1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
|
<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.0 Transitional//EN">
<!--
(c) Copyright 2000-2004 Silicon Graphics Inc. All rights reserved.
Permission is granted to copy, distribute, and/or modify this document
under the terms of the Creative Commons Attribution-Share Alike, Version
3.0 or any later version published by the Creative Commons Corp. A copy
of the license is available at
https://creativecommons.org/licenses/by-sa/3.0/us/ .
-->
<HTML>
<HEAD>
<meta http-equiv="content-type" content="text/html; charset=utf-8">
<meta http-equiv="content-style-type" content="text/css">
<link href="pcpdoc.css" rel="stylesheet" type="text/css">
<link href="images/pcp.ico" rel="icon" type="image/ico">
<TITLE>Understanding system-level processor performance</TITLE>
</HEAD>
<BODY LANG="en-AU" TEXT="#000060" DIR="LTR">
<TABLE WIDTH="100%" BORDER=0 CELLPADDING=0 CELLSPACING=0 STYLE="page-break-before: always">
<TR> <TD WIDTH=64 HEIGHT=64><FONT COLOR="#000080"><A HREF="https://pcp.io/"><IMG SRC="images/pcpicon.png" ALT="pmcharticon" ALIGN=TOP WIDTH=64 HEIGHT=64 BORDER=0></A></FONT></TD>
<TD WIDTH=1><P> </P></TD>
<TD WIDTH=500><P ALIGN=LEFT><A HREF="index.html"><FONT COLOR="#cc0000">Home</FONT></A> · <A HREF="lab.pmchart.html"><FONT COLOR="#cc0000">Charts</FONT></A> · <A HREF="timecontrol.html"><FONT COLOR="#cc0000">Time Control</FONT></A></P></TD>
</TR>
</TABLE>
<H1 ALIGN=CENTER STYLE="margin-top: 0.48cm; margin-bottom: 0.32cm"><FONT SIZE=7>Understanding measures of system-level processor performance</FONT></H1>
<TABLE WIDTH="15%" BORDER=0 CELLPADDING=5 CELLSPACING=10 ALIGN=RIGHT>
<TR><TD BGCOLOR="#e2e2e2"><PRE><IMG SRC="images/system-search.png" ALT="" WIDTH=16 HEIGHT=16 BORDER=0> <I>Tools</I><BR>
sar
mpstat
pmchart
pcp-atop
pcp-atopsar
pcp-uptime
</PRE></TD></TR>
</TABLE>
<P>This chapter of the Performance Co-Pilot tutorial provides guidance
on how to interpret and understand the various measures of system-level
processor (CPU) performance.</P>
<P>All modern operating systems collect processor resource utilization at both
the <B>process</B>-level and the <B>system</B>-level. This tutorial
relates specifically to the <B>system</B>-level metrics.</P>
<P><BR></P>
<TABLE WIDTH="100%" BORDER=0 CELLPADDING=0 CELLSPACING=0 BGCOLOR="#e2e2e2">
<TR><TD WIDTH="100%" BGCOLOR="#081c59"><P ALIGN=LEFT><FONT SIZE=5 COLOR="#ffffff"><B>How the system-level CPU time is computed</B></FONT></P></TD></TR>
</TABLE>
<P>
Utilities like <I>mpstat</I>, <I>sar</I> and Performance Co-Pilot (PCP) use a
common collection of system-level CPU performance instrumentation from the kernel.
This instrumentation is based upon statistical sampling of the state of <B>each</B>
CPU in the kernel's software clock interrupt routine which is commonly called
100 times (PCP metric <I><TT>kernel.all.hz</TT></I>) per second on every CPU.</P>
<P>
At each observation a CPU is attributed a quantum of 10 milliseconds of
elapsed time to one of several counters based on the current state of
the code executing on that CPU.</P>
<P>
This sort of statistical sampling is subject to some anomalies,
particularly when activity is strongly correlated with the clock
interrupts, however the distribution of observations over several
seconds or minutes is often an accurate reflection of the true
distribution of CPU time. The kernel profiling mechanisms offer higher
resolution should that be required, however that is beyond the scope
of this document.</P>
<P>
The CPU state is determined by considering what the CPU was doing just
before the clock interrupt, as follows:</P>
<OL>
<LI>
If executing a userspace task (i.e. above the kernel system call
interface for some program) then
<UL>
<LI>
If executing with modified scheduling priority, the state is
<B>nice</B>.
