+/EE360P

distributed systems

git clone https://git.julia.blue/+/EE360P

vijaygarg1init07ca65d

master
8.0 KiB181 linesraw
1<html>
2<!--
3   Licensed to the Apache Software Foundation (ASF) under one or more
4   contributor license agreements.  See the NOTICE file distributed with
5   this work for additional information regarding copyright ownership.
6   The ASF licenses this file to You under the Apache License, Version 2.0
7   (the "License"); you may not use this file except in compliance with
8   the License.  You may obtain a copy of the License at
9
10       http://www.apache.org/licenses/LICENSE-2.0
11
12   Unless required by applicable law or agreed to in writing, software
13   distributed under the License is distributed on an "AS IS" BASIS,
14   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
15   See the License for the specific language governing permissions and
16   limitations under the License.
17-->
18<body>
19  This package consists of a map/reduce application,
20  <i>distbbp</i>,
21  which computes exact binary digits of the mathematical constant &pi;.
22  distbbp is designed for computing the n<sup>th</sup> bit of &pi;,
23  for large n, say n &gt; 100,000,000.
24  For computing the lower bits of &pi;, consider using <i>bbp</i>.
25
26  <h3>The distbbp Program</h3>
27  The main class is DistBbp
28  and the actually computation is done by DistSum jobs.
29  The steps for launching the jobs are:
30 
31  <ol>
32    <li>Initialize parameters.</li>
33    <li>Create a list of sums.</li>
34    <li>Read computed values from the given local directory.</li>
35    <li>Remove the computed values from the sums.</li>
36    <li>Partition the remaining sums into computation jobs.</li>
37    <li>Submit the computation jobs to a cluster and then wait for the results.</li>
38    <li>Write job outputs to the given local directory.</li>
39    <li>Combine the job outputs and print the &pi; bits.</li>
40  </ol>
41
42<table><tr valign=top><td width=420>
43  <h3>The Bits of &pi;</h3>
44  <p>
45The table on the right are the results computed by distbbp.
46</p>
47<ul>
48<p><li>Row 0 to Row 7
49<ul><li>They were computed by a single machine.</li>
50
51    <li>A single run of Row 7 took several seconds.</li>
52</ul></li></p>
53<p><li>Row 8 to Row 14
54<ul><li>They were computed by a 7600-task-capacity cluster.</li>
55    <li>A single run of Row 14 took 27 hours.</li>
56    <li>The computations in Row 13 and Row 14 were completed on May 20, 2009. 
57        It seems that the corresponding bits were never computed before.</li>
58</ul></li></p>
59<p><li>The first part of Row 15 (<tt>6216B06</tt>)
60
61<ul><li>The first 30% of the computation was done in idle cycles of some 
62        clusters spread over 20 days.</li>
63    <li>The remaining 70% was finished over a weekend on <i>Hammer</i>,
64        a 30,000-task-capacity cluster, which was also used for the
65        <a href="http://developer.yahoo.net/blogs/hadoop/2009/05/hadoop_sorts_a_petabyte_in_162.html">petabyte sort benchmark</a>.</li>
66    <li>The log files are available
67        <a href="https://issues.apache.org/jira/secure/attachment/12408543/1e15log.zip">here</a>.</li>
68    <li>The result was posted in
69        <a href="http://developer.yahoo.net/blogs/hadoop/2009/05/hadoop_computes_the_10151st_bi.html">this YDN blog</a>.</li>
70
71</ul></li></p>
72<p><li>The second part of Row 15 (<tt>D3611</tt>)
73<ul><li>The starting position is 1,000,000,000,000,053, totally 20 bits.</li>
74    <li>Two computations, at positions <i>n</i> and <i>n</i>+4, were performed.
75    <li>A single computation was divided into 14,000 jobs
76        and totally 7,000,000 tasks.
77        It took 208 years of CPU-time
78        or 12 days of cluster (with 7600-task-capacity) time.  </li>
79    <li>The log files are available
80        <a href="https://issues.apache.org/jira/secure/attachment/12412297/D3611.tar.gz">here</a>.</li>
81
82    <li>The computations were completed on June 30, 2009.
83        The last bit, the 1,000,000,000,000,072<sup>nd</sup> bit,
84        probably is the highest bit (or the least significant bit) of &pi;
85        computed ever in the history.</li>
86</ul></li></p>
87
88</td><td width=20></td><td>
89<table border=1 width=400 cellpadding=5>
90<tr><th width=30></th><th>Position <i>n</i></th><th>&pi; bits (in hex) starting at <i>n</i></th></tr>
