|
From: Nicholas N. <nj...@cs...> - 2006-03-14 01:35:28
|
On Mon, 13 Mar 2006, Graydon Hoare wrote: > I've performed a large-ish run (several hours) with massif, and I find that > the numbers which come out of its html results don't seem to add up. For > example, in this profile entry: > > http://people.mozilla.com/~graydon/results/massif.9425.html#b61F64278 > > 3.3 + 3.0 + 2.8 + 1.7 + 1.1 + 1.1 = 13%, yet the entry is supposedly only > responsible for 5.1%. How am I to interpret this? Is it an arithmetic error? > Are the contexts relative portions (eg. is the 13% listed supposed to be > 13%-of-the-5%?) and if so, where is the rest? > > On shorter runs, the numbers appear to add up, so I am thinking this is a > cumulative arithmetic error of some sort. I wonder if anyone else has seen > such things. I think (from what I remember, and judging from the manual) that the percentages are all absolute, ie. a proportion of the total spacetime. So this does look strange. It's possible sometimes that the allocation tree is actually a dag if function pointers are involved. I wonder if that might be confusing Massif in such a way that it does some kind of double-counting somewhere. Nick |