1 02/27/85 response_meters
2
3 Syntax as a command: response_meters -control_args
4
5
6 Function: displays statistics on interactive response time.
7
8
9 Control arguments:
10 -report_reset, -rr
11 generates a report and then performs the reset operation.
12 -reset, -rs
13 resets the metering interval for the invoking process so that the
14 interval begins at the last call with -reset specified. If -reset
15 has never been given in the process, it is equivalent to having been
16 specified at system initialization time.
17 -total, -totals, -tt
18 prints summary information for all work classes.
19 -work_class N, -wc N
20 prints information only for work class number N.
21
22
23 Access required: This command requires access to phcs_ or
24 metering_gate_.
25
26
27 Notes: If neither -reset nor -report_reset is specified,
28 response_meters prints a report.
29
30 The control arguments -total and -work_class cannot be used together.
31 If either of these is used, the remaining control arguments must be
32 such that response_meters prints a report.
33
34
35 The fundamental concept needed to understand the output produced by the
36 response_meters command is that of a terminal interaction. A terminal
37 interaction is essentially an atomic unit of work by a Multics user at
38 a remote terminal. A terminal interaction is triggered by a set of
39 characters keyed by a user at a terminal, which are passed into the
40 user ring as a unit. The activity between the receipt of the set of
41 characters at the Multics mainframe and the next request by user-ring
42 software for more input is a terminal interaction. For a typical user,
43 a terminal interaction corresponds to a line of input. Some user-ring
44 software, however, receives input as sets of characters rather than as
45 lines the Emacs editor is an example of such software. For software
46 of this sort, a terminal interaction corresponds to the set of
47 characters handed out by ring 0 TTY software at one time. If, in
48 processing a set of characters keyed by a user at a terminal, the
49 user-ring software blocks itself on anything other than terminal input,
50 that activity is not considered a terminal interaction. For example,
51 any command that results in tape I/O is not considered a terminal
52 interaction.
53
54
55 The output of response_meters is summarized by work class. Any work
56 class for which no interactions were measured is not displayed. The
57 following are brief descriptions of the variables printed out by
58 response_meters for each work class and for the system.
59 WC
60 is the work class number. Work class "All" represents statistics
61 for the entire system.
62 # Thinks
63 is the number of times a process in the work class blocked itself
64 awaiting terminal input. The term "Think" is used because the delay
65 represented here corresponds to what is commonly called "Think Time"
66 in models of interactive computer systems. This term is derived
67 from the assumption that during this time the interactive user is
68 cogitating about the response just received and is formulating an
69 appropriate next input.
70
71
72 Avg Thinks
73 is the average time in seconds for all blocks awaiting terminal
74 input. This is the average external delay between interactions to
75 the same process in the work class. This external delay includes
76 communications processing time in the FNP, communications line
77 delay, and user idle or "Think" time.
78 # Queues
79 is the number of times a process in the work class was queued for
80 eligibility following receipt of terminal input. It may be less
81 than the number of interactions because of type-ahead. That is, by
82 keying input sufficiently rapidly, a user may cause several
83 interactions to be processed during the same eligibility period. It
84 may be larger than the number of Thinks because of perturbations
85 associated with process initiation.
86
87
88 Avg Queues
89 is the average eligibility queue time in seconds following the
90 receipt of terminal input. This is the average clock time between
91 the receipt of the terminal input at the Multics mainframe and the
92 awarding of eligibility to the target process by traffic control.
93 The average queue time corresponds to what is called "response time"
94 in the output of traffic_control_meters.
95 Load Control Group
96 is the set of Load Control Groups corresponding to the work class.
97 Read access to the Answer Table and to the Master Group Table MGT
98 is required for this information to be displayed. If the requesting
99 process does not have this access, then this field is left blank.
100
101
102 Response times by VCPU Range
103 Up to four detail lines and one summary line are printed, with each
104 line corresponding to a set of interactions. Each interaction
105 measured is categorized by the amount of virtual CPU time consumed
106 by the interaction. Based on the virtual CPU time for the
107 interaction, the interaction is recorded against one of four
108 "buckets." Each of the detail lines represents all terminal
109 interactions for processes in the work class that fell into the
110 corresponding "bucket." A bucket is simply a range of virtual
111 times. Intuitively, the first bucket represents trivial
112 interactions, the last bucket represents exceptionally long
113 interactions, and the remaining two buckets lie between these
114 extremes. If there were no interactions recorded against a bucket,
115 the detail line corresponding to that bucket is not printed. The
116 summary line contains information on all terminal interactions for
117 processes in the work class.
118
119
120 VCPU Range
121 From
122 To
123 is the range of virtual CPU times for interactions presented on that
124 line, in seconds. "From" is the beginning of the range, and "To" is
125 the end of the range. The summary line is identified by "-----" in
126 both the "From" and the "To" fields.
127 # Int
128 is the number of interactions in the bucket for the line. That is,
129 this is the number of interactions whose virtual CPU time fell
130 between the values of "From" and "To."
131 Avg VCPU
132 is the average virtual CPU time per interaction for the bucket.
133
134
135 Avg RT
136 is the average response time per interaction for the bucket. It is
137 the average clock time required to process an interaction, including
138 any eligibility queue time.
139 Resp Fact
140 is the response factor. This is defined as the ratio of the average
141 response time per interaction to the average virtual CPU time per
142 interaction. Intuitively, this ratio represents an "extension
143 factor" for virtual CPU time. That is, it is a factor expressing
144 the average amount of clock time required to process an interaction
145 with a given virtual CPU time. This number can be used as a "quick"
146 indicator of work class or system performance.
147
148
149 In addition to the per-work class data described above, the following
150 summary information is displayed.
151 calls to meter_response_time
152 is the number of subroutine calls to the ring 0 routine that
153 collects the raw response time data.
154 invalid transitions
155 is the number of invalid calls to this routine; meter_response_time
156 implements a finite-state automaton that models a terminal user's
157 interaction with the Multics system. The number of invalid
158 transitions is the number of events that did not fit into this
159 models. Typically, invalid transitions result from perturbations
160 associated with process initiation.
161 Overhead
162 is the CPU time consumed by the measurement routine, expressed as a
163 percentage of total system CPU time, and as the average CPU time in
164 milliseconds per call to this routine.