Slide 1
Slide 1 text
Bloom: A Language For Disorderly
Distributed Programming
Christopher Meiklejohn
Wicked Good Ruby ’13
Sunday, October 13, 13
Slide 2
Slide 2 text
cmeiklejohn / @cmeik
My Name is My Name
Sunday, October 13, 13
Slide 3
Slide 3 text
What do you do?
Sunday, October 13, 13
Slide 4
Slide 4 text
Sunday, October 13, 13
Slide 5
Slide 5 text
Sunday, October 13, 13
Slide 6
Slide 6 text
Sunday, October 13, 13
Slide 7
Slide 7 text
What did you do?
Sunday, October 13, 13
Slide 8
Slide 8 text
Sunday, October 13, 13
Slide 9
Slide 9 text
Sunday, October 13, 13
Slide 10
Slide 10 text
UC Berkeley
Sunday, October 13, 13
Slide 11
Slide 11 text
BOOM
Sunday, October 13, 13
Slide 12
Slide 12 text
The Outline
Sunday, October 13, 13
Slide 13
Slide 13 text
Theory, language.
The Outline
Sunday, October 13, 13
Slide 14
Slide 14 text
von Neumann
Sunday, October 13, 13
Slide 15
Slide 15 text
Instructions, mutable state, executed in order.
von Neumann
Sunday, October 13, 13
Slide 16
Slide 16 text
Imposes a total order.
von Neumann
Sunday, October 13, 13
Slide 17
Slide 17 text
Disorderly actors
Sunday, October 13, 13
Slide 18
Slide 18 text
Allow disorder and races; illusion of total order.
Disorderly actors
Sunday, October 13, 13
Slide 19
Slide 19 text
The idea behind transactions; serializability.
Disorderly actors
Sunday, October 13, 13
Slide 20
Slide 20 text
Tricky to scale.
Disorderly actors
Sunday, October 13, 13
Slide 21
Slide 21 text
Scaling
Sunday, October 13, 13
Slide 22
Slide 22 text
Multiple machines; sharing computation.
Scaling
Sunday, October 13, 13
Slide 23
Slide 23 text
CAP Theorem
Sunday, October 13, 13
Slide 24
Slide 24 text
Consistency, availability, partition tolerance.
CAP Theorem
Sunday, October 13, 13
Slide 25
Slide 25 text
Negative result: linearizability impossible.
CAP Theorem
Sunday, October 13, 13
Slide 26
Slide 26 text
Whoa, Theory!
Sunday, October 13, 13
Slide 27
Slide 27 text
Distributed computing
Sunday, October 13, 13
Slide 28
Slide 28 text
The “new normal”.
Distributed computing
Sunday, October 13, 13
Slide 29
Slide 29 text
Order is di!cult to impose.
Distributed computing
Sunday, October 13, 13
Slide 30
Slide 30 text
Synchronization; coordination.
Distributed computing
Sunday, October 13, 13
Slide 31
Slide 31 text
FLP result; delays, partitions, partial writes.
Distributed computing
Sunday, October 13, 13
Slide 32
Slide 32 text
Inverse methods; commutative operations.
Distributed computing
Sunday, October 13, 13
Slide 33
Slide 33 text
ACID 2.0
Sunday, October 13, 13
Slide 34
Slide 34 text
Associativity, commutativity, idempotence.
ACID 2.0
Sunday, October 13, 13
Slide 35
Slide 35 text
(X · Y) · Z = X · (Y · Z)
ACID 2.0: Associativity
Sunday, October 13, 13
Slide 36
Slide 36 text
X · Y = Y · X
ACID 2.0: Commutativity
Sunday, October 13, 13
Slide 37
Slide 37 text
X · X = X
ACID 2.0: Idempotence
Sunday, October 13, 13
Slide 38
Slide 38 text
Associativity, commutativity, idempotence.
ACID 2.0
Sunday, October 13, 13
Slide 39
Slide 39 text
Familiar?
Sunday, October 13, 13
Slide 40
Slide 40 text
Bounded Join Semilattices
Sunday, October 13, 13
Slide 41
Slide 41 text
Partially ordered set; least upper bound; ACI.
Bounded Join Semilattices
Sunday, October 13, 13
Slide 42
Slide 42 text
Objects grow over time; merge computes LUB.
Bounded Join Semilattices
Sunday, October 13, 13
Slide 43
Slide 43 text
Monotonic and con"uent; convergent.
Bounded Join Semilattices
Sunday, October 13, 13
Slide 44
Slide 44 text
Map into another via monotone functions.
