-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathREADME.pathview
186 lines (133 loc) · 6.29 KB
/
README.pathview
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
Sketchy documentation for Pip GUI
written by Jeff Mogul, 10 March 2006
Warning: this is probably full of bugs!
+++++++++++++++++++
How to run demo under VMware player:
The two things you need are the vmware player, and
Patrick's pre-fabbed Pip GUI VM (sorry for the pun).
VMware player:
go to vmware.com
under the Products pulldown, find + click on "VMware player"
Download it and follow the instructions for installation.
Pip image:
Find the link on http://issg.cs.duke.edu/pip/
unpack it with winzip. My suggestion is to create a folder
on your Windows desktop called "vmware images" and then extract
the entire zip file into that. I think this will result in
a folder called "pipvm" with all of the relevant files.
I downloaded Patrick's file at about 800 KB/sec, so it doesn't
take too long.
Patrick's instructions [with my notes in brackets]:
You should be able to click on 'pip.vmx' [in the pipvm folder] to
run it. [Vmware might complain about a version mismatch; go ahead
anyway.]
It's set up to use 128MB of RAM and run at 1024x768 . [You should
enter full-screen mode]
[Note that in vmware player, you need to select whether the player
or the virtual machine gets your mouse and keyboard input. Ctrl-G
switches control to the VM; Ctrl-Alt switches control back to the
Windows environment.]
[Just exit the vmware player; this should suspend the virtual
machine until the next time]
If anything goes wrong with the VM, reboot it. You can log in as
root (password=piproot) or patrick (password=pippatrick). To run
the GUI, log in as patrick, run "startx", then run "./demo" in the
resulting xterm.
The trace in the database is "fab5", a mostly disk-bound
read/write trace taken shortly before the NSDI deadline.
+++++++++++++++++++
General stuff:
Tables can be sorted on any column header, the usual way.
The two sliders at the bottom of the window focus the view in time.
The top slider is the start time, and the bottom is the end time, both
relative to the first event of the trace.
The search below each list box filter the contents of the list box.
In each case, a search beginning with "!" is a negative search,
yielding every task/host/path/recognizer not containing the string.
+++++++++++++++++++
Tabs at the top left:
Tree view: shows communication path between nodes for one path
only works if you have selected a specific path
from the "Paths" table. Double-click the
path to see the tree view for it.
Node numbers correspond to Pool numbers in the
"Thread pools" table.
Clicking on a node shows which other nodes in
the tree have the same pool number. Also
shows path, host, and program name at the
bottom of the view pane.
Double-clicking a node pops up the traced events
corresponding to that node (e.g, a Recv()
event, a Task step, and then a Send() event).
Double-clicking on a traced event pops up
the event details (e.g., for a task step, it
shows the resources consumed; for a message,
it shows the latency and size).
The zoom slider at the bottom of the view pane allows
you to zoom in or out.
Timeline view: shows communication and threads timeline for one path
The zoom slider at the bottom of the view pane allows
you to zoom in or out. If you zoom in enough, a horizontal
panning slider will appear.
checkboxes at the bottom of the view pane allow you
to see or suppress subtasks, notices, and messages.
colored bars are threads; thin lines are messages.
If several bars of the same color appear stacked
vertically, these are subtasks.
single-click on a bar to see (at bottom of the view
pane) the path, host, and task.
double-click on a bar to see the traced events for the thread.
Double-clicking on a traced event pops up
the event details (e.g., for a task step, it
shows the resources consumed; for a message,
it shows the latency and size).
Communication: shows host-wise communication adjacencies for path
The zoom slider at the bottom of the view pane allows
you to zoom in or out.
The layout is automatic; you have a choice of layout
algorithm.
Performance: this tab changes depending on whether you've
most recently requested a task graph or a paths graph
To graph "Performance: tasks", select one or more
rows in the "Tasks" table. (Shift-click extends
selection; Ctrl-click adds to or deletes from selection.)
Then click on the "Graph" button above the "Tasks" table.
To graph "Performance: paths": there seem to be two ways
to do this, depending on whether you want to graph just
one recognizer or more than one. There are also things
that seem to sort-of work, but which end up yielding
confusing results ... beware.
For multiple recognizers:
1. If the "Filter" button above the "Recognizers"
table is pushed in, unpush it.
2. select one or more rows in the "Recognizers"
table. (Shift-click extends selection; Ctrl-click
adds to or deletes from selection.)
3. Now push in the "Filter" button. This populates
the "Paths" table with paths from the union of the
selected recognizers.
4. Now click "Graph" button above the "Paths" table.
For one recognizer:
1. double-click one row in the "Recognizers" table.
This auto-pushes the "Filter" button and populates
the "Paths" table with paths from the selected recognizer.
2. Now click "Graph" button above the "Paths" table.
You have your choice of "quantity" (i.e., metric) for any
graph (tasks or paths). Some metrics are not meaningful
for both graph types (e.g., "network bytes" is not
meaningul for task performance graphs). Other metrics
might just be too boring. ("Real time," the default metric,
is often a good bet, though.)
You have your choice of graphing the chosen metric
as a CDF, PDF, or time series.
You can choose whether the x-scale is log or linear.
(Log is probably a bad idea for time series, but can be a
good idea for CDFs and PDFs.)
You can use the "Max points" slider to adjust the
number of points plotted for a CDF. This setting currently
has no effect for PDFs or time series.
+++++++++++++++++++
Still to document:
How to use "mismatch details"
details of column headers in "Recognizers" table
how do checkboxes get checked in "Recognizers" table?