ERLANG EVOLVES FOR MULTI-CORE AND CLOUD ENVIRONMENTS

Transcription

1 ERLANG EVOLVES FOR MULTI-CORE AND CLOUD ENVIRONMENTS Torben Hoffmann Erlang Solutions 1

2 Agenda Erlang fundamentals Challenges 2 2

3 Warning 1: The Truth I will do a few simplifications in order to get the main points across. 3

4 Warning 1: The Truth 3 I will do a few simplifications in order to get the main points across. 3

5 Warning 1: The Truth Will you tell the truth? 3 I will do a few simplifications in order to get the main points across. 3

6 Warning 1: The Truth Will you tell the truth? Yes 3 I will do a few simplifications in order to get the main points across. 3

7 Warning 1: The Truth Will you tell the truth? Yes The whole truth? 3 I will do a few simplifications in order to get the main points across. 3

8 Warning 1: The Truth Will you tell the truth? The whole truth? Yes No 3 I will do a few simplifications in order to get the main points across. 3

9 Warning 1: The Truth Will you tell the truth? The whole truth? Yes No So help you OTP? 3 I will do a few simplifications in order to get the main points across. 3

10 Warning 1: The Truth Will you tell the truth? The whole truth? So help you OTP? Yes No Yes 3 I will do a few simplifications in order to get the main points across. 3

11 Warning 2: Serious Love Ahead I love Erlang! 4 It was part of a major career shift and I have never looked back. Apologies if I get too intense. 4

12 Realities Of Software Development Time-to-market pressure Scaling successes Maintenance burden Utilisation of computing resources 5 5

13 What Could Be x productivity over C++/Java Seamless scaling on multicore Scaling nicely over machines Less code per feature The future is here today - it s called Erlang! 6

14 What Could Be x productivity over C++/Java Seamless scaling on multicore Scaling nicely over machines Less code per feature The future is here today - it s called Erlang! 6

15 What Could Be x productivity over C++/Java Seamless scaling on multicore Scaling nicely over machines Less code per feature The future is here today - it s called Erlang! 6

16 What Could Be x productivity over C++/Java Seamless scaling on multicore Scaling nicely over machines Less code per feature The future is here today - it s called Erlang! 6

17 What Could Be x productivity over C++/Java Seamless scaling on multicore Scaling nicely over machines Less code per feature The future is here today - it s called Erlang! 6

18 What Could Be... The future is here x productivity over C++/Java Seamless scaling on multicore Scaling nicely over machines Less code per feature The future is here today - it s called Erlang! 6

19 What Could Be... The future is here... The future is Erlang! 6 3x productivity over C++/Java Seamless scaling on multicore Scaling nicely over machines Less code per feature The future is here today - it s called Erlang! 6

20 Erlang s Original Requirements 7 7

21 Erlang s Original Requirements Large scale concurrency 7 7

22 Erlang s Original Requirements Large scale concurrency Soft real-time 7 7

23 Erlang s Original Requirements Large scale concurrency Soft real-time Distributed systems 7 7

24 Erlang s Original Requirements Large scale concurrency Soft real-time Distributed systems Hardware interaction 7 7

25 Erlang s Original Requirements Large scale concurrency Soft real-time Distributed systems Hardware interaction Very large software systems 7 7

26 Erlang s Original Requirements Large scale concurrency Soft real-time Distributed systems Hardware interaction Complex functionality Very large software systems 7 7

27 Erlang s Original Requirements Large scale concurrency Soft real-time Distributed systems Hardware interaction Complex functionality Very large software systems Continuous operation for many years 7 7

28 Erlang s Original Requirements Large scale concurrency Soft real-time Distributed systems Hardware interaction Complex functionality Very large software systems Continuous operation for many years Software maintenance on-the-fly 7 7

29 Erlang s Original Requirements Large scale concurrency Soft real-time Distributed systems Hardware interaction Complex functionality Very large software systems Continuous operation for many years Software maintenance on-the-fly High quality and reliability 7 7

30 Erlang s Original Requirements Large scale concurrency Soft real-time Distributed systems Hardware interaction Complex functionality Very large software systems Continuous operation for many years Software maintenance on-the-fly High quality and reliability Fault tolerance 7 7