Bounded Join Semilattices
Sunday, October 13, 13
Slide 45
Slide 45 text
Examples.
Bounded Join Semilattices
Sunday, October 13, 13
Slide 46
Slide 46 text
b a c
a, b a, c
a, b, c
Set; merge function: union.
b, c
Sunday, October 13, 13
Slide 47
Slide 47 text
3 5 7
5 7
7
Increaing natual; merge function: max.
Sunday, October 13, 13
Slide 48
Slide 48 text
F F T
F T
T
Booleans; merge function: or.
Sunday, October 13, 13
Slide 49
Slide 49 text
x <= y montone f(x) <= f(y)
Sunday, October 13, 13
Slide 50
Slide 50 text
Sunday, October 13, 13
Slide 51
Slide 51 text
CvRDT examples.
Sunday, October 13, 13
Slide 52
Slide 52 text
[ [{1, a}], [] ] [ [{1, a}], [] ]
Sunday, October 13, 13
Slide 53
Slide 53 text
[ [{1, a}], [] ] [ [{1, a}], [] ]
[ [{1, a}], [{1, a}] ]
[ [{1, a}], [{1, a}] ]
Sunday, October 13, 13
Slide 54
Slide 54 text
[ [{1, a}], [] ] [ [{1, a}], [] ]
[ [{1, a}], [{1, a}] ]
[ [{1, a}], [{1, a}] ]
[ [{1, a}, {2, a}], [{1, a}] ]
Sunday, October 13, 13
Slide 55
Slide 55 text
[ [{1, a}], [] ] [ [{1, a}], [] ]
[ [{1, a}], [{1, a}] ]
[ [{1, a}], [{1, a}] ]
[ [{1, a}, {2, a}], [{1, a}] ]
[ [{1, a}, {2, a}], [{1, a}] ]
Sunday, October 13, 13
Slide 56
Slide 56 text
[ [{1, a}], [] ] [ [{1, a}], [] ]
Sunday, October 13, 13
Slide 57
Slide 57 text
[ [{1, a}], [] ] [ [{1, a}], [] ]
[ [{1, a}, {2, b}], [] ]
Sunday, October 13, 13
Slide 58
Slide 58 text
[ [{1, a}], [] ] [ [{1, a}], [] ]
[ [{1, a}, {2, b}], [] ]
[ [{1, a}], [{1, a}] ]
Sunday, October 13, 13
Slide 59
Slide 59 text
[ [{1, a}], [] ] [ [{1, a}], [] ]
[ [{1, a}, {2, b}], [] ]
[ [{1, a}], [{1, a}] ]
[ [{1, a}, {2, b}], [{1, a}] ]
Sunday, October 13, 13
Slide 60
Slide 60 text
Just data; what about computation?
Bounded Join Semilattices
Sunday, October 13, 13
Slide 61
Slide 61 text
Composability desired.
Bounded Join Semilattices
Sunday, October 13, 13
Slide 62
Slide 62 text
CALM Theorem
Sunday, October 13, 13
Slide 63
Slide 63 text
Consistency as Logical Monotonicity
CALM Theorem
Sunday, October 13, 13
Slide 64
Slide 64 text
Consistent given async; all monotone functions.
CALM Theorem
Sunday, October 13, 13
Slide 65
Slide 65 text
Identify non-monotone functions.
CALM Theorem
Sunday, October 13, 13
Slide 66
Slide 66 text
Disorderly Programming
Sunday, October 13, 13
Slide 67
Slide 67 text
Order-independent, explicit order-dependent.
Disorderly Programming
Sunday, October 13, 13
Slide 68
Slide 68 text
Prototype language: Bloom
Disorderly Programming
Sunday, October 13, 13
Slide 69
Slide 69 text
Bloom
Sunday, October 13, 13
Slide 70
Slide 70 text
Based in a logic called Dedalus.
Bloom
Sunday, October 13, 13
Slide 71
Slide 71 text
Disorderly language.
Bloom
Sunday, October 13, 13
Slide 72
Slide 72 text
Lattices and mappings between lattices.
Bloom
Sunday, October 13, 13
Slide 73
Slide 73 text
Highlights non-monotonicity.
Bloom
Sunday, October 13, 13
Slide 74
Slide 74 text
Operational model.
Bloom
Sunday, October 13, 13
Slide 75
Slide 75 text
Local clock; local execution; local state.
Bloom
Sunday, October 13, 13
Slide 76
Slide 76 text
Timestepped execution.
Bloom
Sunday, October 13, 13
Slide 77
Slide 77 text
Local updates; system events; network.
Sunday, October 13, 13
Slide 78
Slide 78 text
Bloom rules; merge functions.
