Synchronization with Eventcounts and Sequencers David P. Reed Rajendra K. Kanodia

Introduction What is it used for? Synchronizing the use of shared resources How is it different from semaphores and monitors? Mutual exclusion

EventCounts What is an eventcount? Tracks number of events Non-decreasing integer variable

EventCount operations advance(E) read(E) await(E, v)

advance( E ) Signals occurrence of an event Update eventcount value

read( E ) Returns value of eventcount May or may not count events in progress

await( E, v ) Similar to read( E ) Waits for value v to be reached May not return immediately once the vth advance is executed

Producer/Consumer Example N-cell ring buffer buffer[0:N –1] Eventcounts IN and OUT produce() to generate items

Single producer code Procedure producer() begin integer i; for i:= 1 to infinity do begin await( OUT, i – N); buffer[i mod N] := produce( ); advance( IN ); end

Single consumer code Procedure consumer() begin integer i; Procedure producer() begin integer i; for i:= 1 to infinity do begin await( OUT, i – N); buffer[i mod N] := produce( ); advance( IN ); end Procedure consumer() begin integer i; for i := 1 to infinity begin await( IN, i ); consume( buffer[i mod N]); advance( OUT ); end

Possible Situations Fast Producer Fast Consumer Producer will wait until item it is trying to overwrite is consumed. Fast Consumer Consumer will wait until the producer has added the value.

EventCount observations How is this solution different than semaphores? Relative ordering rather than mutual exclusion. Producer & consumer can be concurrent. Never does a process have to wait due to synchronization.

Sequencers Used when synchronization requires arbitration Used to order the events Useful with Eventcounts but not on its own Non-decreasing integer value

Sequencer operations ticket(S) Value returned is the process’s ordering. Two calls to ticket( S ) will always return different values.

Producer/Consumer Example Same as with Eventcounts but multiple producers now Sequencer T Use ticket(T) to synchronize with other producers

buffer[t+1mod N] := produce( ); advance( IN ); end Procedure producer() begin integer i; for i:= 1 to infinity do begin await( OUT, i – N); buffer[i mod N] := produce( ); advance( IN ); end Procedure producer() begin integer t; do forever begin t := ticket(T); await( IN, t); await( OUT, t – N + 1 ) buffer[t+1mod N] := produce( ); advance( IN ); end

Relation to semaphores Lower level than Semaphores Semaphores can be built from Eventcounts and Sequencers

Semaphore Example Semaphore S EventCount is S.E Sequencer is S.T Initial value of S is S.I

Semaphore Wait Procedure P(S) begin integer t; t := ticket( S.T ); await( S.E, t – S.I ); end

Semaphore Signal Procedure V( S ) begin advance( S.E ) end

Deadlock Free Simultaneous P 2 Semaphores R and S Global semaphore G to synchronize part of operation

Procedure Pboth( R, S ) begin integer g, r, s; g := ticket( G.T ); await( G.E, g ); r := ticket( R.T ); s := ticket( S.T ); advance( G.E ); await( R.E, r – R.I ); await( S.E, s – S.I ); end

Observations G is used for obtaining tickets await operation could be deferred Useful in solving the Dining Philosophers Problem

Observations Process destroyed while holding 1 ticket Keep the window during which a process has an unredeemed a ticket short Don’t allow destroying the process during the window

Flow of information operations are: Observer or Signaler unlike the semaphore wait Easily adapted to permissions Observer permission(advance). signaler permission(read,await). Useful in secure systems

Secure Readers - Writers Shared database Readers have observer permission Writers have observer and signaler permission Writer priority S and C are eventcounts T is a sequencer

Reader code Procedure reader() begin integer w; abort: w := read(S); await(C, w); “read DB” if read(S) ≠ w then goto abort; end

Writer code Procedure writer() begin integer t; advance( S ); t := ticket( T ); await( C, t ); advance( C ); end

Conclusion new mechanism for synchronization. not based on mutual exclusion. Provides an interface between processes. Information flow paths are clear. Effective in secure systems. Unnecessary serialization avoided.