Workload Generation for PubSub System

1
Workload Generation for Pub/Sub System

• Yanyan Wang
• 11/25/2002

2
Agenda

• Introduction
• Related Works
• Approach
• Experiences
• Conclusion Future Works

3
Motivation

• Publish-Subscribe Infrastructure

4
A Benchmark Suite

• A benchmark suite composed of three sections
• Interface suitability
• Applications
• Synthetic Scenarios
• Role of a synthetic workload generator
• Generate inputs to system/simulator

5
Agenda

• Introduction
• Related Works
• Approach
• Experiences
• Conclusion Future Works

6
Two Approaches

• Trace-based approach
• Starts with an empirical trace
• Subsamples or permutes the ordering of the
  requests to generate a new workload different in
  some respect from the original.
• Analytic approach
• Uses mathematical models for the workload
  characteristics of interest
• Uses random number generation to produce
  workloads that statistically conform to these
  models.

7
Analytic approach

• Two types
• Resource-oriented modeling approach
• Capture the characteristic of the workload
  itself.
• User behavior modeling approach
• Capture the characteristic of the user behavior
• In general be hierarchical
• The sequence of user interactions at a higher
  level
• Result in a stream of requests at a lower level

8
Resource-oriented modeling approach

• ProWGen (Proxy Workload Generator)
• Analytic approach
• Parameters provide control over five key workload
  characteristics
• Resource-oriented modeling approach does not
  model individual client behaviors, rather, models
  the aggregate workload as generated from many
  clients

9
User behavior modeling approach

• SURGE (Scalable URL Reference Generator)
• UE (User Equivalent) a single ON/OFF process
• Probability distributions are used for each UE
• BISANTE (Broadband Integrated Satellite Network
  Traffic Evaluations)
• User Profiles a hierarchy of independent
  stochastic processes modeled by FSM to represent
  user behavior.
• S-client
• A single process and select system call to
  manage a large number of concurrent active
  connections to the server
• Waspclient
• Adapted from SURGE, excluding the user think time

10
User behavior modeling approach(Cont.)

• SynRGen (Synthetic file Reference Generator)
• User-Behavior modeling approach
• Volume a subtree of files and directories
  exhibiting a unique combination of physical
  characteristics.
• Preprocessed into a C data structure accessed by
  users
• User classes a stochastic finite state machine
• Configuration files describing user behavior
• Preprocessed into a C program representing a
  synthetic user

11
Agenda

• Introduction
• Related Works
• Approach
• Experiences
• Conclusion Future Works

12
Our Approach

• Two Parts
• Topology
• Internet Topology Generator from GIT (includes
  mapping from servers to sites)
• Map clients to sites
• Application behavior
• A process for each client user behavior modeling
  approach
• Creating workloads by simulating clients using a
  generic discrete-event sequential simulator (a
  client can be understood as a program.)
• program complex behaviors
• program inter-related behaviors among clients
• Parameters defined in configuration file to
  describe their behavior

13
Topology
Site
Client
Server
Client/Server
14
Topology File

• C0_at_s77 -gt S_at_s79
• C0_at_s77 -gt S_at_s82
• C1_at_s29 -gt S_at_s0
• C1_at_s29 -gt S_at_s30
• C2_at_s47 -gt S_at_s43
• C2_at_s47 -gt S_at_s45
• C3_at_s15 -gt S_at_s11
• C3_at_s15 -gt S_at_s12
• C4_at_s80 -gt S_at_s2
• C4_at_s80 -gt S_at_s84
• C4_at_s80 -gt S_at_s85
• C5_at_s42 -gt S_at_s43
• C5_at_s42 -gt S_at_s45
• C6_at_s4 -gt S_at_s0
• C6_at_s4 -gt S_at_s7
• C7_at_s59 -gt S_at_s60
• C7_at_s59 -gt S_at_s63

• S_at_s0
• S_at_s2 -gt S_at_s0
• S_at_s3 -gt S_at_s0
• S_at_s4 -gt S_at_s0
• S_at_s12 -gt S_at_s0
• S_at_s17 -gt S_at_s0
• S_at_s29 -gt S_at_s0
• S_at_s69 -gt S_at_s2
• S_at_s80 -gt S_at_s2
• S_at_s88 -gt S_at_s2
• S_at_s1 -gt S_at_s3
• S_at_s95 -gt S_at_s3
• S_at_s5 -gt S_at_s4
• S_at_s7 -gt S_at_s4
• S_at_s9 -gt S_at_s12
• S_at_s10 -gt S_at_s12
• S_at_s11 -gt S_at_s12