<LI>
Otherwise the state is <B>user</B>.
</UL>
<LI>
If servicing an interrupt, then the state is <B>irq</B>.
<LI>
If this is a virtual CPU being made to wait involuntarily for a real
CPU (i.e. while the hypervisor services another virtual processor),
the state is <B>steal</B>. (Linux only)
<LI>
If executing a hypervisor guest (Linux only), then
<UL>
<LI>
If executing with modified scheduling priority, the state is
both <B>nice</B> and <B>guest_nice</B>.
<LI>
Otherwise the state is both <B>user</B> and <B>guest</B>.
<LI>
If otherwise executing code in the kernel (i.e. below the system call
interface, or a kernel task), then the state is <B>kernel</B>.
<LI>
Otherwise if some I/O is pending, then the state is <B>wait</B>.
<LI>
Otherwise the state is <B>idle</B>.
</OL>
<P>
These states are mutually exclusive and complete, so exactly one state
is assigned for each CPU at each clock interrupt.</P>
<P>
The kernel agent for PCP exports the following metrics:</P>
<CENTER>
<TABLE BORDER="1">
<CAPTION ALIGN="BOTTOM"><B>Table 1: Raw PCP CPU metrics</B></CAPTION>
<TR VALIGN="TOP">
<TH>PCP Metric</TH>
<TH>Semantics</TH>
</TR>
<TR VALIGN="TOP">
<TD><I><TT>kernel.all.cpu.user</TT></I></TD>
<TD>Time counted when in <B>user</B> state.</TD>
</TR>
<TR VALIGN="TOP">
<TD><I><TT>kernel.all.cpu.nice</TT></I></TD>
<TD>Time counted when in <B>nice</B> state. (Linux / UNIX)</TD>
</TR>
<TR VALIGN="TOP">
<TD><I><TT>kernel.all.cpu.sys</TT></I></TD>
<TD>Time counted when in <B>kernel</B> state.</TD>
</TR>
<TR VALIGN="TOP">
<TD><I><TT>kernel.all.cpu.intr</TT></I></TD>
<TD>Time counted when in <B>irq</B> state.</TD>
</TR>
<TR VALIGN="TOP">
<TD><I><TT>kernel.all.cpu.steal</TT></I></TD>
<TD>Time counted when in <B>steal</B> state. (Linux only)</TD>
</TR>
<TR VALIGN="TOP">
<TD><I><TT>kernel.all.cpu.wait.total</TT></I></TD>
<TD>Time counted when in <B>wait</B> state. (Linux / UNIX)</TD>
</TR>
<TR VALIGN="TOP">
<TD><I><TT>kernel.all.cpu.idle</TT></I></TD>
<TD>Time counted when in <B>idle</B> state.</TD>
</TR>
</TABLE>
</CENTER>
<P>
These metrics are all "counters", in units of milliseconds
(cumulative since system boot time).
When displayed with most PCP tools they are "rate converted"
(sampled periodically and the differences between consecutive values
converted to time utilization in units of milliseconds-per-second over
the sample interval).
