You are viewing paulmck

(no subject)

Hi Paul -

Thanks for putting together this excellent write-up. I have two questions for you.

1. Do you know where I can find out more about suspended transactions on Power? I googled it and I'm not finding anything.

2. I agree with you that if we have full escape actions, we can using ad hoc debugging techniques and breakpoint debugging for TM programs ... at least in theory. However, I see problems with both of these techniques even if we have support for them with HTM.

First, breakpoint debugging seems, at least to me, to be primarily a sequential program debugging technique. Trying to use breakpoints to reproduce a concurrency bug seems challenging. Second, using ad hoc debugging, such as printf() for pattern analysis, may introduce significant overhead to transactional executions, most especially if those transactions are being executed primarily in hardware. This overhead could change the contention signature of the program and therefore lead to either hiding correctness bugs and/or obfuscating real performance bottlenecks, preventing the programmer from seeing anything useful. Or worse, showing the programmer a performance issue that only exists due to the ad hoc techniques' execution perturbation.

So, my question is to you is this. Let's suppose we have full escape action support. Given the points above, do you believe the current debugging techniques are sufficient for debugging TM programs?

Thanks, as always, for sharing your thoughts and thinking deeply about these questions and problems with me. :)

Justin


Comment Form

No HTML allowed in subject

  
 
   
 

Notice! This user has turned on the option that logs your IP address when posting. 

(will be screened)