rbsim 0.0.3
Sign up to get free protection for your applications and to get access to all the features.
- checksums.yaml +7 -0
- data/.hgignore +6 -0
- data/.rspec +2 -0
- data/Gemfile +12 -0
- data/Gemfile.lock +66 -0
- data/LICENSE.txt +674 -0
- data/README.md +960 -0
- data/TODO +28 -0
- data/basic_sim.rb +62 -0
- data/fast-tcpn.rb +3 -0
- data/lib/rbsim.rb +14 -0
- data/lib/rbsim/dsl.rb +30 -0
- data/lib/rbsim/dsl/infrastructure.rb +48 -0
- data/lib/rbsim/dsl/mapping.rb +32 -0
- data/lib/rbsim/dsl/process.rb +129 -0
- data/lib/rbsim/dsl/program.rb +10 -0
- data/lib/rbsim/experiment.rb +110 -0
- data/lib/rbsim/hlmodel.rb +25 -0
- data/lib/rbsim/hlmodel/infrastructure.rb +116 -0
- data/lib/rbsim/hlmodel/mapping.rb +5 -0
- data/lib/rbsim/hlmodel/process.rb +152 -0
- data/lib/rbsim/numeric_units.rb +107 -0
- data/lib/rbsim/simulator.rb +184 -0
- data/lib/rbsim/statistics.rb +77 -0
- data/lib/rbsim/tokens.rb +146 -0
- data/lib/rbsim/version.rb +3 -0
- data/new_process.rb +49 -0
- data/rbsim.gemspec +42 -0
- data/show_readme.rb +15 -0
- data/sim.rb +142 -0
- data/sim_bamboo.rb +251 -0
- data/sim_process.rb +83 -0
- data/sim_process_dsl.rb +58 -0
- data/spec/dsl/infrastructure_nets_spec.rb +39 -0
- data/spec/dsl/infrastructure_nodes_spec.rb +72 -0
- data/spec/dsl/infrastructure_routes_spec.rb +44 -0
- data/spec/dsl/mapping_spec.rb +70 -0
- data/spec/dsl/process_spec.rb +56 -0
- data/spec/dsl/program_spec.rb +36 -0
- data/spec/dsl_and_hlmodel/new_process_spec.rb +235 -0
- data/spec/hlmodel/net_spec.rb +112 -0
- data/spec/hlmodel/process_spec.rb +242 -0
- data/spec/hlmodel/route_spec.rb +47 -0
- data/spec/hlmodel/routes_spec.rb +44 -0
- data/spec/integration/basic_simulation_spec.rb +104 -0
- data/spec/integration/net_spec.rb +44 -0
- data/spec/integration/process_spec.rb +117 -0
- data/spec/integration/rbsim_spec.rb +40 -0
- data/spec/simulator/logger_spec.rb +35 -0
- data/spec/simulator/stats_spec.rb +93 -0
- data/spec/spec_helper.rb +26 -0
- data/spec/statistics_spec.rb +300 -0
- data/spec/tcpn/add_route_spec.rb +55 -0
- data/spec/tcpn/cpu_spec.rb +53 -0
- data/spec/tcpn/map_data_spec.rb +37 -0
- data/spec/tcpn/network_spec.rb +163 -0
- data/spec/tcpn/register_event_spec.rb +48 -0
- data/spec/tcpn/route_to_self_spec.rb +53 -0
- data/spec/tcpn/stats_spec.rb +77 -0
- data/spec/tokens/data_queue_obsolete.rb +121 -0
- data/spec/tokens/data_queue_spec.rb +111 -0
- data/spec/units_spec.rb +48 -0
- data/tcpn/model.rb +6 -0
- data/tcpn/model/add_route.rb +78 -0
- data/tcpn/model/application.rb +250 -0
- data/tcpn/model/cpu.rb +75 -0
- data/tcpn/model/map_data.rb +42 -0
- data/tcpn/model/network.rb +108 -0
- data/tcpn/model/register_event.rb +89 -0
- data/tcpn/model/stats.rb +46 -0
- metadata +221 -0
data/README.md
ADDED
@@ -0,0 +1,960 @@
|
|
1
|
+
RBSim -- software technical reference
|
2
|
+
=====
|
3
|
+
|
4
|
+
This is a technical description of the RBSim software package.
|
5
|
+
|
6
|
+
RBSim is a simulation tool designed for analysis of architecture of
|
7
|
+
concurrent and distributed applications. Application should be described
|
8
|
+
using convenient Ruby-based DSL. Simulation is based on Timed Colored
|
9
|
+
Petri nets designed by K. Jensen, thus ensuring reliable analysis. Basic
|
10
|
+
statistics module is included.
|
11
|
+
|
12
|
+
In order to provide quick overview of the method here you will
|
13
|
+
find example usage of RBSim. Detailed description of model,
|
14
|
+
simulation and processing statistics can be found in the
|
15
|
+
following sections.
|
16
|
+
|
17
|
+
|
18
|
+
# Example
|
19
|
+
|
20
|
+
Below there is a simple but complete example model of two
|
21
|
+
applications: `wget` sending subsequent requests to specified
|
22
|
+
process and `apache` responding to received requests.
|
23
|
+
|
24
|
+
The `wget` program accepts parameters describing its target
|
25
|
+
(destination of requests) -- `opts[:target]` and count of
|
26
|
+
requests to send -- `opts[:count]`. The parameters are defined
|
27
|
+
when new process is defined using this program. The `apache`
|
28
|
+
program takes no additional parameters.
|
29
|
+
|
30
|
+
Two client processes are started using program `wget`: `client1`
|
31
|
+
and `client2`. Using `apache` program one server is started:
|
32
|
+
`server`. Application uses two nodes: `desktop` with one slower
|
33
|
+
processor and `gandalf` with one faster CPU. The nodes are
|
34
|
+
connected by two nets and one two-way route. Both clients are
|
35
|
+
assigned to the `desktop` node while server is run on `gandalf`.