Since the raw values are aggregated across all CPUs, the time utilization
for any of the metrics above is in the range 0 to N*1000 for an N CPU system;
for some PCP tools this is reported as a percentage in the range 0 to N*100
percent.</P>
<TABLE WIDTH="100%" BORDER=0 CELLPADDING=10 CELLSPACING=20>
<TR><TD BGCOLOR="#e2e2e2" WIDTH="70%"><BR><IMG SRC="images/stepfwd_on.png" ALT="" WIDTH=16 HEIGHT=16 BORDER=0> Using <I>pmchart</I> to display CPU activity (aggregated over all CPUs).<BR>
<PRE><B>
$ pmchart -c CPU
</B></PRE>
<CENTER>
<IMG SRC="images/linux-cpu.png" ALT="" BORDER=0>
</CENTER>
</TD></TR>
</TABLE>
<P><BR></P>
<TABLE WIDTH="100%" BORDER=0 CELLPADDING=0 CELLSPACING=0 BGCOLOR="#e2e2e2">
<TR><TD WIDTH="100%" BGCOLOR="#081c59"><P ALIGN=LEFT><FONT SIZE=5 COLOR="#ffffff"><B>Interrupts</B></FONT></P></TD></TR>
</TABLE>
<P>On Linux, the <B>irq</B> state is further subdivided into components
describing different stages of kernel interrupt handling:</P>
<OL>
<LI>
If executing deferred interrupt handling code, the state is <B>softirq</B>.
<LI>
Otherwise the state is <B>hardirq</B>.
</OL>
<P>
The Linux kernel agent for PCP exports the following CPU interrupt
metrics, the sum of which approximately equals <I><TT>kernel.all.cpu.intr</TT></I>:</P>
<CENTER>
<TABLE BORDER="1">
<CAPTION ALIGN="BOTTOM"><B>Table 2: Raw PCP CPU interrupt metrics</B></CAPTION>
<TR VALIGN="TOP">
<TH>PCP Metric</TH>
<TH>Semantics</TH>
</TR>
<TR VALIGN="TOP">
<TD><I><TT>kernel.all.cpu.irq.soft</TT></I></TD>
<TD>Time counted in <B>softirq</B> state.</TD>
</TR>
<TR VALIGN="TOP">
<TD><I><TT>kernel.all.cpu.irq.hard</TT></I></TD>
<TD>Time counted in <B>hardirq</B> state.</TD>
</TR>
</TABLE>
</CENTER>
<P>
<P><BR></P>
<TABLE WIDTH="100%" BORDER=0 CELLPADDING=0 CELLSPACING=0 BGCOLOR="#e2e2e2">
<TR><TD WIDTH="100%" BGCOLOR="#081c59"><P ALIGN=LEFT><FONT SIZE=5 COLOR="#ffffff"><B>Virtualization</B></FONT></P></TD></TR>
</TABLE>
<P>
On Linux, several additions to the processor utilization metrics from
other platforms have been made, allowing finer-grained observation of
the effects of virtualization.</P>
<P>
In a virtual environment processor cycles are shared amongst virtual
guests on a physical host.
If a guest displays a high <B>steal</B> time, then cycles are being taken
from that guest to serve other purposes.
This indicates one of two possible things:
</P>
<UL>
<LI>
Either the guest is using more than its share of cycles
<LI>
Or the physical server (which may be a hosting provider) is overloaded.
</UL>
<P>
As described earlier the <B>user</B> and <B>nice</B> processor states are
reported <I>inclusive</I> of virtualization guest state.
