[Fwd: Re: Is JAMA thread-safe]

[Ron Boisvert <boisvert@nist.gov>]

Sender: "Ray A. Conner" <raymond.a.conner@saic.com>
Subject: Re: Is JAMA thread-safe


Jama isn't thread-safe.  My personal opinion is that it shouldn't be, 
since the majority of users don't care and this would incur some 
overhead, although not much most of the time.  Any user who requires 
thread-safety can easily enough do that in their own code.

To go into another of my long-winded explanations.... to be thread-safe 
an object must behave properly if two (or more) threads of execution 
simultaneously use it.  The term "thread-safe" is not quite descriptive 
enough, but for Matrix objects there's really only three possible ways 
for Jama to work:

1) Objects are completely unsynchronized (what it is now).

2) Matrix instances are immutable, which makes synchronization 
unnecessary.  This isn't feasible because of the storage and time hits 
that would take, there would be no set methods, plusEquals, etc.

3) Matrix methods contain sufficient synchronization to keep concurrent 
threads from stomping on each other within methods.  While this would be 
easy to do, I don't think it's very useful.  The reason is that it 
wouldn't help you if the two threads were performing sequences of 
operations, for example:

    Thread A                    Thread B
  ----------------------------------------
  op#1: x.plusEquals( a )     op#3: x.arrayTimesEquals( y )
  op#2: x.plusEquals( b )     op#4: x.arrayTimesEquals( z )

Even if the methods plusEquals() and arrayTimesEquals() were 
synchronized so that they couldn't clobber each other, there are
six possible final states for the matrix x depending on how the 
operations interleave.  Sequences of operations are the rule when 
dealing with matrices, not the exception.  Only the calling application 
can possibly know what sequences of operations must be invoked without 
interference, and so the synchronization must be applied at that level.

- Ray A. Conner


Cleve Moler wrote:
> I certainly never considered that issue when I wrote much of the code a
> few years ago.  What does it take to make it thread safe?
> 
>   -- Cleve Moler
>      The MathWorks
> 
> -----Original Message-----
> From: jama@nist.gov [mailto:jama@nist.gov] On Behalf Of Geri Wagner
> Sent: Monday, October 09, 2006 10:27 AM
> To: Multiple recipients of list
> Subject: Is JAMA thread-safe
> 
> Hi
> 
> I wonder whether the JAMA implementation is thread-safe. 
> 
> I could not find a statetment on the website, but perhaps I have 
> overlooked it. Thank you for comments.
>