|
36
|
+
|
37
|
+
The clients are mapped to the `desktop` node and the server is
|
38
|
+
assigned to `gandalf`.
|
39
|
+
|
40
|
+
Logs are printed to STDOUT and statistics are collected. Apache
|
41
|
+
`stats` definitions allow to observe time taken by serving
|
42
|
+
requests. Client `stats` count served requests and allow to
|
43
|
+
verify if responses were received for all sent requests.
|
44
|
+
|
45
|
+
The created model is run and its statistics are printed.
|
46
|
+
|
47
|
+
The `model.rb` file contains model of the application and resources
|
48
|
+
described with DSL:
|
49
|
+
|
50
|
+
```ruby
|
51
|
+
program :wget do |opts|
|
52
|
+
sent = 0
|
53
|
+
on_event :send do
|
54
|
+
cpu do |cpu|
|
55
|
+
(150 / cpu.performance).miliseconds
|
56
|
+
end
|
57
|
+
send_data to: opts[:target], size: 1024.bytes,
|
58
|
+
type: :request, content: sent
|
59
|
+
sent += 1
|
60
|
+
register_event :send, delay: 5.miliseconds if sent < opts[:count]
|
61
|
+
end
|
62
|
+
|
63
|
+
on_event :data_received do |data|
|
64
|
+
log "Got data #{data} in process #{process.name}"
|
65
|
+
stats event: :request_served, client: process.name
|
66
|
+
end
|
67
|
+
|
68
|
+
register_event :send
|
69
|
+
end
|
70
|
+
|
71
|
+
program :apache do
|
72
|
+
on_event :data_received do |data|
|
73
|
+
stats_start server: :apache, name: process.name
|
74
|
+
cpu do |cpu|
|
75
|
+
(100 * data.size.in_bytes / cpu.performance).miliseconds
|
76
|
+
end
|
77
|
+
send_data to: data.src, size: data.size * 10,
|
78
|
+
type: :response, content: data.content
|
79
|
+
stats_stop server: :apache, name: process.name
|
80
|
+
end
|
81
|
+
end
|
82
|
+
|
83
|
+
node :desktop do
|
84
|
+
cpu 100
|
85
|
+
end
|
86
|
+
|
87
|
+
node :gandalf do
|
88
|
+
cpu 1400
|
89
|
+
end
|
90
|
+
|
91
|
+
new_process :client1, program: :wget, args: { target: :server, count: 10 }
|
92
|
+
new_process :client2, program: :wget, args: { target: :server, count: 10 }
|
93
|
+
new_process :server, program: :apache
|
94
|
+
|
95
|
+
net :net01, bw: 1024.bps
|
96
|
+
net :net02, bw: 510.bps
|
97
|
+
|
98
|
+
route from: :desktop, to: :gandalf, via: [ :net01, :net02 ], twoway: true
|
99
|
+
|
100
|
+
put :server, on: :gandalf
|
101
|
+
put :client1, on: :desktop
|
102
|
+
put :client2, on: :desktop
|
103
|
+
```
|
104
|
+
|
105
|
+
Creating empty subclass of `RBsim::Experiment` in `rbsim_example.rb`
|
106
|
+
file is sufficient to start simulation:
|
107
|
+
|
108
|
+
```ruby
|
109
|
+
class Experiment < RBSim::Experiment
|
110
|
+
end
|
111
|
+
```
|
112
|
+
|
113
|
+
The simulation can be started with the following code:
|
114
|
+
|
115
|
+
```ruby
|
116
|
+
require './rbsim_example'
|
117
|
+
|
118
|
+
params = { }
|
119
|
+
|
120
|
+
sim = Experiment.new
|
121
|
+
sim.run './model.rb', params
|
122
|
+
sim.save_stats 'simulation.stats'
|
123
|
+
```
|
124
|
+
|
125
|
+
The `save_stats` method appends statistics to the specified file.
|
126
|
+
Statistics saved can be loaded with:
|
127
|
+
|
128
|
+
```ruby
|
129
|
+
all_stats = Experiment.read_stats 'simulation.stats'
|
130
|
+
```
|
131
|
+
|
132
|
+
The `all_stats` will be an iterator yielding objects of class `Experiement`,
|
133
|
+
so statistics of first experiment can be accessed using:
|
134
|
+
|
135
|
+
```ruby
|
136
|
+
all_stats.first.app_stats # application statistics
|
137
|
+
all_stats.first.res_stats # resource statistics
|
138
|
+
```
|
139
|
+
|
140
|
+
Processing statistics into required results can be implemented in the
|
141
|
+
`Experiement` class (so far this class was empty) like this:
|
142
|
+
|
143
|
+
```ruby
|
144
|
+
class Experiment < RBSim::Experiment
|
145
|
+
|
146
|
+
def print_req_times_for(server)
|
147
|
+
app_stats.durations(server: server) do |tags, start, stop|
|
148
|
+
puts "Request time #{(stop - start).in_miliseconds} ms. "
|
149
|
+
end
|
150
|
+
end
|
151
|
+
|
152
|
+
def mean_req_time_for(server)
|
153
|
+
req_times = app_stats.durations(server: server).to_a
|
154
|
+
sum = req_times.reduce(0) do |acc, data|
|
155
|
+
_, start, stop = *data
|
156
|
+
acc + stop - start
|
157
|
+
end
|
158
|
+
sum / req_times.size
|
159
|
+
end
|
160
|
+
|
161
|
+
end
|
162
|
+
```
|
163
|
+
|
164
|
+
Then, the statistics can be conveniently used like this:
|
165
|
+
|
166
|
+
```ruby
|
167
|
+
all_stats = Experiment.read_stats 'simulation.stats'
|
168
|
+
first_experiment = all_stats.first
|
169
|
+
first_experiment.print_req_times_for(:apache)
|
170
|
+
puts "Mean request time for apache: "
|
171
|
+
puts "#{first_experiment.mean_req_time_for(:apache).in_seconds} s"
|
172
|
+
```
|
173
|
+
|
174
|
+
You can of course iterate over subsequent experiments to get statistics
|
175
|
+
involving a number of different tests.