There are PCP metrics allowing these states to be observed separately.</P>
<TABLE WIDTH="100%" BORDER=0 CELLPADDING=10 CELLSPACING=20>
<TR><TD BGCOLOR="#e2e2e2" WIDTH="70%"><BR><IMG SRC="images/stepfwd_on.png" ALT="" WIDTH=16 HEIGHT=16 BORDER=0> Using <I>pmchart</I> to display separate guest CPU activity (compared to inclusive).<BR>
<PRE><B>
$ pmchart -c CPU -c vCPU
</B></PRE>
<CENTER>
<IMG SRC="images/linux-virt-cpu.png" ALT="" BORDER=0>
</CENTER>
</TD></TR>
</TABLE>
<P>
<CENTER>
<TABLE BORDER="1">
<CAPTION ALIGN="BOTTOM"><B>Table 3: Raw PCP virtualization user time split</B></CAPTION>
<TR VALIGN="TOP">
<TD><I><TT>kernel.all.cpu.vuser</TT></I></TD>
<TD>User time when not running a virtual guest, default scheduling priority.</TD>
</TR>
<TR VALIGN="TOP">
<TD><I><TT>kernel.all.cpu.guest</TT></I></TD>
<TD>User time when running a guest, default scheduling priority.</TD>
</TR>
</TABLE>
</CENTER>
</P>
<P>
<CENTER>
<TABLE BORDER="1">
<CAPTION ALIGN="BOTTOM"><B>Table 4: Raw PCP virtualization nice time split</B></CAPTION>
<TR VALIGN="TOP">
<TD><I><TT>kernel.all.cpu.vnice</TT></I></TD>
<TD>User time when not running a virtual guest, modified scheduling priority.</TD>
</TR>
<TR VALIGN="TOP">
<TD><I><TT>kernel.all.cpu.guest_nice</TT></I></TD>
<TD>User time when running a guest, modified scheduling priority.</TD>
</TR>
</TABLE>
</CENTER>
</P>
<P><BR></P>
<TABLE WIDTH="100%" BORDER=0 CELLPADDING=0 CELLSPACING=0 BGCOLOR="#e2e2e2">
<TR><TD WIDTH="100%" BGCOLOR="#081c59"><P ALIGN=LEFT><FONT SIZE=5 COLOR="#ffffff"><B>The per-CPU variants</B></FONT></P></TD></TR>
</TABLE>
<P>
Inside the kernel, most of the metrics described above are actually
accumulated per-CPU for reasons of efficiency (to reduce the locking
overheads and minimize dirty cache-line traffic).</P>
<P>
PCP exports the per-CPU versions of the system-wide metrics with metric
names formed by replacing <B><I><TT>all</TT></I></B> by <B><I><TT>percpu</TT></I></B>,
e.g. <I><TT>kernel.percpu.cpu.user</TT></I>.</P>
<P>
Note that on some multiprocessor UNIX platforms with one I/O pending,
<B>all</B> otherwise idle CPUs will be assigned the <B>wait</B> state.
This may lead to an over-estimate of the I/O wait time, as discussed in the
companion <A HREF="howto.diskperf.html">How to understand measures of
disk performance</A> document.
This is not the case on Linux, however, where pre-CPU request queues
track counts of processes in I/O wait state.</P>
<P>
The <I>pcp-atop</I> and <I>pcp-atopsar</I> tools can be used to report
the per-CPU utilization metrics, the latter in a fashion similar to the
<I>mpstat</I> <B>-I CPU</B> option.
<P>
<P><BR></P>
<TABLE WIDTH="100%" BORDER=0 CELLPADDING=0 CELLSPACING=0 BGCOLOR="#e2e2e2">
<TR><TD WIDTH="100%" BGCOLOR="#081c59"><P ALIGN=LEFT><FONT SIZE=5 COLOR="#ffffff"><B>Reconciling mpstat, sar -u and PCP CPU performance metrics</B></FONT></P></TD></TR>