15
Simulator

• Class Sim a generic discrete-event sequential
  simulator, maintains a time-ordered schedule of
  discrete events.
• create_process(Process, char mode 0)
• run_simulation(), stop_simulation()
• set_timeout(Time)
• signal_event(Event, ProcessId, Time)
• Virtual class Process representing processes
  running within the simulator.
• process_event(const Event msg)
• process_timeout()
• stop()

16
Simulation
create_process()
create_process()
create_process()
run_simulation()
set_timeout(T)
process_timeout() subscribe signal_event() set_
timeout(T)
process_event() subscribe signal_event()
process_event() subscribe
process_timeout() publish stop()
stop_simulation()
17
Contents of Sub/Pub

• Sub Pattern Type Attr_Name Operator Value
• Pub Pattern Type Attr_Name Value
• Every part comes from weighted dictionaries
  (distribution)
• Type from types.dist
• Attr_Name from attr_names.dist
• Operator from string_operators.dist for string,
  from operators.dist for other types
• Value from int_values.dist for int, from
  str_values.dist for string

18
Parameters

• Includes dictionary files as parameters.
• Different parameters for different kinds of
  activities. E.g
• For non-interactive activities
• Type Random, Constant or Poisson
• For Random Min, Max
• For Constant Value
• For Poisson Mid
• For triggered activities modeled through
  simulator

19
Parameters (Content)
20
Parameters (Behavior)
21
Workload Generated

• T3, C0, S0
• s optics SF belch
• i relocated 29
• s panelist lt dismissing
• i Massachusetts 85
• i lilacs 80
• T4, C2, S0
• s absolution adiabatic
• i signify gt 21
• i hourly 89
• T5, C0, U0
• T5, C1, S0
• s opposites gt epistemological
• T5, C2, U0
• T6, C7, S0
• s pooling SF lender

T6, C0, N0 i hourly 22 i permanently 56 i
Bontempo 7 T6, C6, N0 s Muslims smuggle i
influenced 55 T7, C0, N1 i roping 6 i syllogisms
47 T7, C4, N0 s archdiocese belch b greengrocer
0 s wheels sates T9, C7, U0 T10, C1, U0
22
Agenda

• Introduction
• Related Works
• Approach
• Experiences
• Conclusion Future Works

23
Formulate implement scenarios (Wave Scenario)
Film
4
2
Film News
2
3
2
Film News shopping
1
5
Film
shopping
5
2
3
News shopping
4
News shopping
6
2
shopping
7
24
Configuration File(1)

• SUBJECT "News"
• CONSTR_MIN 1
• CONSTR_MAX 3
• ATTR_MIN 2
• ATTR_MAX 6
• ATTR_DICT_F "attr_names.dict
• TYPES_DIST_F "types.dist"
• OP_DIST_F "operators.dist"
• OP_STRING_DIST_F "string_operators.dist"
• STR_DIST_F "str_values.dist"
• INT_DIST_F "int_values.dist"
• BOOL_DIST_F "bool_values.dist"

25
Configuration File(2)

• CLIENT1
• SUB_FRIENDS 20_at_2Film"News",
  30_at_5"News""Shopping"
• SUBJECT_DICT_F "subject_names.dict"
• SUBSCRIPTION 7
• NOTIFICATION 3
• FIRST 0
• SUB_SERIAL_NUM 2
• NOTIFY_SERIAL_NUM 1
• SUB_NOTIFY_TIME 1
• SUB_TIME 0
• NOTIFY_TIME 0
• UNSUB_TIME 2
• 20_at_2Film News means client 2 will be
  triggered to subscribe on the subject Film and
  News after 2 seconds.

26
Simulation
subscribe
publish
Be triggered to sub/pub
signal_event
Time (s)
Client 1
Client 2
Client 3
Client 4
Client 5
Client 6
27
Application to Siena
Generate topology
Client 1
Script 1
Topology file
Server 1
Script Generator
Script Dispatcher
Script 2
Client 2
Script 3
Workload file
scripts for each clients
Server 2
Client 3
28
Agenda

• Introduction
• Related Works
• Approach
• Experiences
• Conclusion Future Works

29
Conclusion

• User behavior modeling approach
• Two Parts
• Topology generated by Internet Topology
  generator
• Application Behavior
• Every client is a program
• Regular non-interrelated activity through
  analytic approach
• Complex and interrelated activity through
  simulating
• Defined by configuration file

30
Future Works

• More reasonable workload content
• Experiment on remote servers
• Improvement of configuration file
• Application to other systems
• Validation with data from real life

31
Questions?

• Happy ThanksGiving! ?