|
176
|
+
|
177
|
+
In next sections you will find description of subsequent parts of the
|
178
|
+
RBSim tool.
|
179
|
+
|
180
|
+
## Usage
|
181
|
+
|
182
|
+
There are two ways to create model. Preferred one is to subclass the
|
183
|
+
`Experiment` class. But it is also possible to directly create model
|
184
|
+
with RBSim class.
|
185
|
+
|
186
|
+
### Using `RBsim::Experiment` class
|
187
|
+
|
188
|
+
Create a class that inherits `RBsim::Experiment` class.
|
189
|
+
|
190
|
+
```ruby
|
191
|
+
class MyTests < RBSim::Experiment
|
192
|
+
end
|
193
|
+
```
|
194
|
+
|
195
|
+
Thereafter you can use it to load model and perform simulations:
|
196
|
+
|
197
|
+
```ruby
|
198
|
+
sim = MyTests.new
|
199
|
+
sim.run 'path/to/model_file.rb'
|
200
|
+
```
|
201
|
+
|
202
|
+
Finally, you can save statistics gathered from simulation to a file:
|
203
|
+
|
204
|
+
```ruby
|
205
|
+
sim.save_stats 'file_name.stats'
|
206
|
+
```
|
207
|
+
|
208
|
+
If the file exists, new statistics will be appended at its end.
|
209
|
+
|
210
|
+
The saved statistics can be later loaded and analyzed with the same
|
211
|
+
class inheriting from the `RBSim::Experiment`:
|
212
|
+
|
213
|
+
```ruby
|
214
|
+
stats = MyTests.read_stats 'file_name.stats'
|
215
|
+
```
|
216
|
+
|
217
|
+
The `RBSim.read_stats` method will return an array of `MyTests` objects
|
218
|
+
for each experiment saved in the file. The objects can then be used to
|
219
|
+
process statistics as described further.
|
220
|
+
|
221
|
+
You can limit simulation time by setting `sim.time_limit` before you
|
222
|
+
start simulation. In order to use time units you need to `require
|
223
|
+
'rbsim/numeric_units'` before.
|
224
|
+
|
225
|
+
```ruby
|
226
|
+
sim.time_limit = 10.seconds
|
227
|
+
```
|
228
|
+
|
229
|
+
RBSim uses coarse model of network transmission. If you application
|
230
|
+
requires more precise estimation of network transmission time, you can
|
231
|
+
set:
|
232
|
+
|
233
|
+
```ruby
|
234
|
+
sim.data_fragmentation = number_of_fragments
|
235
|
+
```
|
236
|
+
|
237
|
+
This will cause each data package transmitted over the network to be
|
238
|
+
divided into at most `number_of_fragments` (but not smaller then 1500B).
|
239
|
+
If your application efficiency depends significantly on numerous data
|
240
|
+
transmissions influencing each other this may improve accuracy, but it
|
241
|
+
will also increase simulation time. If this is not sufficient and
|
242
|
+
efficiency of your application depends on network transmission time
|
243
|
+
(network bounded), not on application logic and logic of communication
|
244
|
+
in the system, you should probably revert to a dedicated network
|
245
|
+
simulator.
|
246
|
+
|
247
|
+
### Using `RBSim.model`
|
248
|
+
|
249
|
+
You can also define your model using `RBSim.model` method:
|
250
|
+
|
251
|
+
```ruby
|
252
|
+
model = RBSim.model some_params do |params|
|
253
|
+
# define your model here
|
254
|
+
# use params passed to the block
|
255
|
+
end
|
256
|
+
```
|
257
|
+
|
258
|
+
Or read the model from a file:
|
259
|
+
|
260
|
+
```ruby
|
261
|
+
model = RBSim.read file_name, some_params_hash
|
262
|
+
```
|
263
|
+
|
264
|
+
`some_params_hash` will be available in the model loaded from the file
|
265
|
+
as `params` variable.
|
266
|
+
|
267
|
+
|
268
|
+
Run simulator:
|
269
|
+
|
270
|
+
```ruby
|
271
|
+
model.run
|
272
|
+
```
|
273
|
+
|
274
|
+
When simulation is finished, the statistics can be obtained with
|
275
|
+
`model.stats` method.
|
276
|
+
|
277
|
+
## Model
|
278
|
+
|
279
|
+
Use `RBSim.model` to create model described by DSL in a block or
|
280
|
+
`RBSim.read` to load a model from separate file.
|
281
|
+
|
282
|
+
The model is a set of `process`es that are `put` on `nodes` and
|
283
|
+
communicate over `net`s. Processes can be defined by `program`s or
|
284
|
+
directly by blocks, `route`s define sequence of `net`s that should be
|
285
|
+
traversed by data while communication between `node`s. Application
|
286
|
+
logic implemented in `process`es is described in terms of events.
|
287
|
+
|
288
|
+
So to summarize, the most important parts of the model are:
|
289
|
+
|
290
|
+
- Application described as a set of `processes`
|
291
|
+
- Resources described as
|
292
|
+
- `nodes` (computers)
|
293
|
+
- `nets` (network segemtns)
|
294
|
+
- `routes` (from one node to another, over the `net` segments
|
295
|
+
- Mapping of application `processes` to `nodes`
|
296
|
+
|
297
|
+
The application can be modeled independently, resources can be
|
298
|
+
modeled separately and at the end, the application can be mapped
|
299
|
+
to the resources.
|
300
|
+
|
301
|
+
### Processes
|
302
|
+
|
303
|
+
Processes are defined by `new_process` statement.
