blades (2+12) 6 core Intel x86 12 SMT threads total ~ 432 simultaneously running processes Backplane: 1 or 10Gbps Guest Lecture PPxT VT2011 Karol Ostrovsky Ericsson AB Page 16

17 SGSN-MME architecture sketch NCB NCB FS SB FSB AP AP... AP DP DP... DP Guest Lecture PPxT VT2011 Karol Ostrovsky Ericsson AB Page 17

18 Sieve of Eratosthenes Architecture nums N+1 pipeline channels One shared output channel filter1 filter2 filtern print output eat PPVT10 Message Passing 46 Guest Lecture PPxT VT2011 Karol Ostrovsky Ericsson AB Page 18

19 pipeline of processes NC CB AP_1 AP_2 AP_N LOG log gging g Guest Lecture PPxT VT2011 Karol Ostrovsky Ericsson AB Page 19

20 Typical concurrency patterns One mobile one process (replicated worker) Isolation Synchronisation only with resources Central resources Resource allocator Master/coordinator slave/worker Transaction handler Node management Configuration management Fault management (e.g. barrier sync. during start-up) Timeouts, alarms (e.g. heartbeat) Guest Lecture PPxT VT2011 Karol Ostrovsky Ericsson AB Page 20

SGSN-MME A huge version of the assignment 2 The pub SGSN-MME node Customers mobile phones The landlord NCB The assistant node management The tables system resources

21 SGSN-MME A huge version of the assignment 2 The pub SGSN-MME node Customers mobile phones The landlord NCB The assistant node management The tables system resources Extension of the assignment SGSN-MME can run in a pool Operations & Maintenance? multiple-bars? Guest Lecture PPxT VT2011 Karol Ostrovsky Ericsson AB Page 21

22 Operations & Maintenance Configuration management Maintaining working and consist tent settings Logging Black box (flight data recorder) Fault management Alarm handling Performance management Gathering statistics Monitoring health Guest Lecture PPxT VT2011 Karol Ostrovsky Ericsson AB Page 22

23 Haskell patterns monads Good Keeps pure and side-effecting effecting c omputations apart Good separation of concerns Improved compositionality Possible performance gain Gather writes together and write to DB once amortise the cost of transactions: Bad 1 item write costs items write is not 100 but only 20! In rapid prototyping it can present a big hurdle to jump over So, it is good that Erlang does not have static types Guest Lecture PPxT VT2011 Karol Ostrovsky Ericsson AB Page 23

24 OO-design patterns Factory method Improve memory sharing Object pool Bounded parallelisation of algorithms thread pool Overload protection Guest Lecture PPxT VT2011 Karol Ostrovsky Ericsson AB Page 24

25 What they do not teach you Software maintenance Software lives long Especially telecom systems (decades) Banking systems live even longer (think COBOL) People change Organisations change Hardware changes Requirements change Documentation often does not change Guest Lecture PPxT VT2011 Karol Ostrovsky Ericsson AB Page 25

26 Software maintenance The developer s challenge Write simple (readable) and effic cient code: 1. Write a straightforward and working solution first 2. Optimise later (or even better skip this step) Think smart but do not over-optimise Optimisations complicate mainte enance The code is often the only reliable document Types can be very good documentation Guest Lecture PPxT VT2011 Karol Ostrovsky Ericsson AB Page 26

27 Synthesis and analysis Emphasis on synthesis so far Software development Perhaps 30% of manpower is spent on testing Analytical work Do you like to break a system? But testing ti can also be synth hetical Testing frameworks Quickcheck SGSN-MME has its own Would you like to formally prove the system correct? Guest Lecture PPxT VT2011 Karol Ostrovsky Ericsson AB Page 27

28 Erlang in practice pros Well suited for Control handling of telecom traff fic Application layer (OSI model) applications Web servers, etc. Domain Specific Language framework Test scripting Reasonably high-level (as co mpared to for example C) Good for software maintenance OTP includes useful building blocks Supervisor Generic server Finite state machine Guest Lecture PPxT VT2011 Karol Ostrovsky Ericsson AB Page 28

29 Erlang in practice cons Hard to find good Erlang programmers (?) A bit too low-level language Given current HW limitations one must sometimes optimise to the point where the code is not portable (with the same performance) Raise the abstraction and provide a customisable compiler, VM Where is the type system? A static type system of Haskell-nature would be a hindrance But good static analysis tools are desperately needed Guest Lecture PPxT VT2011 Karol Ostrovsky Ericsson AB Page 29

30 More than true Sayings The greatest performance improvement of all is when a system goes from not working to working The only thing worse than a problem that happens all the time is a problem that doesn't happen all the time PPVT10 Introduction 54 Guest Lecture PPxT VT2011 Karol Ostrovsky Ericsson AB Page 30

31 Concurrent programming xp You are finishing a very good course Provides great general/theoretic cal background Mixed with almost production level practical examples The industrial experience It is more difficult that you expect, but Good systematic approach helps Lateral thinking is an asset Guest Lecture PPxT VT2011 Karol Ostrovsky Ericsson AB Page 31

concurrent programming XXL

concurrent programming XXL Industrial Use of Erlang Introduction Karol Ostrovský (karol.ostrovsky@gmail.com) Motivation Machine Ericsson Blade System 3 sub-racks 14 blades 2 routers 12 compute nodes 6