Handoff Counters Simulator

Simulates distributed executions of Handoff Counters - eventually consistent distributed counters, as presented in http://arxiv.org/abs/1307.3207

Usage

Have Rust installed. Then, clone repo:

git clone git@github.com:pssalmeida/handoff_counter_simulator-rs.git

Change to created dir and build project:

cd handoff_counter_simulator-rs
cargo build --release

There are 4 executables (cn1, cn2, cn3, cn4), which can be invoked as, e.g.,

./target/release/cn1 10 100 1000 10 10000 10 1000000 1000

Or better, go to the runs dir and use sim.py, which takes as arguments the name of one of these executables and its arguments, and redirects the output to a suitably named file. E.g., invoking:

./sim.py cn1 10 100 1000 10 10000 10 1000000 1000

will execute cn1 as above, redirecting the output to file cn1-10-100-1000-10-10000-10-1000000-1000.

In the runs dir there are already several files corresponding to some runs, as well as some other Python scripts to produce plots (through pandas and Matplotlib).

Scenarios

All 4 executables simulate scenarios with three tiers, with a fixed number of tier 0 nodes and tier 1 nodes, and a varying number of tier 2 nodes along the run. We call clients to tier 2 nodes and servers to tier 1 nodes.

• cn1 where clients keep affinity to a randomly choosen server;

• cn2 where clients disconnect abruptly and choose a different random server when reconnecting, exchanging messages only with this new server;

• cn3 as cn2 but now with more intelligent message exchange, to discard unwanted slots;

• cn4 with client retirement, where clients both arrive and retire at a given rate, while keeping a given number of active clients.

These executables take from 8 to 10 arguments:

cn1 t0 t1 t2 arrival_period activity_period active_percentage end_time stat_interval

cn2 t0 t1 t2 arrival_period activity_period active_percentage end_time stat_interval

cn3 t0 t1 t2 arrival_period activity_period active_percentage handler_period end_time stat_interval

cn4 t0 t1 t2 arrival_period activity_period active_percentage handler_period partition_probability end_time stat_interval

These arguments determine some simulation parameters:

• t0: number of tier 0 nodes;
• t1: number of tier 1 nodes;
• t2: starting number of tier 2 nodes;
• arrival_period: one new tier 2 node arrives each arrival_period miliseconds;
• activity_period: time for each period of client behavior;
• active_percentage: percentage of the activity_period that a client is active;
• handler_period: period of the invocations of the periodic handlers at both clients and servers, that send messages to appropriate peers;
• partition_probability: probability that a client is partitioned when it decides to retire;
• end_time: time at which the simulation ends;
• stat_interval: statistics are gathered each stat_interval miliseconds;