|
304
|
+
|
305
|
+
```ruby
|
306
|
+
new_process :sender1 do
|
307
|
+
delay_for time: 100
|
308
|
+
cpu do |cpu|
|
309
|
+
(10000 / cpu.performance).miliseconds
|
310
|
+
end
|
311
|
+
end
|
312
|
+
```
|
313
|
+
|
314
|
+
First parameter of the statement is the process name which must
|
315
|
+
be unique in the whole model. The block defines behavior of the process
|
316
|
+
using statements described below.
|
317
|
+
|
318
|
+
#### Delay and CPU Load
|
319
|
+
|
320
|
+
A process can do nothing for some time. This is specified with
|
321
|
+
`delay_for` statement.
|
322
|
+
|
323
|
+
```ruby
|
324
|
+
delay_for 100.seconds
|
325
|
+
```
|
326
|
+
|
327
|
+
or
|
328
|
+
|
329
|
+
```ruby
|
330
|
+
delay_for time: 100.seconds
|
331
|
+
```
|
332
|
+
|
333
|
+
Using `delay_for` causes process to stop for specified time. It
|
334
|
+
will not occupy resources, but it will not serve any incoming
|
335
|
+
event either! If you need to put a delay between recurring
|
336
|
+
events, you should use `delay` option of `register_event`
|
337
|
+
statement.
|
338
|
+
|
339
|
+
It can also load node's CPU for specified time. This is defined
|
340
|
+
with `cpu` statement. CPU load time is defined by results of block
|
341
|
+
passed to the statement. The parameter passed to the block
|
342
|
+
represents CPU to which this work is assigned. Performance of
|
343
|
+
this CPU can be checked using `cpu.performance`.
|
344
|
+
|
345
|
+
```ruby
|
346
|
+
cpu do |cpu|
|
347
|
+
(10000 / cpu.performance).miliseconds
|
348
|
+
end
|
349
|
+
```
|
350
|
+
|
351
|
+
Time values defined by `delay_for` and returned by the `cpu` block
|
352
|
+
can be random.
|
353
|
+
|
354
|
+
#### Events and Handlers
|
355
|
+
|
356
|
+
Defining process one can use `on_event` statement, to define
|
357
|
+
process behavior when an event occurs. Process can also register
|
358
|
+
event's occurence using `register_event` statement. This is
|
359
|
+
recommended method of describing processes behavior, also
|
360
|
+
recurring behaviors. The following example will repeat sending
|
361
|
+
data 10 times.
|
362
|
+
|
363
|
+
```ruby
|
364
|
+
new_process :wget do
|
365
|
+
sent = 0
|
366
|
+
on_event :send do
|
367
|
+
cpu do |cpu|
|
368
|
+
(150 / cpu.performance).miliseconds
|
369
|
+
end
|
370
|
+
sent += 1
|
371
|
+
register_event :send, delay: 5.miliseconds if sent < 10
|
372
|
+
end
|
373
|
+
|
374
|
+
register_event :send
|
375
|
+
end
|
376
|
+
```
|
377
|
+
|
378
|
+
The optional `delay:` option of the `register_event` statement
|
379
|
+
will cause the event to be registered with specified delay. Thus
|
380
|
+
the above example will register the `:send` events with 5
|
381
|
+
milisecond gaps. By default events are registered immediatelly,
|
382
|
+
without any delay.
|
383
|
+
|
384
|
+
All statements that can be used to describe processe's behavior
|
385
|
+
can also be used inside `on_event` statement. In fact the event
|
386
|
+
handlers are preferred place to describe behavior of a process.
|
387
|
+
|
388
|
+
#### Reusable functions
|
389
|
+
|
390
|
+
The model allows to define functions that can be called from blocks
|
391
|
+
defining event handlers. The functions can hol reusable code useful
|
392
|
+
in one or more then handlers. Functions are defined using `function`
|
393
|
+
statement like this:
|
394
|
+
|
395
|
+
```ruby
|
396
|
+
function :do_something do
|
397
|
+
# any statements allowed in
|
398
|
+
# event handler blocks can be
|
399
|
+
# put here
|
400
|
+
end
|
401
|
+
```
|
402
|
+
|
403
|
+
Functions take parameters defined as usually for Ruby blocks:
|
404
|
+
|
405
|
+
```ruby
|
406
|
+
function :do_something_with_params do |param1, param2|
|
407
|
+
# any statements allowed in
|
408
|
+
# event handler blocks can be
|
409
|
+
# put here
|
410
|
+
end
|
411
|
+
```
|
412
|
+
|
413
|
+
They return values like Ruby methods: either defined by `return`
|
414
|
+
statement or the last stateent in the function block.
|
415
|
+
|
416
|
+
##### Note on `def`
|
417
|
+
|
418
|
+
You don't have to understand this. You don't even have to read this
|
419
|
+
unless you insist on using Ruby's `def` defined methods instead of above
|
420
|
+
described `function` statement.
|
421
|
+
|
422
|
+
You can use Ruby `def` statement to define reusable code in the model,
|
423
|
+
but it is **not recommended** unless you know exactly what you are
|
424
|
+
doing! The `def` defined methods will be accessible from event handler
|
425
|
+
blocks, but when called, they will be evaluated in the context in which
|
426
|
+
they were defined, not in the context of the block calling the function.
|
427
|
+
Consequently, any side effects caused by the function (like
|
428
|
+
`register_event`, `log`, `stat` or anything the DSL gives you) will be
|
429
|
+
reflected in the wrong context! Using `def` defined methods is safe only
|
430
|
+
if they are strictly functional-like -- i.e. cause no side effects.
|
431
|
+
|
432
|
+
#### Reading simulation clock
|
433
|
+
|
434
|
+
I am not convinced that this is necessary, but for now it is available.
|
435
|
+
Read and understand all this section if you think you need this.