</TABLE>
<P>
The <I>mpstat</I> and <I>sar</I> metrics are scaled based on the number of CPUs
and expressed in percentages, PCP metrics are in units of milliseconds per
second after rate conversion; this explains the PCP metric <I><TT>hinv.ncpu</TT></I>
and the constants 100 and 1000 in the expressions below.</P>
<P>
When run without options, <I>sar</I> reports the following:</P>
<CENTER>
<TABLE BORDER="1">
<CAPTION ALIGN="BOTTOM"><B>Table 5: PCP and sar (default) metric equivalents</B></CAPTION>
<TR>
<TH><I>sar</I><BR> metric</TH>
<TH>PCP equivalent (assuming rate conversion)</TH>
</TR>
<TR VALIGN="TOP">
<TD><I><TT>%user</TT></I></TD>
<TD>100 * <I><TT>kernel.all.cpu.user</TT></I> / (<I><TT>hinv.ncpu</TT></I> * 1000)</TD>
</TR>
<TR VALIGN="TOP">
<TD><I><TT>%nice</TT></I></TD>
<TD>100 * <I><TT>kernel.all.cpu.nice</TT></I> / (<I><TT>hinv.ncpu</TT></I> * 1000)</TD>
</TR>
<TR VALIGN="TOP">
<TD><I><TT>%system</TT></I></TD>
<TD>100 * (<I><TT>kernel.all.cpu.sys</TT></I> + <I><TT>kernel.all.cpu.intr</TT></I>) / (<I><TT>hinv.ncpu</TT></I> * 1000)</TD>
</TR>
<TR VALIGN="TOP">
<TD><I><TT>%iowait</TT></I></TD>
<TD>100 * <I><TT>kernel.all.cpu.wait.total</TT></I> / (<I><TT>hinv.ncpu</TT></I> * 1000)</TD>
</TR>
<TR VALIGN="TOP">
<TD><I><TT>%steal</TT></I></TD>
<TD>100 * <I><TT>kernel.all.cpu.steal</TT></I> / (<I><TT>hinv.ncpu</TT></I> * 1000)</TD>
</TR>
<TR VALIGN="TOP">
<TD><I><TT>%idle</TT></I></TD>
<TD>100 * <I><TT>kernel.all.cpu.idle</TT></I> / (<I><TT>hinv.ncpu</TT></I> * 1000)</TD>
</TR>
</TABLE>
</CENTER>
<P>
On the other hand, <I>mpstat</I>, and <I>sar</I> with the <B>-u ALL</B> option,
report the following:</P>
<CENTER>
<TABLE BORDER="1">
<CAPTION ALIGN="BOTTOM"><B>Table 6: PCP and mpstat (sar -u ALL) metric equivalents</B></CAPTION>
<TR>
<TH><I>mpstat</I><BR> metric</TH>
<TH>PCP equivalent (assuming rate conversion)</TH>
</TR>
<TR VALIGN="TOP">
<TD><I><TT>%usr</TT></I></TD>
<TD>100 * <I><TT>kernel.all.cpu.vuser</TT></I> / (<I><TT>hinv.ncpu</TT></I> * 1000)</TD>
</TR>
<TR VALIGN="TOP">
<TD><I><TT>%nice</TT></I></TD>
<TD>100 * <I><TT>kernel.all.cpu.vnice </TT></I> / (<I><TT>hinv.ncpu</TT></I> * 1000)</TD>
</TR>
<TR VALIGN="TOP">
<TD><I><TT>%sys</TT></I></TD>
<TD>100 * <I><TT>kernel.all.cpu.sys</TT></I> / (<I><TT>hinv.ncpu</TT></I> *
1000)</TD>
</TR>
<TR VALIGN="TOP">
<TD><I><TT>%iowait</TT></I></TD>
<TD>100 * <I><TT>kernel.all.cpu.wait.total</TT></I> / (<I><TT>hinv.ncpu</TT></I> * 1000)</TD>
</TR>
<TR VALIGN="TOP">
<TD><I><TT>%steal</TT></I></TD>
<TD>100 * <I><TT>kernel.all.cpu.steal</TT></I> / (<I><TT>hinv.ncpu</TT></I> * 1000)</TD>
</TR>
<TR VALIGN="TOP">
<TD><I><TT>%irq</TT></I></TD>
<TD>100 * <I><TT>kernel.all.cpu.irq.hard</TT></I> / (<I><TT>hinv.ncpu</TT></I> * 1000)</TD>
</TR>
<TR VALIGN="TOP">
<TD><I><TT>%soft</TT></I></TD>
<TD>100 * <I><TT>kernel.all.cpu.irq.soft</TT></I> / (<I><TT>hinv.ncpu</TT></I> * 1000)</TD>
