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


Groups > comp.programming > #3913 > unrolled thread

Transactional memory and lockfree...

Started byaminer <aminer@toto.net>
First post2013-10-10 18:26 -0700
Last post2013-10-10 18:26 -0700
Articles 1 — 1 participant

Back to article view | Back to comp.programming


Contents

  Transactional memory and lockfree... aminer <aminer@toto.net> - 2013-10-10 18:26 -0700

#3913 — Transactional memory and lockfree...

Fromaminer <aminer@toto.net>
Date2013-10-10 18:26 -0700
SubjectTransactional memory and lockfree...
Message-ID<l379jc$e02$1@news.albasani.net>
Hello,

I have thought more about lockfree algorithms, they look
like transactional memory but they only work for small applications
such us FIFO queues and stacks and hash tables and Skiplists.. but 
transactional mmemory is suited also for bigger parallel applications, but
i was asking myself a question about the resiliency of the lockfree
algorithms, such us if a thread have crached or stopped inside the CAS 
,that is
also a kind of small critical section, it will not stop the progress
of the other threads, this requirement can, in my opinion, be satisfied 
in lock based algorithms by just using exceptions, cause a thread can 
have only
a some limited problems such as when a memory is full or the disk is 
full .. so we can
use also exceptions to avoid some of those problems and requeue
the work in a FIFO queue for example and leave the critical section,
so Lock based algorithms are also very interresting cause they can
satisfy the requirement of the FIFO fairness and  Locks such
us scalable Anderson Lock or the MCS lock can also minimize the
cache-coherence traffic, this is not the case with the Lockfree traffic, 
so as i said , the Lock based algorithms coupled with
exceptions can satisfy many requirements and can be excellent.



Thank you,
Amine Moulay Ramdane.



[toc] | [standalone]


Back to top | Article view | comp.programming


csiph-web