|
436
|
+
|
437
|
+
It is possible to check value of simulation clock at which given event
|
438
|
+
is being handled. This clock has the same value inside the whole block
|
439
|
+
handling the event and it equals to the time at which the event handling
|
440
|
+
started (not the time when the event occured/was registered!). The clock
|
441
|
+
can be read using:
|
442
|
+
|
443
|
+
* `event_time` method that returns clock value
|
444
|
+
|
445
|
+
Value returned by the `event_time` method does not change inside
|
446
|
+
the event handling block even if you used `delay_for` or `cpu`
|
447
|
+
statements in this block before the `event_time` method was
|
448
|
+
called. But if you call `event_time inside the `cpu` block it
|
449
|
+
will return the time at which this block is valueted i.e. time at
|
450
|
+
which the CPU processing starts.
|
451
|
+
|
452
|
+
Concluding:
|
453
|
+
|
454
|
+
* the `event_time` method can be used inside each block inside program
|
455
|
+
definition,
|
456
|
+
* inside the whole block it returns the same value -- time when the event
|
457
|
+
handling (i.e. block evaluation) started,
|
458
|
+
* if called inside a block which is inside a block... it returns time at
|
459
|
+
which the most inner event handling (block evaluation) started.
|
460
|
+
|
461
|
+
If you need to measure time between two occurrences in whatever you
|
462
|
+
simulate, just define these occurences as two separate events in
|
463
|
+
the model. Then you will be able to read the time at which
|
464
|
+
handling of each of these events started.
|
465
|
+
|
466
|
+
The truth is that you should not need this. Not ever. For instance if
|
467
|
+
you need to model timeout waiting for a response, just register a
|
468
|
+
timeout event when you send request and inside this event either mark
|
469
|
+
request as timed out or do nothing it response was receivd before.
|
470
|
+
|
471
|
+
#### Communication
|
472
|
+
|
473
|
+
##### Sending data
|
474
|
+
|
475
|
+
A process can send data to another process using its name as
|
476
|
+
destination address.
|
477
|
+
|
478
|
+
```ruby
|
479
|
+
new_process :sender1 do
|
480
|
+
send_data to: :receiver, size: 1024.bytes, type: :request, content: 'anything useful for your model'
|
481
|
+
end
|
482
|
+
```
|
483
|
+
|
484
|
+
Data will be sent to process called `:receiver`, size will be
|
485
|
+
1024 bytes, `type` and `content` of the data can be set to anything
|
486
|
+
considered useful.
|
487
|
+
|
488
|
+
##### Receiving data
|
489
|
+
|
490
|
+
Every process is capable of receiving data, but by default the
|
491
|
+
data will be dropped and a warning issued. To process received
|
492
|
+
data, process must define event handler for `:data_received`
|
493
|
+
event.
|
494
|
+
|
495
|
+
```ruby
|
496
|
+
on_event :data_received do |data|
|
497
|
+
cpu do |cpu|
|
498
|
+
(data.size / cpu.performance).miliseconds
|
499
|
+
end
|
500
|
+
end
|
501
|
+
```
|
502
|
+
|
503
|
+
The parameter passed to the event handler (`data` in the example
|
504
|
+
above) contains a Hash describing received data.
|
505
|
+
|
506
|
+
```ruby
|
507
|
+
{ src: :source_process_name,
|
508
|
+
dst: :destination_process_name,
|
509
|
+
size: data_size,
|
510
|
+
type: 'a value_given_by_sender',
|
511
|
+
content: 'a value given by sender' }
|
512
|
+
```
|
513
|
+
|
514
|
+
The complete example of wget -- sending requests and receiving
|
515
|
+
responses can look like this:
|
516
|
+
|
517
|
+
```ruby
|
518
|
+
new_process :wget do
|
519
|
+
sent = 0
|
520
|
+
on_event :send do
|
521
|
+
cpu do |cpu|
|
522
|
+
(150 / cpu.performance).miliseconds
|
523
|
+
end
|
524
|
+
send_data to: opts[:target], size: 1024.bytes, type: :request, content: sent
|
525
|
+
sent += 1
|
526
|
+
register_event :send, delay: 5.miliseconds if sent < 10
|
527
|
+
end
|
528
|
+
|
529
|
+
on_event :data_received do |data|
|
530
|
+
log "Got data #{data} in process #{process.name}"
|
531
|
+
stats event: :request_served, where: process.name
|
532
|
+
end
|
533
|
+
|
534
|
+
register_event :send
|
535
|
+
end
|
536
|
+
```
|
537
|
+
|
538
|
+
##### Logs and statistics
|
539
|
+
|
540
|
+
The `log` statement can be used in the process description, to
|
541
|
+
send a message to logs (by default to STDOUT).
|
542
|
+
|
543
|
+
The `stats` statements can be used to collect running statistics.
|
544
|
+
|
545
|
+
* `stats_start tags` marks start of an activity described by `tags`
|
546
|
+
* `stats_stop tags` marks start of an activity described by `tags`
|
547
|
+
* `stats tags` marks that a counter marked by `tags` should be
|
548
|
+
incremented
|
549
|
+
* `stats_save value, tags` saves given value and current time
|
550
|
+
|
551
|
+
The `tags` parameter allows one to group
|
552
|
+
statistics by required criteria, e.g. name of specific process
|
553
|
+
(apache1, apache2, apache2, ...) in which they were collected, action
|
554
|
+
performred by the process (begin_request, end_request) etc. The
|
555
|
+
parameter should be a hash and it is your responsibility to design
|
556
|
+
structure of these parameters to be able to conveniently collect
|
557
|
+
required events.
|
558
|
+
|
559
|
+
Simulator automatically collects statistics for resource usage
|
560
|
+
(subsequent CPUs and net segments).
|
561
|
+
|
562
|
+
##### Variables, Conditions, Loops
|
563
|
+
|
564
|
+
As shown in examples above (see model of wget), variables can be
|
565
|
+
used to steer behavior of a process. Their visibility should be
|
566
|
+
intuitive. Don't use Ruby's instance variables -- `@something`
|
567
|
+
-- didn't test it, but no guarantee given!
|
568
|
+
|
569
|
+
You can also get name of current process from `process.name`.
|
570
|
+
|
571
|
+
Using conditional statements is encouraged wherever it is useful.