31 Erlang s Original Requirements Large scale concurrency Soft real-time Distributed systems Hardware interaction Complex functionality Very large software systems Continuous operation for many years Software maintenance on-the-fly High quality and reliability Sounds familiar? Fault tolerance 7 7

32 Erlang s Original Requirements Large scale concurrency Soft real-time Distributed systems Hardware interaction Complex functionality Very large software systems Continuous operation for many years Software maintenance on-the-fly High quality and reliability Sounds familiar? Sounds good, right? Fault tolerance 7 7

33 General vs Domain Specific Small semantic gap Telecom 8 8

34 General vs Domain Specific Small semantic gap Telecom C++/Java 8 8

35 General vs Domain Specific Small semantic gap Telecom C++/Java 8 8

36 General vs Domain Specific Small semantic gap Telecom C++/Java Erlang 8 8

37 General vs Domain Specific Small semantic gap Telecom C++/Java Erlang 8 8

38 General vs Domain Specific Small semantic gap Telecom C++/Java Erlang Smaller gap = benefits! 8 8

39 Erlang s Sweet Spot GUI Middleware Erlang Coordination Control Drivers 9 Erlang was intended to deal with the control plane in telecom, which is all about orchestration of what goes on. GUI and low-level things are not what Erlang was created for - hence Erlang has good support for integration with other languages. Read the wonderful doctor thesis by Bjarne Däcker if you want to learn more: 9

40 Other Erlang Domains Messaging - XMPP et al - ejabberd, MongooseIM Webservers - Yaws, Chicago Boss Payment switches & soft switches - Vocalink, OpenFlow/LINC Distributed Databases - Riak, CouchDB, Scalaris 10 10

41 Other Erlang Domains Messaging - XMPP et al - ejabberd, MongooseIM Webservers - Yaws, Chicago Boss Payment switches & soft switches - Vocalink, OpenFlow/LINC Distributed Databases - Riak, CouchDB, Scalaris If the tool fits, you must select! Tech Mesh Conference 4-5 December London 10 10

42 To Share Or Not To Share 11 Death propagates in shared memory unless you do a ton of defensive programming. Due to the actor model with no shared memory it is custom in Erlang to do fail-fast programming. No shared memory allows you to fail fast when suitable. Erlang uses message passing between processes to exchange information. 11

43 To Share Or Not To Share Memory 11 Death propagates in shared memory unless you do a ton of defensive programming. Due to the actor model with no shared memory it is custom in Erlang to do fail-fast programming. No shared memory allows you to fail fast when suitable. Erlang uses message passing between processes to exchange information. 11

44 To Share Or Not To Share Memory P1 11 Death propagates in shared memory unless you do a ton of defensive programming. Due to the actor model with no shared memory it is custom in Erlang to do fail-fast programming. No shared memory allows you to fail fast when suitable. Erlang uses message passing between processes to exchange information. 11

45 To Share Or Not To Share Memory P1 P2 11 Death propagates in shared memory unless you do a ton of defensive programming. Due to the actor model with no shared memory it is custom in Erlang to do fail-fast programming. No shared memory allows you to fail fast when suitable. Erlang uses message passing between processes to exchange information. 11

46 To Share Or Not To Share Memory Corrupt P2 11 Death propagates in shared memory unless you do a ton of defensive programming. Due to the actor model with no shared memory it is custom in Erlang to do fail-fast programming. No shared memory allows you to fail fast when suitable. Erlang uses message passing between processes to exchange information. 11

47 To Share Or Not To Share Memory Corrupt 11 Death propagates in shared memory unless you do a ton of defensive programming. Due to the actor model with no shared memory it is custom in Erlang to do fail-fast programming. No shared memory allows you to fail fast when suitable. Erlang uses message passing between processes to exchange information. 11

48 To Share Or Not To Share Memory Corrupt Memory 11 Death propagates in shared memory unless you do a ton of defensive programming. Due to the actor model with no shared memory it is custom in Erlang to do fail-fast programming. No shared memory allows you to fail fast when suitable. Erlang uses message passing between processes to exchange information. 11