91
92<tr><td align=right>0</td><td align=right>1</td><td><tt>243F6A8885A3</tt><sup>*</sup></td></tr>
93<tr><td align=right>1</td><td align=right>11</td><td><tt>FDAA22168C23</tt></td></tr>
94<tr><td align=right>2</td><td align=right>101</td><td><tt>3707344A409</tt></td></tr>
95<tr><td align=right>3</td><td align=right>1,001</td><td><tt>574E69A458F</tt></td></tr>
96
97<tr><td align=right>4</td><td align=right>10,001</td><td><tt>44EC5716F2B</tt></td></tr>
98<tr><td align=right>5</td><td align=right>100,001</td><td><tt>944F7A204</tt></td></tr>
99<tr><td align=right>6</td><td align=right>1,000,001</td><td><tt>6FFFA4103</tt></td></tr>
100<tr><td align=right>7</td><td align=right>10,000,001</td><td><tt>6CFDD54E3</tt></td></tr>
101<tr><td align=right>8</td><td align=right>100,000,001</td><td><tt>A306CFA7</tt></td></tr>
102
103<tr><td align=right>9</td><td align=right>1,000,000,001</td><td><tt>3E08FF2B</tt></td></tr>
104<tr><td align=right>10</td><td align=right>10,000,000,001</td><td><tt>0A8BD8C0</tt></td></tr>
105<tr><td align=right>11</td><td align=right>100,000,000,001</td><td><tt>B2238C1</tt></td></tr>
106<tr><td align=right>12</td><td align=right>1,000,000,000,001</td><td><tt>0FEE563</tt></td></tr>
107<tr><td align=right>13</td><td align=right>10,000,000,000,001</td><td><tt>896DC3</tt></td></tr>
108
109<tr><td align=right>14</td><td align=right>100,000,000,000,001</td><td><tt>C216EC</tt></td></tr>
110<tr><td align=right>15</td><td align=right>1,000,000,000,000,001</td><td><tt>6216B06</tt> ... <tt>D3611</tt></td></tr>
111</table>
112
113<p><sup>*</sup>
114By representing &pi; in decimal, hexadecimal and binary, we have
115
116<ul><table><tr>
117  <td>&pi;</td><td>=</td><td><tt>3.1415926535 8979323846 2643383279</tt> ...</td>
118</tr><tr>
119  <td></td><td>=</td><td><tt>3.243F6A8885 A308D31319 8A2E037073</tt> ...</td>
120</tr><tr>
121  <td></td><td>=</td><td><tt>11.0010010000 1111110110 1010100010</tt> ...</td>
122
123</td></tr></table></ul>
124The first ten bits of &pi; are <tt>0010010000</tt>.
125</p>
126</td></tr></table>
127
128
129  <h3>Command Line Usages</h3>
130  The command line format is:
131  <ul><pre>
132$ hadoop org.apache.hadoop.examples.pi.DistBbp \
133         &lt;b&gt; &lt;nThreads&gt; &lt;nJobs&gt; &lt;type&gt; &lt;nPart&gt; &lt;remoteDir&gt; &lt;localDir&gt;</pre></ul>
134  And the parameters are:
135  <ul><table>
136    <tr>
137      <td>&lt;b&gt;</td>
138      <td>The number of bits to skip, i.e. compute the (b+1)th position.</td>
139    </tr>
140    <tr>
141      <td>&lt;nThreads&gt;</td>
142      <td>The number of working threads.</td>
143    </tr>
144    <tr>
145      <td>&lt;nJobs&gt;</td><td>The number of jobs per sum.
146    </tr>
147    <tr>
148      <td>&lt;type&gt;</td><td>'m' for map side job, 'r' for reduce side job, 'x' for mix type.</td>
149    </tr>
150    <tr>
151      <td>&lt;nPart&gt;</td>
152      <td>The number of parts per job.</td>
153    </tr>
154    <tr>
155      <td>&lt;remoteDir&gt;</td>
156      <td>Remote directory for submitting jobs.</td>
157    </tr>
158    <tr>
159      <td>&lt;localDir&gt;</td>
160      <td>Local directory for storing output files.</td>
161    </tr>
162  </table></ul>
163   Note that it may take a long time to finish all the jobs when &lt;b&gt; is large.
164   If the program is killed in the middle of the execution, the same command with
165   a different &lt;remoteDir&gt; can be used to resume the execution.  For example, suppose
166   we use the following command to compute the (10^15+57)th bit of &pi;.
167   <ul><pre>
168$ hadoop org.apache.hadoop.examples.pi.DistBbp \
169	 1,000,000,000,000,056 20 1000 x 500 remote/a local/output</pre></ul>
170   It uses 20 threads to summit jobs so that there are at most 20 concurrent jobs.
171   Each sum (there are totally 14 sums) is partitioned into 1000 jobs.
172   The jobs will be executed in map-side or reduce-side.  Each job has 500 parts.
173   The remote directory for the jobs is remote/a and the local directory
174   for storing output is local/output.  Depends on the cluster configuration,
175   it may take many days to finish the entire execution.  If the execution is killed,
176   we may resume it by
177   <ul><pre>
178$ hadoop org.apache.hadoop.examples.pi.DistBbp \
179         1,000,000,000,000,056 20 1000 x 500 remote/b local/output</pre></ul>
180</body>
181</html>