Skip to content

HTTPS clone URL

Subversion checkout URL

You can clone with
or
.
Download ZIP
Fetching contributors…

Cannot retrieve contributors at this time

116 lines (72 sloc) 3.442 kB
OTP Supervision tree:
====================
The application is now split in two (see tsunami-inside.png for an overview):
** a single controller (tsunami_controller)
* ts_config_server (gen_server). Configuration server. Session's
definitions are kept by the config server.
* ts_mon (gen_server). Each client send reports of stats to this
server. Several types of messages are handled by ts_mon.
* ts_os_mon (gen_server). Use to monitor remote node activity (cpu,
memory, network traffic). Currently, use an erlang agent on remote nodes.
* ts_timer (used by ts_client when ack is global) (gen_fsm)
servers used to construct messages:
* ts_msg_server (gen_server)
* ts_user_server (gen_server) used by jabber_* for unicity of users id
** several clients (tsunami) Several nodes can be used simultaneously
This application is simpler:
* ts_launcher (gen_fsm) launch simulated users.
* ts_session_cache (gen_server) cache the sessions's definition (ask
the config_server if it's not yet in the cache)
* 1 process per simultated client (ts_client), under
the supervision of ts_clients_sup ( using simple_one_for_one )
Main modules:
============
1/ ts_launcher. the master process that spawns other simulated clients:
1.1/ client processes: at each simulated client correspond 1 erlang
process (ts_client)
1.2/ monitoring process (ts_mon)
2/ statistical module (ts_stats)
3/ protocol-specific modules (ts_jabber and ts_http, for
example).
IDX-TSUNAMI use different types of acknoledgements to determine
when a the response of a request is over. For each requests, 4
options are possible:
* parse -> the receiving process parse the response from the
server dans can warn the sending process when the response is
finish (function parse/2). This is used for HTTP.
* no_ack: as soon as the request has been sent, the next one is
executed (it can be a thinktime)
* local: the request is acknoledge once a packet is received
* global: the request is acknoledge once all clients has
received an acknoledgement. This has been introduced for
Jabber: with that, you can set that users starts talking when
everyone is connected.
How to add a new protocol, or extend an existing one:
====================================================
To add a new protocol, you have to create a module that implement and
exports:
-export([init_dynparams/0,
add_dynparams/4,
get_message/1,
session_defaults/0,
parse/2,
parse_config/2,
new_session/0]).
References:
==========
- Erlang
http://www.erlang.org/
Design principles:
http://www.erlang.org/doc/r7b/doc/design_principles/part_frame.html
- Jabber
http://docs.jabber.org/general/html/protocol.html
- Stochastics models:
For more details on stochastics models and application to Web
workload generators, have a look at:
Nicolas Niclausse. Modélisation, analyse de performance et
dimensionnement du World Wide Web. Thèse de Doctorat (PhD), Université
de Nice - Sophia Antipolis, Juin 1999.
http://www-sop.inria.fr/mistral/personnel/Nicolas.Niclausse/these.html
Z. Liu, N. Niclausse, C. Jalpa-Villanueva & S. Barbier. Traffic
Model and Performance Evaluation of Web Servers Rapport de
recherche INRIA, RR-3840 (http://www.inria.fr/rrrt/rr-3840.html)
Jump to Line
Something went wrong with that request. Please try again.