Comments on: Transactions Everywhere http://www.lshift.net/blog/2005/09/26/transactions-everywhere What happens at LShift Sat, 11 Oct 2008 11:17:33 +0000 http://wordpress.org/?v=2.0.12-alpha by: matthias http://www.lshift.net/blog/2005/09/26/transactions-everywhere#comment-13 Tue, 11 Oct 2005 09:37:19 +0000 http://www.lshift.net/blog/2005/09/26/transactions-everywhere#comment-13 <p>While STM operations are cheap compared to other methods for controlling access to shared data in a multi-threaded system, they are still substantially more expensive than straightforward memory access. However, only shared, mutable storage needs STM semantics and incur the cost of STM operations.</p> <p>It thus follows that STMs are only suitable as a memory model for languages where mutation and sharing is rare and static program analysis can easily determine where it takes place.</p> <p>This makes languages like Haskell - with its pure functional characteristics guaranteed by the type system - ideal candidates for STM implementations.</p> While STM operations are cheap compared to other methods for controlling access to shared data in a multi-threaded system, they are still substantially more expensive than straightforward memory access. However, only shared, mutable storage needs STM semantics and incur the cost of STM operations.

It thus follows that STMs are only suitable as a memory model for languages where mutation and sharing is rare and static program analysis can easily determine where it takes place.

This makes languages like Haskell - with its pure functional characteristics guaranteed by the type system - ideal candidates for STM implementations.

]]> by: matthias http://www.lshift.net/blog/2005/09/26/transactions-everywhere#comment-11 Thu, 29 Sep 2005 15:35:29 +0000 http://www.lshift.net/blog/2005/09/26/transactions-everywhere#comment-11 <p>There is one other, rather important question <em>Are STMs the right abstraction?</em> The literature suggests that they really only work well in shared memory architectures, and when there is little contention.</p> There is one other, rather important question Are STMs the right abstraction? The literature suggests that they really only work well in shared memory architectures, and when there is little contention.

]]>