49 To Share Or Not To Share Memory Corrupt Memory P1 11 Death propagates in shared memory unless you do a ton of defensive programming. Due to the actor model with no shared memory it is custom in Erlang to do fail-fast programming. No shared memory allows you to fail fast when suitable. Erlang uses message passing between processes to exchange information. 11

50 To Share Or Not To Share Memory Corrupt Memory Memory P1 11 Death propagates in shared memory unless you do a ton of defensive programming. Due to the actor model with no shared memory it is custom in Erlang to do fail-fast programming. No shared memory allows you to fail fast when suitable. Erlang uses message passing between processes to exchange information. 11

51 To Share Or Not To Share Memory Corrupt Memory Memory P1 P2 11 Death propagates in shared memory unless you do a ton of defensive programming. Due to the actor model with no shared memory it is custom in Erlang to do fail-fast programming. No shared memory allows you to fail fast when suitable. Erlang uses message passing between processes to exchange information. 11

52 To Share Or Not To Share Memory Corrupt Corrupt Memory P2 11 Death propagates in shared memory unless you do a ton of defensive programming. Due to the actor model with no shared memory it is custom in Erlang to do fail-fast programming. No shared memory allows you to fail fast when suitable. Erlang uses message passing between processes to exchange information. 11

53 To Share Or Not To Share Memory Corrupt Memory P2 11 Death propagates in shared memory unless you do a ton of defensive programming. Due to the actor model with no shared memory it is custom in Erlang to do fail-fast programming. No shared memory allows you to fail fast when suitable. Erlang uses message passing between processes to exchange information. 11

54 Dealing With Failures P1 P2 link = die together P1 P2 monitor = notification of death 12 The ability to link processes and monitor them is the key to adopting fail-fast fully. Link & monitor works across machines! 12

55 Supervision Trees supervisor supervisor worker worker worker worker worker The OTP library is built on this principle 13 Robust systems does not happen by accident - even in Erlang! You have to think about the consequences of a worker process that fails and let the supervisor take appropriate actions. Using the OTP library s components makes it straightforward to implement the supervision tree, which has the added benefit that all things are started in the right order. 13

56 Distribution Over Cores Core Core Scheduler Scheduler Processes Processes 14 There is 1 scheduler per core. The VM tries to load balance across the available cores. Scales extremely well with the addition of extra cores - WITHOUT changing the programs! 14

57 Distribution Over Machines 15 Each instance of the Erlang runtime is called a node. There can be several nodes on one machine if you fancy that. Nodes detect when other nodes are not around any more - the programmer can then decide what to do. If you have the PID (Process Identifier) of a process you do not care which node it is on. You 15

58 Distribution Over Machines Node 15 Each instance of the Erlang runtime is called a node. There can be several nodes on one machine if you fancy that. Nodes detect when other nodes are not around any more - the programmer can then decide what to do. If you have the PID (Process Identifier) of a process you do not care which node it is on. You 15

59 Distribution Over Machines Node Node 15 Each instance of the Erlang runtime is called a node. There can be several nodes on one machine if you fancy that. Nodes detect when other nodes are not around any more - the programmer can then decide what to do. If you have the PID (Process Identifier) of a process you do not care which node it is on. You 15

60 Distribution Over Machines Node Node Node 15 Each instance of the Erlang runtime is called a node. There can be several nodes on one machine if you fancy that. Nodes detect when other nodes are not around any more - the programmer can then decide what to do. If you have the PID (Process Identifier) of a process you do not care which node it is on. You 15

61 Distribution Over Machines Node Node Node Node 15 Each instance of the Erlang runtime is called a node. There can be several nodes on one machine if you fancy that. Nodes detect when other nodes are not around any more - the programmer can then decide what to do. If you have the PID (Process Identifier) of a process you do not care which node it is on. You 15

62 Staying Alive... Learn New Moves! 16 As style changed a new group entered the top of the hip scale. For most it meant a serious restart or a stop altogether. 16

63 Staying Alive... Learn New Moves! 16 As style changed a new group entered the top of the hip scale. For most it meant a serious restart or a stop altogether. 16

