RE: [sv-ec]E-mail Vote: Closes 12am PST October 26th 2007

From: Bresticker, Shalom <shalom.bresticker_at_.....>
Date: Fri Oct 26 2007 - 01:31:18 PDT
Mantis 2055:
 
I'll try to make this my last email on the subject instead of having all
of us repeating the same points over and over again.
 
The current rule can create greate distortions in the coverage bin
distribution. The LRM hides that by illustrating it with a 'nice'
example where the distortion does not occur. So many users probably
don't realize it. But consider the following example:
 
400 values in 100 bins distributes nicely: 100 bins of 4 values.
So do 401 values in 100 bins: 99 bins of 4 values, and 1 of 5 values.
 
But now take 399 values in those 100 bins. Now you get 99 bins of 3
values, and 1 bin of 102 values! 
 
This is certainly non-intuitive, as a normal observer would expect that
if there is 1 less value, then remove 1 value from 1 of the bins, not
remove 1 value from 99 bins and add 98 values to 1 bin. 
 
Furthermore, the a typical use of distributing those values among those
100 bins is to get a finer-grained more-or-less uniform distribution
along the entire range of values. What has now happened to those 399
values is that only 297 of them (approximately 3/4) have been
distributed among the bins, and an entire 1/4 of the range, the last 102
values, have been lumped into a single bin. What's the point? I've lost
the observability of the value distribution in the entire last part of
the range. For example, I use this to see that in my constrained random
environment, I get to the entire range of values with reasonable
distribution. I've now lost that ability.
 
And if we are talking about user calculations, then users now have to
count the number of values they have, see how many bins they have, and
make a calculation to see how serious the distortion is in their case,
for every time they use this feature. Then if it is not nice, they will
have to specify the distribution manually, which can be a big pain.
 
Yes, the change is not back-compatible. But the current rule is simply
bad.
 
Shalom


________________________________

	From: owner-sv-ec@server.eda.org
[mailto:owner-sv-ec@server.eda.org] On Behalf Of Arturo Salz
	Sent: Thursday, October 25, 2007 11:29 PM
	To: Bresticker, Shalom; sv-ec@server.eda-stds.org
	Subject: RE: [sv-ec]E-mail Vote: Closes 12am PST October 26th
2007
	
	

	Shalom,

	 

	I have consulted existing users that echo my concerns. So, the
degree of intuitiveness is perhaps subjective.

	 

	However, I do recognize that sophisticated users may want
different distributions, not limited to the more uniform distribution
you suggest. I would be supportive of an enhancement that extends
coverage distributions to encompass statistical distributions. Examples
of such distributions are: Gaussian , Normal. Cauchy, Binomial, Poisson,
Gamma, Exponential, Weibull, Lognormal, Chi-Square, etc...

	 

	Syntactically, we might add such distributions by including the
optional distribution as part of the declaration, as in:

	 

	bins sizes [10:Gaussian] = { ... };

	 

	bins addresses [10:Uniform] = { ... };

	 

	Then, we would only be arguing about the default distribution,
which I would suggest we leave as is due to backward compatibility and
the reasons I've already stated.

	 

	            Arturo

	 

	
________________________________


	From: Bresticker, Shalom [mailto:shalom.bresticker@intel.com] 
	Sent: Thursday, October 25, 2007 1:02 PM
	To: Arturo Salz; sv-ec@eda-stds.org
	Subject: RE: [sv-ec]E-mail Vote: Closes 12am PST October 26th
2007

	 

	The new distribution is better because it is far more uniform,
which is the entire purpose. The proposed rule is extremely simple. The
email exchange was about how to implement the formula in a
computationally efficient algorithm without going through a lookup
table. That is something quite different. I would say that the current
LRM algorithm is highly non-intuitive and therefore misleading and
unexpected. Also not what users would want.

	 

	Shalom

		 

		 

		 2055  ___ Yes   _X_ No  

		 

		It is unclear how the new distribution is any better. As
I wrote earlier, "the rule needs to be unambiguous, but it must also be
simple enough for users to figure out what happened." Coverage reports
must be actionable, that is, users need to be able to easily determine
how to cover certain bins, and this proposal complicates that
determination.

		I believe that simplicity trounces the need for
uniformity. As proof of how complex the new rules are, I submit the
email exchange between Steven Sharp and Shalom - it took these two
experts several iterations to arrive at a correct formula. Are we
seriously suggesting users must do this computation in their heads? The
current rule is simplistic but predictable.

		 

	
---------------------------------------------------------------------
	Intel Israel (74) Limited
	 
	This e-mail and any attachments may contain confidential
material for
	the sole use of the intended recipient(s). Any review or
distribution
	by others is strictly prohibited. If you are not the intended
	recipient, please contact the sender and delete all copies.

	-- 
	This message has been scanned for viruses and 
	dangerous content by MailScanner <http://www.mailscanner.info/>
, and is 
	believed to be clean. 

---------------------------------------------------------------------
Intel Israel (74) Limited

This e-mail and any attachments may contain confidential material for
the sole use of the intended recipient(s). Any review or distribution
by others is strictly prohibited. If you are not the intended
recipient, please contact the sender and delete all copies.

-- 
This message has been scanned for viruses and
dangerous content by MailScanner, and is
believed to be clean.
Received on Fri Oct 26 01:35:38 2007

This archive was generated by hypermail 2.1.8 : Fri Oct 26 2007 - 01:36:15 PDT