1 02/28/85 system_performance_graph, spg
2
3 Syntax as a command: spg sample_time -control_args
4
5
6 Function: generates a system of graphs that meter information
7 concerning system performance and operation. The output can be
8 directed to a file or to the controlling terminal. Periodically,
9 metering information is presented in an output line. The initial line
10 contains the cumulative values since system initialization. Whenever
11 there is a change in system configuration or any of several parameters
12 affecting system performance, an additional line noting the change is
13 issued before the sample line. In this way, a system of graphs is
14 developed where various metered quantities are plotted against time.
15 Because the sampling is implemented by means of an event call channel,
16 it is possible to use the terminal in a restricted way for other
17 purposes while metering is in progress. All output is produced on the
18 I/O switch spg_output_.
19
20
21 Arguments:
22 sample_time
23 is a decimal integer giving the time, in minutes, desired between
24 meter display lines.
25
26
27 Control arguments:
28 -halt, -ht
29 terminates plotting.
30 -output_file path, -of path
31 directs output to a file called spg_output, or if a path is
32 supplied, directs output the the file specified by path.
33 -short
34 compresses the width of the meter display lines see "Notes" below.
35
36
37 Access required: This command requires access to phcs_ or
38 metering_gate_.
39
40
41 Notes: Description of the output pattern.
42
43 1. An initial line gives the date and time that metering sampling
44 began.
45
46 2. A line is given describing configuration and scheduling parameter
47 settings.
48
49 3. The current state of the meters since system initialization is on
50 the next line where the sample time is replaced by the system
51 initialization line.
52
53
54 4. Each subsequent meter display line gives the incremental status of
55 the meters since the previous line. In addition, whenever the
56 configuration or scheduling parameter settings change, a
57 notification line is interspersed.
58
59
60 The following is a description of the meter display line.
61
62 Each line contains, in the left margin, the time that the sample was
63 taken. Each sample is scheduled to be taken at an exact minute so that
64 the amount that the time given exceeds the minute represents the
65 response time. Strictly interpreted, the time discrepancy is the
66 response time of a trivial request only if the metering computation is
67 less than the quantum and if the command argument sample_time is
68 greater than one minute so that interactive scheduling occurs.
69
70
71 The remainder of the meter display line consists of a sequence of
72 superimpositions over a grid 100 units wide. When the -short control
73 argument is given, the total width of the grid is only 50 units, so all
74 individual components are correspondingly compressed. The "Example"
75 section below shows the output when the command is invoked with the
76 -short control argument. The wider display would, of course, be easier
77 to read. The 100-unit grid is created by vertical bars every 10 spaces
78 with periods at the intervening midpoints between the bars. Over this
79 grid, various metering quantities are superimposed in the order shown
80 below. When the superimposition is complete, the resultant line
81 containing only the last characters superimposed is printed.
82
83
84 1. Time Usage Percentages
85 blank
86 located to the right of y to right margin user processing not in
87 ring 0. The position of y is an estimate; it is a figure that
88 divides user processing into ring 0 and non-ring 0 sections.
89 blank
90 located to the right of s to left of y user processing in ring 0.
91 s
92 time spent handling segment faults.
93 p
94 time spent handling page faults.
95 t
96 time spent in the traffic controller.
97
98
99 i
100 interrupt processing.
101 multiprogramming idle, *
102 nonmultiprogramming idle.
103 blank
104 located from the left margin to left of *'s zero idle.
105
106 2. Other Values
107
108 The current average is determined from samples taken at one-second
109 intervals weighted backwards in time exponentially, with a smoothing
110 constant of 1/64. The effect is to average over roughly the last
111 minute.
112
113
114 q
115 relative to the left margin
116 current average of the ready list length.
117 e
118 relative to the left margin
119 current average of the number of eligible processes.
120 r
121 relative to the left margin
122 current average of the response time in seconds, for trivial
123 requests.
124 Q
125 relative to the left margin
126 average over a sample of quits/minute.
127
128
129 S
130 relative to the left margin
131 average over a sample of schedulings/10 seconds.
132 D
133 relative to the right margin
134 average over a sample of disk read and write traffic in units of
135 pages/.1 seconds. Full scale equals 1000/sec.
136 P
137 relative to the right margin
138 average over a sample of all read and write traffic bulk store
139 and disk in units of pages/.1 seconds. Full scale equals
140 1000/sec.
141
142
143 V
144 relative to the right margin
145 average over a sample of VTOC entry read and write traffic in
146 units of VTOCES/.1 seconds. Full scale equals 1000 VTOCES
147 transferred per second.
148 -
149 relative to the left margin
150 number of load units at the time of the sample. If this number
151 is greater than 100, 100 is assumed.
152 +
153 relative to the left margin
154 number of users at the time of the sample. If this number is
155 greater than 100, 100 is assumed.