</TR>
<TR VALIGN="TOP">
<TD><I><TT>%guest</TT></I></TD>
<TD>100 * <I><TT>kernel.all.cpu.guest</TT></I> / (<I><TT>hinv.ncpu</TT></I> * 1000)</TD>
</TR>
<TR VALIGN="TOP">
<TD><I><TT>%gnice</TT></I></TD>
<TD>100 * <I><TT>kernel.all.cpu.guest_nice</TT></I> / (<I><TT>hinv.ncpu</TT></I> * 1000)</TD>
</TR>
<TR VALIGN="TOP">
<TD><I><TT>%idle</TT></I></TD>
<TD>100 * <I><TT>kernel.all.cpu.idle</TT></I> / (<I><TT>hinv.ncpu</TT></I> * 1000)</TD>
</TR>
</TABLE>
</CENTER>
<P><BR></P>
<TABLE WIDTH="100%" BORDER=0 CELLPADDING=0 CELLSPACING=0 BGCOLOR="#e2e2e2">
<TR><TD WIDTH="100%" BGCOLOR="#081c59"><P ALIGN=LEFT><FONT SIZE=5 COLOR="#ffffff"><B>The load average</B></FONT></P></TD></TR>
</TABLE>
<P>
The "load average" is reported by <I>pcp-uptime</I>, <I>pcp-atop</I>, etc.
and the PCP metric <I><TT>kernel.all.load</TT></I>.</P>
<TABLE WIDTH="100%" BORDER=0 CELLPADDING=10 CELLSPACING=20>
<TR><TD BGCOLOR="#e2e2e2" WIDTH="70%"><BR><IMG SRC="images/stepfwd_on.png" ALT="" WIDTH=16 HEIGHT=16 BORDER=0> Using <I>pcp-uptime</I> to display historical uptime and load average measures.<BR>
<PRE><B>
$ source /etc/pcp.conf
$ tar xzf $PCP_DEMOS_DIR/tutorials/cpuperf.tgz
$ pcp --origin 'Tuesday 4:30pm' --archive cpuperf/system -z uptime
16:30:07 up 5:15, 11 users, load average: 1.59, 1.54, 1.23
</B></PRE>
</TD></TR>
</TABLE>
<P>
The load average is an indirect measure of the demand for CPU
resources. It is calculated using the previous load average (<I>load</I>)
the sum of the currently runnable processes (<I>nr_active</I>), and an
exponential dampening expression, e.g. for the "1 minute" average,
the expression is:</P>
<PRE>
load = exp(-5/60) * load + (1 - exp(-5/60)) * nr_active
</PRE>
<P>
The three load averages use different exponential constants and are all
re-computed every 5 seconds.</P>
<P>
On Linux, <I>nr_active</I> is computed as the sum across all processors
(i.e. using per-CPU counters) of the number of running tasks and the number
of uninterruptible tasks.</P>
<P>
Note that the "run queue length" (reported by the <B>-q</B> option
of <I>sar</I>) uses the aggregation across all processors of the number of
running tasks only (does not include uninterruptible tasks).
This is the <I><TT>kernel.all.runnable</TT></I> PCP metric.</P>
<P><BR></P>
<HR>
<CENTER>
<TABLE WIDTH="100%" BORDER=0 CELLPADDING=0 CELLSPACING=0>
<TR> <TD WIDTH="50%"><P>Copyright © 2007-2010 <A HREF="https://www.aconex.com/"><FONT COLOR="#000060">Aconex</FONT></A><BR>Copyright © 2000-2004 <A HREF="https://www.sgi.com/"><FONT COLOR="#000060">Silicon Graphics Inc</FONT></A></P></TD>
<TD WIDTH="50%"><P ALIGN=RIGHT><A HREF="https://pcp.io/"><FONT COLOR="#000060">PCP Site</FONT></A><BR>Copyright © 2012-2018 <A HREF="https://www.redhat.com/"><FONT COLOR="#000060">Red Hat</FONT></A></P></TD> </TR>
</TABLE>
</CENTER>
</BODY>
</HTML>
|
