Delivery-Date: Sun Oct 26 19:44:19 2008
Message-Id: <200810270255.m9R2t5xv025678@bourbon.usc.edu>
To: cs551@merlot.usc.edu
Subject: Re: CS551 final1 - flood packet
Date: Sun, 26 Oct 2008 19:55:05 -0700
From: Bill Cheng <william@bourbon.usc.edu>

Someone wrote:

  > I have an question on flood mechanism which is among
  > 1) duplicate received packet and put into each node's send() queue (so there
  > are n copies of the original packet)
  > 2) use a pointer to received packet, put the pointer into each node's send()
  > queue
  > 
  > 2nd way is obviously scalable, light weighting and efficiency. However I
  > encounter memory corruption problem.
  > Do I have to use a lock when node's are send() by using 2nd way (pointer to
  > packet)?
  > I guess no because nobody modify the data. Is this correct?
  > 
  > If no, then I probably have to add lock send(). This arises an efficiency
  > problem: each send() have to wait each other.
  > So both way have there down side. Which one is more popular design?

I think you can make the 2nd way work without losing
efficienty.  You can use a reference count with the object.
The key here is that the object is read-only (i.e., it does
not change).  When one sender finishes sending the data, it
decrements the reference count.  When the reference count
reaches zero, it deletes the object.

Of course, when you need to increment or decrement the
reference count, you need to lock a mutex.  But when you are
sending data, I don't think you have to lock the object since
the object never changes.  Also, there is no way for the
object to disappear as long as its reference count is > 0.
--
Bill Cheng // bill.cheng@usc.edu <URL:http://merlot.usc.edu/william/usc/>