[concurrency-interest] Enforcing total sync order on modern hardware

Alexander Terekhov TEREKHOV at de.ibm.com
Wed Mar 25 06:02:53 EDT 2015


thats about atomic 'stores' and 'loads'...  not 'write' atomicity... (note
different words)

Marko Topolnik <marko at hazelcast.com> on 25.03.2015 10:49:39

To:	Alexander Terekhov/Germany/IBM at IBMDE
cc:	concurrency-interest <Concurrency-interest at cs.oswego.edu>
Subject:	Re: [concurrency-interest] Enforcing total sync order on modern
       hardware


You have confused my complaint as being about the lack of any definition,
whereas I actually complained that there are several competing definitions
used by various sources. The most natural definition, at least for me, is
this:

    When an atomic store is performed on a shared variable, no other thread
    can observe the modification half-complete.

Source: http://preshing.com/20130618/atomic-vs-non-atomic-operations

---
Marko

On Wed, Mar 25, 2015 at 10:37 AM, Alexander Terekhov <TEREKHOV at de.ibm.com>
wrote:
      try google... quick search yielded

      https://books.google.de/books?isbn=1608459578

      Michael L. Scott - 2013 - ‎Computers
      Here the problem is not bypassing, but a lack of write atomicity—one
      thread
      sees the value written by a store and another thread subsequently
      sees the
      value ...


      Marko Topolnik <marko at hazelcast.com>@cs.oswego.edu on 25.03.2015 08
      :07:05

      Sent by:        concurrency-interest-bounces at cs.oswego.edu


      To:     Alexander Terekhov/Germany/IBM at IBMDE
      cc:     concurrency-interest <Concurrency-interest at cs.oswego.edu>
      Subject:        Re: [concurrency-interest] Enforcing total sync order
      on modern
             hardware


      On Wed, Mar 25, 2015 at 4:01 AM, Alexander Terekhov <
      TEREKHOV at de.ibm.com>
      wrote:
            T1: X = 1;
            T2: if (X) Y = 1;
            T3: A = Y; B = X;

            is about atomicity of absolutely independent write T1: X = 1;
      just
            like in
            IRIW.

      IRIW requires all writes to be independent, not just one, and I have
      provided a clear argument why that is important (testing for TSO of
      independent writes). So I can't really make sense of your current
      argument.

      BTW "write atomicity" just means that all the constituent bits are
      observed
      to be written at once. Obviously, you don't have that meaning in
      mind, but
      it is not crystal-clear what exactly you _do_ mean by it.

      ---
      Marko_______________________________________________
      Concurrency-interest mailing list
      Concurrency-interest at cs.oswego.edu
      http://cs.oswego.edu/mailman/listinfo/concurrency-interest



More information about the Concurrency-interest mailing list