Abort Free SemanticTM by Dependency Aware Scheduling of Transactional Instructions Shlomi Dolev Ben-Gurion University of the Negev Israel WTM 2013 Panagiota Fatourou University of Crete & FORTH-ICS Greece Eleftherios Kosmas University of Crete & FORTH-ICS Greece

WTM 2013 S. Dolev & P. Fatourou & E. Kosmas  When two transactions conflict, most TM systems abort one of them to ensure consistency  Ideally:  All transactions should commit  Parallelism should not be sacrificed  TMs that never abort transactions are highly desirable  support of irrevocable transactions  avoid the cost of re-executing aborted transactions Two transactions conflict, if they both access the same t-variable and at least one of these accesses is a write Introduction

WTM 2013 S. Dolev & P. Fatourou & E. Kosmas SemanticTM  an opaque TM algorithm  no transaction ever aborts guarantees wait-freedom/local progress for the execution of transactions  fine-grain parallelism at the transactional instruction level An execution of a TM algorithm is opaque if it satisfies strict serializability and active transactions read “consistent” values for t-variables Our Result

WTM 2013 S. Dolev & P. Fatourou & E. Kosmas Transaction Level Parallelism p1p1 p2p2 pnpn 2. T 1.read(z) x yzw 1. T 1.read(x) 5. T 1.write(w) 4. T 1.read(w) 3. T 1.write(y) T 2.read T 2.write T 2.read T m.read T2T2 T1T1 TmTm

WTM 2013 S. Dolev & P. Fatourou & E. Kosmas Transactional Instruction Level Parallelism p1p1 p2p2 pnpn 2. T1.read x yzw 1. T1.read 5. T1.write 4. T1.read 3. T1.write  dependencies may exist between instructions of the same transaction T2. read T2. write T2. read Tm. read T2T2 TmTm

WTM 2013 S. Dolev & P. Fatourou & E. Kosmas SemanticTM p1p1 p2p2 pnpn T1.read x yz w T1.write T1.read T1.write T2.readT2.write T2.read Tm.read t-var lists  a scheduler places transactional intructions into t-var lists  together with their dependencies  transactions are processed by the scheduler the one after the other  each process randomly chooses a t-var list and executes its ready instructions  no conflicts occur between transactions p3p3  For relatively simple transactions the work of the scheduler can be done at compile time.

WTM 2013 S. Dolev & P. Fatourou & E. Kosmas  If the execution of a transactional instruction e 2 depends on the execution of a transactional instruction e 1  there is a dependency between e 1 and e 2 input dependency for e 2 output dependency for e 1  A dependency between  a read and a write instruction → data dependency  a cond and a read (or a write) instruction → control dependency Dependencies x:=3 x++ y:=x if ( ) then x:=1 x+y = 5

WTM 2013 S. Dolev & P. Fatourou & E. Kosmas x:=3 x++ y:=x write(x,3) tmp 1 := read(x) write(x,tmp 1 +1) tmp 2 := read(x) write(y,tmp 2 ) write 0 0 xy read write data dep p1p1 p2p2 read Reads and Writes

WTM 2013 S. Dolev & P. Fatourou & E. Kosmas if(x+y = 5) then x := 1 tmp 1 := read(x) tmp 2 := read(y) tmp 3 := tmp 1 +tmp 2 cond(tmp 3 =5) write(x,1) xy read cond write data dep control dep p1p1 p2p2  If cond evaluates to TRUE then the write is executed  Otherwise it is invalidated read Conditionals – if statement

WTM 2013 S. Dolev & P. Fatourou & E. Kosmas if(x = 1) then x := 2 else x := 4 y := x tmp 1 := read(x) cond(tmp 1 = 1) write(x,2) cond(tmp 1 <> 1) write (x,4) tmp 2 := read(x) write(y,tmp 2 ) xy readwrite cond write read data dep control dep p1p1 p2p2 Conditionals – if..else statements 04 04

WTM 2013 S. Dolev & P. Fatourou & E. Kosmas Loops while(x < 10) do y := x x := 2* x tmp 1 := read(x) cond(tmp 1 < 10, loop) tmp 2 := read(x) write(y,tmp 2 ) tmp 3 := 2 * tmp 2 write(x,tmp 3 ) tmp 1 := read(x) xy read cond read write read control dep data dep across iteration data dep

WTM 2013 S. Dolev & P. Fatourou & E. Kosmas Loops 0 1 xy read cond read output data dep control dep write read across iteration data dep Metadata while (x < 10) y := x x := x* Metadata p1p1 p2p

WTM 2013 S. Dolev & P. Fatourou & E. Kosmas Synchronization for each transactional instruction  status ( ACTIVE, DONE ) together with the value of each input data dependency  iteration number  CAS object for each control dependency  CAS object x y status val iter for the value of each t- variable  version  CAS object  update using oldvalue iteration number CAS object valread ver 1 oldval iter write 1

WTM 2013 S. Dolev & P. Fatourou & E. Kosmas Support of transactions accessing t-variables that are known at run time  accessing an element of an array Extensions A[1] A[2]A[3]A[m] read write read read: A[*] write: A[*] read: A[3] write: A[2] read: A[*] write: A[*] A[*] Dynamic transactions  similar strategy as above, if we consider the memory heap as an array Nested conditionals & loops (cond c2 in the block of outer cond c1)  add a control dependency from c1 to c2 but not to the instructions of the block of c2.

WTM 2013 S. Dolev & P. Fatourou & E. Kosmas Summary We presented SemanticTM:  executes transactions without ever causing any aborts  parallelism is fine-grained achieved at the level of transactional instructions Blocking version of SemanticTM Future Work Implement an optimized version of SemanticTM Experimental Study Conclusion & Future Work

WTM 2013 S. Dolev & P. Fatourou & E. Kosmas QUESTIONS? Thank you!