J ava M odelling T ools Marco Bertoli, Giuliano Casale, Giuseppe - - PowerPoint PPT Presentation
Politecnico di Milano EECS Dept. Milan, Italy User-Friendly Approach to Capacity Planning studies with J ava M odelling T ools Marco Bertoli, Giuliano Casale, Giuseppe Serazzi 1 SIMUTOOLS09 March 5th, 2009 outline the JMT suite of
SIMUTOOLS09 March 5th, 2009
1
Politecnico di Milano EECS Dept. Milan, Italy
Marco Bertoli, Giuliano Casale, Giuseppe Serazzi
SIMUTOOLS09 March 5th, 2009
2
the JMT suite of tools the JSIM simulator Case Study: optimal admission control
SIMUTOOLS09 March 5th, 2009
3
SIMUTOOLS09 March 5th, 2009
4
JMT Tools JMT Tools JSIMwiz JSIMwiz JSIMgraph JSIMgraph
XSLT XSLT
“Model-View-Controller”-like pattern
Better reuse and isolation of components
SIMUTOOLS09 March 5th, 2009
5
SIMUTOOLS09 March 5th, 2009
6
discrete-event simulator for queueing networks several distributions (exp, Erlang, Pareto, burst/MMPP2, …) support for NPF features:
general arrival and service processes Fork-Join centers blocking and finite capacity regions priority Classes state-dependent routing:
route to least utilized center, to shortest queue route to the center with shortest response time fastest service time, round robin, random
Logger component (debugging, processing of transient data, ...)
SIMUTOOLS09 March 5th, 2009
7
Fork and Join components
fork node: jobs are forked into P tasks Synchronization at the join node
a group of queues can be
non-admitted jobs can be
either in a FCFS waiting buffer or dropped
SIMUTOOLS09 March 5th, 2009
8
Automatic removal of the initial bias
R-5 Heuristic MSER-5 Rule (Marginal Standard
C.I. generation using spectral methods
Spectral Analysis [Heidelberger &
Used also for run-length control
SIMUTOOLS09 March 5th, 2009
9
Exponential insufficient for many models Pareto, Hyperexponential, Erlang, Gamma, burst
Custom distribution (external text file, from log,
Random number generation Mersenne Twister Load-dependent service process Server speed variable with the current queue-
Building block for Hierarchical Modeling
SIMUTOOLS09 March 5th, 2009
10
automatic maximum relative error control [Pawlikowski 1990]
ratio half-width marginal CI / estimated mean
automatic removal of the initial bias (transient filtering) max n. of samples (long run analysis) and simulation time CI generation using spectral methods
SIMUTOOLS09 March 5th, 2009
11
simulations control parameters
SIMUTOOLS09 March 5th, 2009
12
Solve open/closed/mixed BCMP queueing nets
Native support for what-if analyses Integrated with JSIMgraph (reuse models)
SIMUTOOLS09 March 5th, 2009
13
SIMUTOOLS09 March 5th, 2009
14
Finite Capacity Region with constant population of requests
13.95 86.86 Database Server service demand
DDB [ms]
55.18 69.15 Storage Server service demand
DSS [ms]
68.07 28.48 Front Server service demand
DFS [ms]
Web Service WS2 Web service WS1 Parameters
bottlenecks
SIMUTOOLS09 March 5th, 2009
15
FC Region
SIMUTOOLS09 March 5th, 2009
16
SIMUTOOLS09 March 5th, 2009
17
com m on saturation sectors 0 2 0 8
SIMUTOOLS09 March 5th, 2009
18
potential bottlenecks
SIMUTOOLS09 March 5th, 2009
19
I II III IV V 12 13 14 15 16 17 18 Experiment throughput [req/s]
(0.20, 0.80) (0.50, 0.50) (0.80, 0.20) (0.05, 0.95) (0.95, 0.05)
m axim um throughput - optim al m ix
SIMUTOOLS09 March 5th, 2009
20
the project
> 11000 downloads since April 2006