|
572
|
+
Using loops is definitely NOT encouraged. It can (and most probably
|
573
|
+
will) create long event queues which will slow down simulation.
|
574
|
+
You should rather use recurring events, as in the example with
|
575
|
+
wget model.
|
576
|
+
|
577
|
+
### Programs
|
578
|
+
|
579
|
+
Programs can be used to define the same logic used in a numebr of
|
580
|
+
processes. Their names can be the used to define processes.
|
581
|
+
Behavior of programs ca be described using the same statemets
|
582
|
+
that are used to describe processes.
|
583
|
+
|
584
|
+
|
585
|
+
```ruby
|
586
|
+
program :waiter do |time|
|
587
|
+
delay_for time: time
|
588
|
+
end
|
589
|
+
|
590
|
+
program :worker do |volume|
|
591
|
+
cpu do |cpu|
|
592
|
+
( (volume * volume).in_bytes / cpu.performance ).miliseconds
|
593
|
+
end
|
594
|
+
end
|
595
|
+
```
|
596
|
+
|
597
|
+
These two programs can be used to define processes:
|
598
|
+
|
599
|
+
```ruby
|
600
|
+
new_process program: waiter, args: 100.miliseconds
|
601
|
+
new_process program: worker, args: 2000.bytes
|
602
|
+
```
|
603
|
+
|
604
|
+
`args` passed to the `new_process` statement will be passed to
|
605
|
+
the block defining program. So in the example above `time`
|
606
|
+
parameter of `:waiter` process will be set to 100 and `volume`
|
607
|
+
parameter of the `:worker` process will be set to 2000.
|
608
|
+
|
609
|
+
### Resources
|
610
|
+
|
611
|
+
Resources are described in terms of `node`s equipped with `cpu`s
|
612
|
+
of given performance and `net`s with given `name` bandwidth
|
613
|
+
(`bw`). Routes between `node`s are defined using `route`
|
614
|
+
statement.
|
615
|
+
|
616
|
+
#### Nodes
|
617
|
+
|
618
|
+
Nodes are defined using `node` statement, cpus inside nodes using
|
619
|
+
`cpu` statement with performance as parameter.
|
620
|
+
|
621
|
+
```ruby
|
622
|
+
node :laptop do
|
623
|
+
cpu 1000
|
624
|
+
cpu 1000
|
625
|
+
cpu 1000
|
626
|
+
end
|
627
|
+
```
|
628
|
+
|
629
|
+
The performance defined here, can be used in `cpu` statement in
|
630
|
+
process description.
|
631
|
+
|
632
|
+
|
633
|
+
#### Nets
|
634
|
+
|
635
|
+
Nets used in communication are defined with `net` statement with
|
636
|
+
name as parameter and a Hash definind other parameters of the
|
637
|
+
segment. The most important parameter of each net segmetn is its
|
638
|
+
bandwidth:
|
639
|
+
|
640
|
+
```ruby
|
641
|
+
net :lan, bw: 1024.bps
|
642
|
+
net :subnet1, bw: 20480.bps
|
643
|
+
```
|
644
|
+
|
645
|
+
Additionally it is possible to specify probability that a packet
|
646
|
+
transmitted over this network will be dropped. Currently, each message
|
647
|
+
sent between two processes is treated as a single packet, so this
|
648
|
+
probability will apply to dropping the whole message -- the message will
|
649
|
+
be sent, but nothing will be received. By default drop probability is
|
650
|
+
set to 0 and all sent messages are delivered.
|
651
|
+
|
652
|
+
There are two ways to define probability drop probability. First,
|
653
|
+
specify a `Float` number between 0 and 1. The packets will be dropped
|
654
|
+
with this this probability and uniform distribution:
|
655
|
+
|
656
|
+
```ruby
|
657
|
+
net :lan, bw: 1024.bps, drop: 0.01
|
658
|
+
```
|
659
|
+
|
660
|
+
Second, it is possible to define a block of code. That block will be
|
661
|
+
evaluated for each packet transmitted over this network and should
|
662
|
+
return true if packet should be dropped and false otherwise. This block
|
663
|
+
can use Ruby's `rand` function and any desired logic to produce require
|
664
|
+
distribution of dropped packets. Fo example, to drop packets according
|
665
|
+
to exponential distribution with lambda = 2:
|
666
|
+
|
667
|
+
```ruby
|
668
|
+
net :lan, bw: 1024.bps, drop: ->{ -0.5*Math.log(rand) < 0.1 }
|
669
|
+
```
|
670
|
+
|
671
|
+
Currently, it is not possible to make probability of dropping a packet
|
672
|
+
dependent on dropping previous packets.
|
673
|
+
|
674
|
+
#### Routes
|
675
|
+
|
676
|
+
Routes are used to define which `net` segments should be traversed
|
677
|
+
by data transmitted between two given `node`s. Routes can be
|
678
|
+
one-way (default) or two-way.
|
679
|
+
|
680
|
+
```ruby
|
681
|
+
route from: :laptop, to: :node02, via: [ :net01, :net02 ]
|
682
|
+
route from: :node04, to: :node05, via: [ :net07, :net01 ], twoway: true
|
683
|
+
route from: :node06, to: :node07, via: [ :net07, :net01 ], twoway: :true
|
684
|
+
```
|
685
|
+
|
686
|
+
Communication between processes located on different nodes
|
687
|
+
requires a route defined between the nodes. If there is more then
|
688
|
+
one route between a pair of nodes, random one is selected. A
|
689
|
+
node can communicate with itself without any route defined and
|
690
|
+
without traversing any `net` segemtns
|
691
|
+
|
692
|
+
### Mapping of Application to Resources
|
693
|
+
|
694
|
+
When application is defined as a set of processes and resources
|
695
|
+
are defined as nodes connected with net segments, the application
|
696
|
+
can be mapped to the nodes. Mapping is defined using `put`
|
697
|
+
statement.