64 Staying Alive... Learn New Moves! 16 As style changed a new group entered the top of the hip scale. For most it meant a serious restart or a stop altogether. 16

65 Staying Alive... Learn New Moves! 16 As style changed a new group entered the top of the hip scale. For most it meant a serious restart or a stop altogether. 16

66 Staying Alive Erlang Style 17 With Erlang you can survive upgrades without loosing service. Along with the code change signal you specify how the internal state of the process should be updated before continuing. 17

67 Staying Alive Erlang Style Process running v1 17 With Erlang you can survive upgrades without loosing service. Along with the code change signal you specify how the internal state of the process should be updated before continuing. 17

68 Staying Alive Erlang Style Process running v1 Code loaded: v2 17 With Erlang you can survive upgrades without loosing service. Along with the code change signal you specify how the internal state of the process should be updated before continuing. 17

69 Staying Alive Erlang Style Process running Code change signal v1 Code loaded: v2 17 With Erlang you can survive upgrades without loosing service. Along with the code change signal you specify how the internal state of the process should be updated before continuing. 17

70 Staying Alive Erlang Style Process running Code change signal v1 v2 Code loaded: v2 17 With Erlang you can survive upgrades without loosing service. Along with the code change signal you specify how the internal state of the process should be updated before continuing. 17

71 Challenges 18 18

72 How Will I Know If It Really Scales? Performance??? Cores/Machines 19 A Scalability Benchmark Suite for Erlang/OTP. A number of synthetic benchmarks plus real-world (dialyzer and scalaris). Extendable to test your own application. 19

73 How Will I Know If It Really Scales? Performance Bencherl Cores/Machines 19 A Scalability Benchmark Suite for Erlang/OTP. A number of synthetic benchmarks plus real-world (dialyzer and scalaris). Extendable to test your own application. 19

74 Who Is Doing What? Core Core Scheduler Scheduler Processes Processes 20 20

75 Who Is Doing What? Core Core Scheduler When is this Scheduler process running? Processes Processes 20 20

76 Who Is Doing What? Core Core Scheduler When is this Scheduler process running? Processes Processes Who started this guy? 20 20

77 Percept2 To The Rescue Active functions 21 Percept: Erlang Concurrency Profiling Tool, utilizes trace informations and profiler events to form a picture of the processes's and ports runnability. Percept2 is an extension of Percept (part of the OTP release). Extensions: # of schedulers active, active functions, process migration, message passing stats, inter-node communication 21

78 Memory Alloc Previously R12B-1 Scheduler N Scheduler 2 Scheduler 1 Thread N Thread 1 sl_alloc sl_alloc sl_alloc eheap_alloc eheap_alloc eheap_alloc binary_alloc binary_alloc binary_alloc ets_alloc ets_alloc ets_alloc mseg_alloc 22 One central memory allocator for all schedulers on the same machine 22

79 Memory Alloc Now R15B Scheduler N Scheduler 2 Scheduler 1 Thread N Thread 1 sl_alloc sl_alloc sl_alloc sl_alloc eheap_alloc eheap_alloc eheap_alloc eheap_alloc binary_alloc binary_alloc binary_alloc binary_alloc ets_alloc ets_alloc ets_alloc ets_alloc mseg_alloc mseg_alloc mseg_alloc mseg_alloc 23 One central memory allocator for all schedulers on the same machien 23

80 Upgrading Blocks Current Code Next Code Load code All schedulers blocked 24 24

81 Upgrade Without Blocking Current Code Next Code Last Code Load code Each scheduler does it when needed 25 Coming in R16! 25

82 Fully Connected 26 Erlang connects all nodes fully. So you get a lot of connections. 26

83 Scaling Over Machines Node Node Node Node Node 27 A fully connected system might not be right for every problem. s_groups allows you to create clusters of nodes. Nodes inside a cluster are fully connected. Connections between clusters can be arbitrary. 27

84 Managing Erlang Systems 28 Basic Erlang has the ability to go in and monitor what is going on in any node you can attach yourself to. But no tool exists to manage a big number of nodes in a coherent fashion. This is no different from any other language/technology! 28

