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

Marko Topolnik marko at hazelcast.com
Tue Mar 24 08:54:03 EDT 2015


On Tue, Mar 24, 2015 at 1:47 PM, Oleksandr Otenko <
oleksandr.otenko at oracle.com> wrote:

> I don't know why you call IRIW a test of TSO, if total order of reads is
> also important. Don't forget the load-load barrier, which is implicit on
> x86, and a small theorem about what happens when you don't order concurrent
> reads totally.
>

But note that you still don't need a total _global_ order of reads for
IRIW, it just has to be preserved locally for each processor. Dekker's
idiom exercises this more strongly because there has to be global ordering
between stores and loads by different processors.

---
Marko
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://cs.oswego.edu/pipermail/concurrency-interest/attachments/20150324/b490fa39/attachment-0001.html>


More information about the Concurrency-interest mailing list