|
698
|
+
|
699
|
+
```ruby
|
700
|
+
put :wget, on: :laptop
|
701
|
+
put :server1, on: :gandalf
|
702
|
+
```
|
703
|
+
|
704
|
+
First parameter is process name, second (after `on:`) is node
|
705
|
+
name.
|
706
|
+
|
707
|
+
Thus application logic and topology does not depent in any way on
|
708
|
+
topology of resources. The same application can be mapped to
|
709
|
+
different resources in different ways. The same resource set can
|
710
|
+
be used for different applications.
|
711
|
+
|
712
|
+
### Units
|
713
|
+
|
714
|
+
Simulator operates on three kinds of values:
|
715
|
+
|
716
|
+
* data volume
|
717
|
+
* network bandwidth
|
718
|
+
* time
|
719
|
+
|
720
|
+
For each value one can use specific measurement units:
|
721
|
+
|
722
|
+
* for data volume
|
723
|
+
* bits
|
724
|
+
* bytes
|
725
|
+
* for network bandwidth
|
726
|
+
* bps (bits per second)
|
727
|
+
* Bps (bytes per second)
|
728
|
+
* for time:
|
729
|
+
* microseconds
|
730
|
+
* miliseconds
|
731
|
+
* seconds
|
732
|
+
* minutes
|
733
|
+
* hours
|
734
|
+
* days (24 hours)
|
735
|
+
|
736
|
+
In every place where data volume should be given, it can be
|
737
|
+
defined using expressions like
|
738
|
+
|
739
|
+
```ruby
|
740
|
+
1024.bytes
|
741
|
+
128.bits
|
742
|
+
```
|
743
|
+
|
744
|
+
Similarly network bandwidth can be defined using
|
745
|
+
|
746
|
+
```ruby
|
747
|
+
128.Bps
|
748
|
+
1024.bps
|
749
|
+
```
|
750
|
+
|
751
|
+
Finally, if time should be given, one should use a unit, to
|
752
|
+
ensure correct value, e.g.
|
753
|
+
|
754
|
+
```ruby
|
755
|
+
10.seconds
|
756
|
+
100.microseconds
|
757
|
+
2.hours
|
758
|
+
```
|
759
|
+
|
760
|
+
For values returned from simulator, to ensure value in correct
|
761
|
+
units use `in_*` methods, e.g.
|
762
|
+
|
763
|
+
```ruby
|
764
|
+
data.in_bytes
|
765
|
+
time.in_seconds
|
766
|
+
```
|
767
|
+
|
768
|
+
So to define that CPU load time in milliseconds should be equal to
|
769
|
+
10 * data volume in bytes, use:
|
770
|
+
|
771
|
+
```ruby
|
772
|
+
cpu do |cpu|
|
773
|
+
(data.size.in_bytes * 10).miliseconds
|
774
|
+
end
|
775
|
+
```
|
776
|
+
|
777
|
+
Every measurement unit has its equivalent `in_*` method.
|
778
|
+
|
779
|
+
## Using simulation statistics
|
780
|
+
|
781
|
+
The way of obtaining statistics depends on the method used to create
|
782
|
+
simulation. The most convenient is to use subclass of the
|
783
|
+
`RBSim::Experiment` class, but it can also be done with the simulation
|
784
|
+
performed with `RBSim` class.
|
785
|
+
|
786
|
+
### Statistics with subclass of `RBSim::Experiment`
|
787
|
+
|
788
|
+
The `RBSim::Experiment` class provides convenient method to obtain
|
789
|
+
simulation statistics and its subclass created to run simulation is a
|
790
|
+
natural place to implement own methods that process the statistics into
|
791
|
+
required collective results.
|
792
|
+
|
793
|
+
There are two methods available for every instance method in a subclass
|
794
|
+
of the `RBSim::Experiment` class. The `app_stats` gives access to
|
795
|
+
statistics concerning modeled application, and the `res_stats`
|
796
|
+
contains statistics concerning used resources, the ones automatically
|
797
|
+
collected by the simulator.
|
798
|
+
|
799
|
+
For application stats as well as for resource stats the actual data is
|
800
|
+
available with three iterators, for three types of
|
801
|
+
collected statistics:
|
802
|
+
|
803
|
+
* data from basic counters collected using `stats`
|
804
|
+
statement in the model can be obtained with `counters` iterator
|
805
|
+
* data from duration statistics collected with `stats_start` and
|
806
|
+
`stats_stop` statements can be obtained with `duration` iterator
|
807
|
+
* data from value stats saved with `stats_save` are available via
|
808
|
+
`values` iterator.
|
809
|
+
|
810
|
+
Each iterator accepts optional parameter describing filters, so it is
|
811
|
+
possible to limit amount of data that should be processed. The filters
|
812
|
+
allow to select required values on the basis of parameters passed to the
|
813
|
+
`stats_*` statements in the model. For example, in order to get
|
814
|
+
durations of `operation` called `:update` on can use the following
|
815
|
+
snippet:
|
816
|
+
|
817
|
+
```ruby
|
818
|
+
app_stats.durations(operation: :update)
|
819
|
+
```
|
820
|
+
|
821
|
+
assuming that the data are collected in the model with statements
|
822
|
+
|
823
|
+
```ruby
|
824
|
+
stats_start operation: :update
|
825
|
+
```
|
826
|
+
|
827
|
+
and
|
828
|
+
|
829
|
+
```ruby
|
830
|
+
stats_stop operation: :update
|
831
|
+
```
|
832
|
+
|
833
|
+
Depending on the type of collected data (counters, durations, values)
|
834
|
+
different data are passed by the iterator.
|
835
|
+
|
836
|
+
#### Values of counters
|
837
|
+
|
838
|
+
Counters are grouped according to parameters (tags) passed to the
|
839
|
+
`stats` statement in the model. Every time such statement is reached
|
840
|
+
during simulation, current values of the simulation clock is saved. The
|
841
|
+
`counters` iterator yields two arguments: `tags` and array of timestamps
|
842
|
+
when counters with these tags was triggered.