85 Managing Erlang Systems Provision machines 28 Basic Erlang has the ability to go in and monitor what is going on in any node you can attach yourself to. But no tool exists to manage a big number of nodes in a coherent fashion. This is no different from any other language/technology! 28

86 Managing Erlang Systems Provision machines Deploy Erlang application 28 Basic Erlang has the ability to go in and monitor what is going on in any node you can attach yourself to. But no tool exists to manage a big number of nodes in a coherent fashion. This is no different from any other language/technology! 28

87 Managing Erlang Systems Provision machines Attach to node Deploy Erlang application 28 Basic Erlang has the ability to go in and monitor what is going on in any node you can attach yourself to. But no tool exists to manage a big number of nodes in a coherent fashion. This is no different from any other language/technology! 28

88 Managing Erlang Systems Provision machines Attach to node Dig out metrics Deploy Erlang application 28 Basic Erlang has the ability to go in and monitor what is going on in any node you can attach yourself to. But no tool exists to manage a big number of nodes in a coherent fashion. This is no different from any other language/technology! 28

89 Managing Erlang Systems Provision machines Attach to node Dig out metrics Deploy Erlang application Memory usage CPU load Process hierarchy 28 Basic Erlang has the ability to go in and monitor what is going on in any node you can attach yourself to. But no tool exists to manage a big number of nodes in a coherent fashion. This is no different from any other language/technology! 28

90 Managing Big Systems CCL /sɪˈsɪlɪ/ 29 Erlang Solutions are building a tool as part of the RELEASE project to manage and operate big Erlang systems. CCL = Cloud Computing Lace or Cloud Cuckoo Land depending on your mood. 29

91 Erlang And Parallelism 30 30

92 Erlang And Parallelism Created for 30 30

93 Erlang And Parallelism Created for - explicit concurrency 30 30

94 Erlang And Parallelism Created for - - explicit concurrency fault tollerance 30 30

95 Erlang And Parallelism Created for explicit concurrency fault tollerance highly concurrent systems 30 30

96 Erlang And Parallelism Created for explicit concurrency fault tollerance highly concurrent systems No direct support for 30 30

97 Erlang And Parallelism Created for explicit concurrency fault tollerance highly concurrent systems No direct support for - matrix multiplication 30 30

98 Erlang And Parallelism Created for explicit concurrency fault tollerance highly concurrent systems No direct support for - - matrix multiplication ray tracing 30 30

99 Erlang And Parallelism Created for explicit concurrency fault tollerance highly concurrent systems No direct support for matrix multiplication ray tracing coarse grained parallel problems 30 30

100 Intensional Parallelism 31 We are taking the good things from what has been learnt in the Haskell & Data-flow language communities and building a DSL which helps us leverage these types of parallel optimisations" 31

101 Intensional Parallelism Lucid like: demand-driven data computation Find short comings in the Erlang VM Variables are infinite streams of values 31 We are taking the good things from what has been learnt in the Haskell & Data-flow language communities and building a DSL which helps us leverage these types of parallel optimisations" 31

102 Intensional Parallelism Lucid like: demand-driven data computation Find short comings in the Erlang VM Variables are infinite streams of values running_avg where sum = first(input) fby sum + next(input); n = 1 fby n + 1; running_avg = sum / n; end; 31 We are taking the good things from what has been learnt in the Haskell & Data-flow language communities and building a DSL which helps us leverage these types of parallel optimisations" 31

103 Intensional Parallelism Lucid like: demand-driven data computation Find short comings in the Erlang VM Variables are infinite streams of values running_avg Black magic where sum = first(input) fby sum + next(input); n = 1 fby n + 1; running_avg = sum / n; end; 31 We are taking the good things from what has been learnt in the Haskell & Data-flow language communities and building a DSL which helps us leverage these types of parallel optimisations" 31

104 Going Forward 32 32

105 Going Forward Consider Erlang when the problem fits 32 32

106 Going Forward Consider Erlang when the problem fits More focus on right tool for the job 32 32

107 Going Forward Consider Erlang when the problem fits More focus on right tool for the job 32 32