On 22/10/13 22:56, Dimitri Fontaine wrote:
Tom Lane <t...@sss.pgh.pa.us> writes:
Hm.  It's been a long time since college statistics, but doesn't the
entire concept of standard deviation depend on the assumption that the
underlying distribution is more-or-less normal (Gaussian)?  Is there a
I just had a quick chat with a statistician friends of mine on that
topic, and it seems that the only way to make sense of an average is if
you know already the distribution.

In our case, what I keep experiencing with tuning queries is that we
have like 99% of them running under acceptable threshold and 1% of them
taking more and more time.

In a normal (Gaussian) distribution, there would be no query time
farther away from the average than any other, so my experience tells me
that the query time distribution is anything BUT normal (Gaussian).

good reason to suppose that query runtime is Gaussian?  (I'd bet not;
in particular, multimodal behavior seems very likely due to things like
plan changes.)  If not, how much does that affect the usefulness of
a standard-deviation calculation?
I don't know what multi-modal is.


Multi-modal is basically having more than one hump when you graph the 
frequencies of values.

If you gave a series of mathematical questions of varying degrees of difficulty and 
divers areas in mathematics to a group of people between the ages of 20 & 25 
selected at random in New Zealand, then you would have at least 2 humps.  One hump 
would be those who had little mathematical training and/or no interest and those 
that had had more advanced mathematical training and/or were interested in 

You would also get at least 2 humps if you plotted numbers of people under the 
age of 50, with the number of visits to medical practioners.  Basically those 
people with chronic illnesses with those who tend not to have extended periods 
of illness - this implies 2 humps, but it may be more complicated.

Grabbing people at random and getting them to fire a rifle at targets would 
also be multi modal.  A lot of people with low scores and a lessor percentage 
with reasonable scores.  I would expect this to be quite pronounced, people 
with lots of rifle practice will tend to do significantly better.


Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)
To make changes to your subscription:

Reply via email to