|
843
|
+
|
844
|
+
```ruby
|
845
|
+
app_stats.counters(event: :finished) do |tags, timestamps|
|
846
|
+
# do something with :finished events
|
847
|
+
# if you need you can check the other
|
848
|
+
# tags saved with these events
|
849
|
+
end
|
850
|
+
```
|
851
|
+
|
852
|
+
#### Duration times
|
853
|
+
|
854
|
+
Duration times are saved from the model using `stats_start` and
|
855
|
+
`stats_stop` statements. They are also grouped according to the tags
|
856
|
+
passed to these statements. Data can be subsequently filtered using
|
857
|
+
these tags and the `durations` iterator yields three values: `tags`,
|
858
|
+
`start_time`, `stop_time`, where start and stop times are timestamps at
|
859
|
+
which simulation reached corresponding statement.
|
860
|
+
|
861
|
+
For example if an operation in a model is braced with statement:
|
862
|
+
|
863
|
+
```ruby
|
864
|
+
stats_start operation: :update
|
865
|
+
```
|
866
|
+
|
867
|
+
and
|
868
|
+
|
869
|
+
```ruby
|
870
|
+
stats_stop operation: :update
|
871
|
+
```
|
872
|
+
|
873
|
+
duration of this operation can be obtained using the following
|
874
|
+
statement:
|
875
|
+
|
876
|
+
```ruby
|
877
|
+
app_stats.durations(operation: :update) do |tags, start, stop|
|
878
|
+
# do something with the single duration
|
879
|
+
# you can use the values of the tags
|
880
|
+
end
|
881
|
+
```
|
882
|
+
|
883
|
+
If there were more then one event with the same tags save, the block
|
884
|
+
will be yielded for each of them. Similarly, if there were additional
|
885
|
+
tags set for the `stats_*` statements, the block will be yielded for
|
886
|
+
each of them.
|
887
|
+
|
888
|
+
#### Values
|
889
|
+
|
890
|
+
Values (e.g. queue length) can be saved while simulation using
|
891
|
+
`stats_save` statement with require tags. The save values are grouped
|
892
|
+
using tags and time at which they were saved. The `values` iterator
|
893
|
+
yields three parameters: `tags`, `timestamp`, `values` where values is a
|
894
|
+
list of values saved for the same tags and the same time.
|
895
|
+
|
896
|
+
#### Statistics of resources
|
897
|
+
|
898
|
+
Statistics of resources are automatically collected by simulator in a
|
899
|
+
predefined way. They are available in the `RBSim::Experiment` subclass
|
900
|
+
with the `res_stats` method. They are grouped by the predefined tags
|
901
|
+
that allow to identify resource that generated specific reading and type
|
902
|
+
of the value.
|
903
|
+
|
904
|
+
* CPU usage can be obtained using duration events tagged with `resource:
|
905
|
+
'CPU'` tag, and also with `node:` with name of the node the CPU
|
906
|
+
belongs to,
|
907
|
+
* network usage for subsequent net segments can be obtained with
|
908
|
+
duration iterator using network `resource: 'NET'` tag; these values
|
909
|
+
are also tagged with and `name:` tag corresponding to the name of the
|
910
|
+
network segment,
|
911
|
+
* number of packages dropped by a network segment is available via
|
912
|
+
`counter` iterator using tags `event: 'NET DROP'` with additional
|
913
|
+
`net:` tag corresponding to the name of the segment,
|
914
|
+
* waiting time for data packages that were transmitted over the network,
|
915
|
+
but not yet handled by a busy processes are tagged with: `resource
|
916
|
+
'DATAQ WAIT` and `process:` tag corresponding to name of the receiving
|
917
|
+
process.
|
918
|
+
* length of the queue that holds data that were transmitted over the
|
919
|
+
network but not yet received by a busy process can be read with
|
920
|
+
`values` counter using `resource: 'DATAQ LEN'` tag with additional tag
|
921
|
+
`process:` corresponding to the process name.
|
922
|
+
|
923
|
+
If for a more complicated model of resources there is a need to
|
924
|
+
additionally group the resources, it is possible to put additional tags
|
925
|
+
to
|
926
|
+
|
927
|
+
* nets
|
928
|
+
* cpus
|
929
|
+
* processes
|
930
|
+
|
931
|
+
and these tags will be saved together with statistics corresponding to
|
932
|
+
these processes. So it is e.g. possible to create a group of processes
|
933
|
+
of the same type:
|
934
|
+
|
935
|
+
```ruby
|
936
|
+
new_process :apache1, program: :webserver, tags { type: :apache }
|
937
|
+
new_process :apache2, program: :webserver, tags { type: :apache }
|
938
|
+
```
|
939
|
+
|
940
|
+
and then obtain queue lengths for all of them with:
|
941
|
+
|
942
|
+
```ruby
|
943
|
+
res_stats.values(resource: 'DATAQ LEN', type: :apache)
|
944
|
+
```
|
945
|
+
|
946
|
+
### Statistics when using `RBSim.model`
|
947
|
+
|
948
|
+
The only difference from `RBSim::Experiment` is the way to obtain
|
949
|
+
objects holding the actual statistics. When model was created using
|
950
|
+
`RBSim.model` or `RBSim.read` and saved in a `model` variable, the
|
951
|
+
statistics can be obtained using `model.stats` method. The method
|
952
|
+
returns hash with two elements. Under `:app_stats` key there is the same
|
953
|
+
object that is available in a subclass of `RBSim::Experiment` using
|
954
|
+
method `app_stats`. Similarly there is `:res_stats` key holding the
|
955
|
+
the resource statistics.
|
956
|
+
|
957
|
+
|
958
|
+
# Copyright
|
959
|
+
|
960
|
+
Copyright (c) 2014-2018 Wojciech Rząsa. See LICENSE.txt for further details. Contact me if interested in different license conditions.
|