Sunday, October 13, 13
Slide 79
Slide 79 text
Network; next.
Sunday, October 13, 13
Slide 80
Slide 80 text
Chat server.
Bloom
Sunday, October 13, 13
Slide 81
Slide 81 text
module ChatProtocol
state do
channel :connect, [:@addr, :client] => [:nick]
channel :mcast
end
end
examples/chat/chat_protocol.rb
Sunday, October 13, 13
Slide 82
Slide 82 text
class ChatServer
include Bud
include ChatProtocol
state { table :nodelist }
bloom do
nodelist <= connect.map { |c| [c.client, c.nick] }
mcast <~ (mcast * nodelist).pairs { |m,n| [n.key, m.val] }
end
end
examples/chat/chat_server.rb
Sunday, October 13, 13
Slide 83
Slide 83 text
Lattices; monotonic functions.
Bloom
Sunday, October 13, 13
Slide 84
Slide 84 text
class ChatServer
include Bud
include ChatProtocol
state do
table :nodelist
channel :disconnect
end
bloom do
nodelist <= connect.map { |c| [c.client, c.nick] }
mcast <~ (mcast * nodelist).pairs { |m,n| [n.key, m.val] }
nodelist <- disconnect.map { |c| [c.client, c.nick] }
end
end
examples/chat/chat_server.rb
Sunday, October 13, 13
Slide 85
Slide 85 text
Non-monotonic; not eventually consistent.
Bloom
Sunday, October 13, 13
Slide 86
Slide 86 text
bloom do
mcast <~ stdio do |s|
[@server, [ip_port, @nick,
Time.new.strftime("%I:%M.%S"), s.line]]
end
stdio <~ mcast { |m| [pretty_print(m.val)] }
end
examples/chat/chat.rb
Sunday, October 13, 13
Slide 87
Slide 87 text
KV Store.
Bloom
Sunday, October 13, 13
Slide 88
Slide 88 text
module BasicKVS
include KVSProtocol
state do
table :kvstate, [:key] => [:value]
end
bloom :mutate do
kvstate <+ kvput {|s| [s.key, s.value]}
kvstate <- (kvstate * kvput).lefts(:key => :key)
end
bloom :get do
temp :getj <= (kvget * kvstate).pairs(:key => :key)
kvget_response <= getj do |g, t|
[g.reqid, t.key, t.value]
end
end
bloom :delete do
kvstate <- (kvstate * kvdel).lefts(:key => :key)
end
end
kvs/kvs.rb
Sunday, October 13, 13
Slide 89
Slide 89 text
module ReplicatedKVS
include KVSProtocol
include MulticastProtocol
include MembershipProtocol
import BasicKVS => :kvs
bloom :local_indir do
kvs.kvdel <= kvdel
kvs.kvget <= kvget
kvget_response <= kvs.kvget_response
end
bloom :puts do
# if I am the master, multicast store requests
mcast_send <= kvput do |k|
unless member.include? [k.client]
[k.reqid, ["put", [@addy, k.key, k.reqid, k.value]]]
end
end
kvs.kvput <= mcast_done do |m|
if m.payload[0] == "put"
m.payload[1]
end
end
# if I am a replica, store the payload of the multicast
kvs.kvput <= pipe_out do |d|
if d.payload.fetch(1) != @addy and d.payload[0] == "put"
d.payload[1]
end
end
end
bloom :dels do
mcast_send <= kvdel do |k|
unless member.include? [k.client]
[k.reqid, ["del", [@addy, k.key, k.reqid]]]
end
end
kvs.kvdel <= mcast_done do |m|
if m.payload[0] == "del"
m.payload[1]
end
end
kvs.kvdel <= pipe_out do |d|
if d.payload.fetch(1) != @addy and d.payload[0] == "del"
d.payload[1]
end
end
end
end
kvs/kvs.rb
Sunday, October 13, 13
Slide 90
Slide 90 text
CALM Analysis
Sunday, October 13, 13
Slide 91
Slide 91 text
Label asynchronous, non-monotonic.
CALM Analysis
Sunday, October 13, 13
Slide 92
Slide 92 text
Identify where coordination is needed.
CALM Analysis
Sunday, October 13, 13
Slide 93
Slide 93 text
The Conclusion
Sunday, October 13, 13
Slide 94
Slide 94 text
Prototype language.
The Conclusion
Sunday, October 13, 13
Slide 95
Slide 95 text
https://github.com/bloom-lang/bud
https://github.com/bloom-lang/bud-sandbox
Thanks! Questions?
Sunday, October 13, 13