blob: fe78fb4ab818ae932cf65f2af1b3ba0f9b3e4b41 [file] [log] [blame]
osdl.net!shemmingere7f02a52004-08-23 20:21:21 +00001I. About the distribution tables
2
3The table used for "synthesizing" the distribution is essentially a scaled,
4translated, inverse to the cumulative distribution function.
5
6Here's how to think about it: Let F() be the cumulative distribution
7function for a probability distribution X. We'll assume we've scaled
8things so that X has mean 0 and standard deviation 1, though that's not
9so important here. Then:
10
11 F(x) = P(X <= x) = \int_{-inf}^x f
12
13where f is the probability density function.
14
15F is monotonically increasing, so has an inverse function G, with range
160 to 1. Here, G(t) = the x such that P(X <= x) = t. (In general, G may
17have singularities if X has point masses, i.e., points x such that
18P(X = x) > 0.)
19
20Now we create a tabular representation of G as follows: Choose some table
21size N, and for the ith entry, put in G(i/N). Let's call this table T.
22
23The claim now is, I can create a (discrete) random variable Y whose
24distribution has the same approximate "shape" as X, simply by letting
25Y = T(U), where U is a discrete uniform random variable with range 1 to N.
26To see this, it's enough to show that Y's cumulative distribution function,
27(let's call it H), is a discrete approximation to F. But
28
29 H(x) = P(Y <= x)
30 = (# of entries in T <= x) / N -- as Y chosen uniformly from T
31 = i/N, where i is the largest integer such that G(i/N) <= x
32 = i/N, where i is the largest integer such that i/N <= F(x)
33 -- since G and F are inverse functions (and F is
34 increasing)
35 = floor(N*F(x))/N
36
37as desired.
38
39II. How to create distribution tables (in theory)
40
41How can we create this table in practice? In some cases, F may have a
42simple expression which allows evaluating its inverse directly. The
43pareto distribution is one example of this. In other cases, and
44especially for matching an experimentally observed distribution, it's
45easiest simply to create a table for F and "invert" it. Here, we give
46a concrete example, namely how the new "experimental" distribution was
47created.
48
491. Collect enough data points to characterize the distribution. Here, I
50collected 25,000 "ping" roundtrip times to a "distant" point (time.nist.gov).
51That's far more data than is really necessary, but it was fairly painless to
52collect it, so...
53
542. Normalize the data so that it has mean 0 and standard deviation 1.
55
563. Determine the cumulative distribution. The code I wrote creates a table
57covering the range -10 to +10, with granularity .00005. Obviously, this
58is absurdly over-precise, but since it's a one-time only computation, I
59figured it hardly mattered.
60
614. Invert the table: for each table entry F(x) = y, make the y*TABLESIZE
62(here, 4096) entry be x*TABLEFACTOR (here, 8192). This creates a table
63for the ("normalized") inverse of size TABLESIZE, covering its domain 0
64to 1 with granularity 1/TABLESIZE. Note that even with the granularity
65used in creating the table for F, it's possible not all the entries in
66the table for G will be filled in. So, make a pass through the
67inverse's table, filling in any missing entries by linear interpolation.
68
69III. How to create distribution tables (in practice)
70
71If you want to do all this yourself, I've provided several tools to help:
72
731. maketable does the steps 2-4 above, and then generates the appropriate
74header file. So if you have your own time distribution, you can generate
75the header simply by:
76
77 maketable < time.values > header.h
78
792. As explained in the other README file, the somewhat sleazy way I have
80of generating correlated values needs correction. You can generate your
81own correction tables by compiling makesigtable and makemutable with
82your header file. Check the Makefile to see how this is done.
83
843. Warning: maketable, makesigtable and especially makemutable do
85enormous amounts of floating point arithmetic. Don't try running
86these on an old 486. (NIST Net itself will run fine on such a
87system, since in operation, it just needs to do a few simple integral
88calculations. But getting there takes some work.)
89
904. The tables produced are all normalized for mean 0 and standard
91deviation 1. How do you know what values to use for real? Here, I've
92provided a simple "stats" utility. Give it a series of floating point
93values, and it will return their mean (mu), standard deviation (sigma),
94and correlation coefficient (rho). You can then plug these values
95directly into NIST Net.