shared memory & synchronization

8/13/2019 Shared Memory & Synchronization

1/19

Multicore Programming

Shared Memory & Synchronization

Tutorial 5CS 0368-3469Spring 2010


2/19

2

Summary

Registers

AtomicMRSWRegister Q.

Consensus

Queue Q.

Team consensus Q.


3/19

3 2007 Herlihy & Shavit 3

Safe Register

write(1001)

Some valid value ifreads and writes do

overlap

read(????)

0000 1001 1111

$*&v


4/19


Regular Register

write(0)

read(1)

write(1)

read(0)

Single Writer Readers return:

Old value if no overlap (safe) Old or one of new values if overlap


5/19


Atomic Register

write(1001)

read(1001)

Linearizable to sequential saferegister

write(1010)

read(1010)

read(1010)


6/19

6Art of Multiprocessor Programming 6

Safe Boolean MRSW fromSafe Boolean SRSW

public classSafeBoolMRSWRegisterimplementsRegister {privateSafeBoolSRSWRegister[] r =newSafeBoolSRSWRegister[N];public voidwrite booleanx) {for intj = 0; j < N; j++)r[j].write x);}public booleanread ) {inti = ThreadID.get );returnr[i].read );}}


7/19

7

Safe Boolean MRSW Qeustion

True/False:

If we replace the safe Boolean SRSWregisterswith regular Boolean SRSWregisters, then the constructionyields an regular BooleanMRSW

register.


8/19


Safe Boolean MRSW Qeustion

cont.

time

R[0]=1

1/0read(R[1])

1read(R[0])

R[1]=1 R[2]=1

0

read(R[2])


9/19

9

AtomicMRSWRegister

public classAtomicMRSWRegister implements Register {ThreadLocal lastStamp; // last timestamp writtenprivateStampedValue[][] a_table; // each entry is SRSW atomic

publicAtomicMRSWRegister(T init, intreaders) {this.lastStamp = newThreadLocal() {

protected LonginitialValue () { return 0; };};a_table = newStampedValue[readers][readers];StampedValue value = newStampedValue(init);for(inti = 0; i < readers; i++){

for(intj = 0; j < readers; j++){a_table[ i ][ j ] = value;}

}}


10/19

10

AtomicMRSWRegister

publicT read() {intme = ThreadID.get();StampedValue value = a_table[me][me]; // former read

for(inti = 0; i < a_table.length ; i++) // finds max in rowvalue = StampedValue.max(value, a_table[me][i]);

for(inti = 0; i < a_table.length ; i++) // writes max in columna_table[i][me] = value;

returnvalue ;}


11/19

11

AtomicMRSWRegister

publicvoid write (T v){longstamp = lastStamp.get() + 1;

lastStamp.set( stamp);// remember for next time

StampedValue value = newStampedValue(stamp, v);

// writes a column

for(inti = 0; i < a_table.length ; i++)a_table [i][0] = value;}


12/19


AtomicMRSWRegister

1:45 1234 1:45 12341:45 1234

1:45 1234 1:45 1234

1:45 1234

One per thread

1:45 1234

1:45 1234

1:45 1234

1 2 3

1

23


13/19

13

AtomicMRSWRegister

QeustionTrue/False:

If we replace the atomicSRSWregisters with regularSRSWregisters, then the construction stillyields an atomicMRSWregister.


14/19


AtomicMRSWRegister

Qeustion cont.

time

write(2:00 5678)

read(1:45 1234)

1:451234

read(2:00 5678)

In which caseits OK to read

1234

Yes it does, because regular registers act differently only whentwo reads are done concurrently with a write.


15/19


AtomicMRSWRegister

Qeustion cont.

time

write(2:00 5678)

read(2:00 5678)

1:451234

read(2:00 5678)

In which case Bluewill complete writing2:00 5678 to its

column

read(1:45 1234)


16/19

16

AcmRegister Qeustion

public classAcmRegister {//Atomic MRSW registersprivateBoolRegister[] b = newBoolRegister[3*N];

public voidwrite(intx){

boolean[] v = intToBooleanArray(x);

for( inti=0 ; i


17/19

17

AcmRegister Qeustion cont.public classAcmRegister {

public intread(){

for( inti=0 ; i


18/19

18

AcmRegister Qeustion cont.

If first == middle

return firstElse If last == middlereturn last

Else //writer in the middlereturn first
http://creativecommons.org/licenses/by-sa/2.5/http://creativecommons.org/licenses/by-sa/2.5/


19/19

19Art of MultiprocessorProgramming 19

This work is licensed under a Creative Commons Attribution-ShareAlike 2.5 License.

You are free:

to Shareto copy, distribute and transmit the work

to Remixto adapt the work

Under the following conditions:

Attribution. You must attribute the work to The Art ofMultiprocessor Programming (but not in any way that suggests thatthe authors endorse you or your use of the work).

Share Alike. If you alter, transform, or build upon this work, youmay distribute the resulting work only under the same, similar or acompatible license.

For any reuse or distribution, you must make clear to others the licenseterms of this work. The best way to do this is with a link to

http://creativecommons.org/licenses/by-sa/3.0/.

Any of the above conditions can be waived if you get permission fromthe copyright holder.

Nothing in this license impairs or restricts the author's moral rights.
http://creativecommons.org/licenses/by-sa/2.5/http://creativecommons.org/licenses/by-sa/2.5/http://creativecommons.org/licenses/by-sa/2.5/http://creativecommons.org/licenses/by-sa/2.5/http://creativecommons.org/licenses/by-sa/2.5/http://creativecommons.org/licenses/by-sa/2.5/http://creativecommons.org/licenses/by-sa/2.5/http://creativecommons.org/licenses/by-sa/2.5/http://creativecommons.org/licenses/by-sa/2.5/

shared memory & synchronization

Documents