Re: [Exist-open] Concurrency issues with BTree classes

[Tom B. <bra...@db...>]

On Saturday, November 3, 2001, at 05:04 AM, Wolfgang Meier wrote:
> thanks for the message. I started to work on the concurrency issue 
> myself on
> a page locking level, but my first approach does not seem to work and
> produces lots of deadlocks.

Yeah, the LockManager class is prone to deadlocks, especially when 
called recursively.  After I stopped using it, and switch to Java 
synchronization exclusively, every concurrency related issue went away.

The way I see it is that the disk head can only be in one place at any 
one time for a single file anyway, so trying to have many descriptors 
open to a single file and manage page locking will simply add overhead 
and complexity to the process that the kernel will only be able to 
remedy via very fast caching anyway, so it all seems to offset itself in 
the end.

The only thing I plan on changing is burying the synchronization a 
little deeper so that preparation code doesn't have to be synchronized.

--
Tom Bradford
The dbXML Project
Open Source Native XML Database
http://www.dbxml.org/