Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > comp.programming.threads > #2074

I eaplain my new LW_RWLockX algorithm

Show all headers | View raw

Hello,

I will explain my new LW_RWLockX algorithm...

you have to undertand first my LW_RWLock algorithm that is
not starvation-free..

Here it's:

http://pages.videotron.com/aminer/rwlock1.html


Here is what i have changed, first i have to not
use spin-wait cause if you loop back it will not be starvation-free and
so you can still have starvation , so i have got rid of looping back and 
i have done this:


In the TRWLOCK.RLock() method you will read this:


procedure TRWLOCK.RLock(var myid:integer);


begin

myid:=GetCurrentProcessorNumber;

if (FCount3^.fcount3 = 1)
  then
    begin
     LockedExchangeAdd(nbr2,1);
     while nbr3=0 do sleep(0);
     LockedExchangeAdd(nbr2,-1);
    end;

LockedExchangeAdd(FCount1^[myid].fcount1,1);

if (FCount3^.fcount3 = 0)
then
else
  begin
   LockedExchangeAdd(nbr,1);
   LockedExchangeAdd(FCount1^[myid].fcount1,-1);
   while nbr1=0 do sleep(0);
   LockedExchangeAdd(FCount1^[myid].fcount1,1);;
   LockedExchangeAdd(nbr,-1);

  end;

end;

and in the TRWLOCK.WLock() method you will read this:


procedure TRWLOCK.WLock;

var i:integer;

begin

lock.enter;

FCount3^.fcount3:=1;

repeat;
  nbr1:=1;
  asm pause end;
  nbr1:=0;
until nbr=0;

FCount3^.fcount3:=0;

repeat;
  nbr3:=1;
  asm pause end;
  nbr3:=0;
until nbr2=0;

FCount3^.fcount3:=1;

for i:=0 to GetSystemThreadCount-1 do
  begin
    while (FCount1^[i].fcount1<>0)
     do
      begin
       //if mysleep=0 then sleep(0)
       //else sleep(1);
      end;
  end;

end;

--

Notice in TRWLOCK.WLock() that i have first to
set FCount3^.fcount3 to 1 to block the reader threads
and i have done also this in the reader side:

repeat;
  nbr1:=1;
  asm pause end;
  nbr1:=0;
until nbr=0;


this repeat until loop will just unblock the reader threads that entered 
  between the following two LockedExchangeAdd() in the reader side:

LockedExchangeAdd(nbr,1);
LockedExchangeAdd(nbr,1);


after that i am setting FCount3^.fcount3 to 0 to unblock the new reader
that are coming and after that i am using the following repeat loop

repeat;
  nbr3:=1;
  asm pause end;
  nbr3:=0;
until nbr2=0;

this repeat until loop will just unblock the reader threads that entered 
between the following two LockedExchangeAdd() in the reader side:

  LockedExchangeAdd(nbr2,1);
  LockedExchangeAdd(nbr2,-1);


this way i the reader threads will not starve forever the writers,
and this way the algorithm will be starvation-free and it is
fast and scalable, i have just tested it and it's scalable.

You can download all the variants of my scalable RWLock 2.15 from:

http://pages.videotron.com/aminer/


And please take a look at my new algorithm inside LW_RWLockX.pas
algorithm inside the zip file.




Thank you,
Amine Moulay Ramdane.


Thank you,
Amine Moulay Ramdane

Back to comp.programming.threads | Previous | Next | Find similar | Unroll thread

Thread

I eaplain my new LW_RWLockX algorithm aminer <aminer@toto.net> - 2014-01-29 18:50 -0800

csiph-web