Re: [Webmacro-devel] ContextTool replacement and thread-safety

[Keats <ke...@ea...>]

Sounds great.  I'll have to read up on this!

----- Original Message -----
From: "Brian Goetz" <br...@qu...>
To: "Keats" <ke...@ea...>
Cc: <web...@li...>
Sent: Thursday, July 24, 2003 6:13 PM
Subject: Re: [Webmacro-devel] ContextTool replacement and thread-safety


>
> >The question is, would such an approach reduce the cost of
synchronization
> >to the point where it is worth giving up the quick, unsynchronized
lookups?
> >Should we maintain a separate table for preconfigured, thread-safe tools?
> >Or is hot deploy/undeploy really important?
> >
> >Brian, I know you are expert in this area, so please chime in.
>
> Short answer: the ConcurrentHashMap class from Doug Lea's util.concurrent
> gives you the best of both worlds.  It is a thread-safe Map that avoids
> locking on the common code path.  It is blazingly fast, and a masterpiece
> of coding.  If I hadn't audited the code myself, I wouldn't believe it.
>
> The common case, a lookup for an object already in the map, succeeds
> without locking.  So in the case where you load up the map once at
startup,
> its no slower than HashMap.  (And, unlike other similar attempts at the
> same result, this one is actually thread-safe, under both the old and new
> Java Memory Models.)
>
> So basically, with CHM, its a non issue.  (I'm going to rip out all the
> fancy cache-management crap and replace it with a ConcurrentHashMap.)
>