-
Notifications
You must be signed in to change notification settings - Fork 10
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
- Loading branch information
1 parent
6626079
commit b762303
Showing
1 changed file
with
168 additions
and
1 deletion.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Original file line number | Original file line | Diff line number | Diff line change |
---|---|---|---|
@@ -1,2 +1,169 @@ | |||
Aggregate is a ruby implementation of a statistics aggregator including histogram support | Aggregate is an intuitive ruby implementation of a statistics aggregator | ||
including both default and configurable histogram support. It does this | |||
without recording/storing any of the actual sample values, making it | |||
suitable for tracking statistics across millions/billions of sample | |||
without any impact on performance or memory footprint. Originally | |||
inspired by the Aggregate support in SystemTap (http://sourceware.org/systemtap/) | |||
|
|
||
Aggregates are easy to instantiate, populate with sample data, and examine | |||
statistics: | |||
|
|||
#After instantiation use the << operator to add a sample to the aggregate: | |||
stats = Aggregate.new | |||
|
|||
loop do | |||
# Take some action that generates a sample measurement | |||
stats << sample | |||
end | |||
|
|||
# The number of samples | |||
stats.count | |||
|
|||
# The average | |||
stats.mean | |||
|
|||
# Max sample value | |||
stats.max | |||
|
|||
# Min sample value | |||
stats.min | |||
|
|||
# The standard deviation | |||
stats.std_dev | |||
|
|||
Perhaps more importantly than the basic aggregate statistics detailed above | |||
Aggregate also maintains a histogram of samples. Good explanation of why | |||
its important: http://37signals.com/svn/posts/1836-the-problem-with-averages | |||
|
|||
The histogram is maintained as a set of "buckets". Each bucket represents a | |||
range of possible sample values. The set of all buckets represents the range | |||
of "normal" sample values. By default this is a binary histogram, where | |||
each bucket represents a range twice as large as the preceding bucket i.e. | |||
[1,1], [2,3], [4,5,6,7], [8,9,10,11,12,13,14,15]. The default binary histogram | |||
provides for 128 buckets, theoretically covering the range [1, (2^127) - 1] | |||
(See NOTES below for a discussion on the effects in practice of insufficient | |||
precision.) | |||
|
|||
Binary histograms are useful when we have little idea about what the | |||
sample distribution may look like as almost any positive value will | |||
fall into some bucket. After using binary histograms to determine | |||
the coarse-grained characteristics of your sample space you can | |||
configure a linear histogram to examine it in closer detail. | |||
|
|||
Linear histograms are specified with the three values low, high, and width. | |||
Low and high specifiy a range [low, high) of values included in the | |||
histogram (all others are outliers). Width specifies the number of | |||
values represented by each bucket and therefore the number of | |||
buckets i.e. granularity of the histogram. The histogram range | |||
(high - low) must be a multiple of width: | |||
|
|||
#Want to track aggregate stats on response times in ms | |||
response_stats = Aggregate.new(0, 2000, 50) | |||
|
|||
The example above creates a linear histogram that tracks the | |||
response times from 0 ms to 2000 ms in buckets of width 50 ms. Hopefully | |||
most of your samples fall in the first couple buckets! Any values added to the | |||
aggregate that fall outside of the histogram range are recorded as outliers: | |||
|
|||
# Number of samples that fall below the normal range | |||
stats.outliers_low | |||
|
|||
# Number of samples that fall above the normal range | |||
stats.outliers_high | |||
|
|||
Once a histogram is populated Aggregate provides iterator support for | |||
examining the contents of buckets. The iterators provide both the | |||
number of samples in the bucket, as well as its range: | |||
|
|||
#Examine every bucket | |||
@stats.each do |bucket, count| | |||
end | |||
|
|||
#Examine only buckets containing samples | |||
@stats.each_nonzero do |bucket, count| | |||
end | |||
|
|||
Finally Aggregate contains sophisticated pretty-printing support that for | |||
any given number of columns >= 80 (defaults to 80) and sample distribution | |||
properly sets a marker weight based on the samples per bucket and aligns all | |||
output. Empty buckets are skipped to conserve screen space. | |||
|
|||
# Generate and display an 80 column histogram | |||
puts stats.to_s | |||
|
|||
# Generate and display a 120 column histogram | |||
puts stats.to_s(120) | |||
|
|||
The following code populates both a binary and linear histogram with the same | |||
set of 65536 values generated by rand to produce two histograms: | |||
|
|||
require 'rubygems' | |||
require 'aggregate' | |||
|
|||
# Create an Aggregate instance | |||
binary_aggregate = Aggregate.new | |||
linear_aggregate = Aggregate.new(0, 65536, 8192) | |||
|
|||
65536.times do | |||
x = rand(65536) | |||
binary_aggregate << x | |||
linear_aggregate << x | |||
end | |||
|
|||
puts binary_aggregate.to_s | |||
puts linear_aggregate.to_s | |||
|
|||
** OUTPUT ** | |||
** Binary Histogram** | |||
value |------------------------------------------------------------------| count | |||
1 | | 3 | |||
2 | | 1 | |||
4 | | 5 | |||
8 | | 9 | |||
16 | | 15 | |||
32 | | 29 | |||
64 | | 62 | |||
128 | | 115 | |||
256 | | 267 | |||
512 |@ | 523 | |||
1024 |@ | 970 | |||
2048 |@@@ | 1987 | |||
4096 |@@@@@@@@ | 4075 | |||
8192 |@@@@@@@@@@@@@@@@ | 8108 | |||
16384 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@ | 16405 | |||
32768 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@| 32961 | |||
~ | |||
Total |------------------------------------------------------------------| 65535 | |||
|
|||
** Linear (0, 65536, 4096) Histogram ** | |||
value |------------------------------------------------------------------| count | |||
0 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@ | 4094 | |||
4096 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@| 4202 | |||
8192 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@ | 4118 | |||
12288 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@ | 4059 | |||
16384 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@ | 3999 | |||
20480 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@ | 4083 | |||
24576 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@ | 4134 | |||
28672 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@ | 4143 | |||
32768 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@ | 4152 | |||
36864 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@ | 4033 | |||
40960 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@ | 4064 | |||
45056 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@ | 4012 | |||
49152 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@ | 4070 | |||
53248 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@ | 4090 | |||
57344 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@ | 4135 | |||
61440 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@ | 4144 | |||
Total |------------------------------------------------------------------| 65532 | |||
|
|||
We can see from these histograms that Ruby's rand function does a relatively good | |||
job of distributing returned values in the requested range. | |||
|
|||
** NOTES ** | |||
Ruby doesn't have a log2 function built into Math, so we approximate with | |||
log(x)/log(2). Theoretically log( 2^n - 1 )/ log(2) == n-1. Unfortunately due | |||
to precision limitations, once n reaches a certain size (somewhere > 32) | |||
this starts to return n. The larger the value of n, the more numbers i.e. | |||
(2^n - 2), (2^n - 3), etc fall trap to this errors. Could probably look into | |||
using something like BigDecimal, but for the current purposes of the binary | |||
histogram i.e. a simple coarse-grained view the current implementation is | |||
sufficient. |