parity

Manual deployment of Parity on a cluster

Generate config files

First, we need to generate a list of config files for launching the nodes.

  `python gen_config.py <#nodes>` 

will generate the following files:

• chain_spec.json: specification of the genesis chain, which include list of validators, etc.
• config.tom.[1..<#nodes>]: multiple config files each contain network parameter for the node, such as who are the signer, mining parameters, etc.

Other important files

• keys: directory contains all the keys for the accounts
• node.pwds: password strings

For more details, see https://github.com/ethcore/parity/wiki/Proof-of-Authority-Chains

Launch Parity server

  `./start_parity.sh <data dir> <host/IP> <log dir> <parity binary> <host index>`

where:

• <data dir>: directory where chain data + keys are store. Make sure it is non-NFS
• <host/IP: IP address of the host. The RPC end point will be <hosts/IP>:8545
• <parity binary>: where the Parity executable is
• <host index>: from 1 to <#nodes>

Add peers to the network

This step is needed to get the Parity nodes connected to each other. There are several steps needed (best to use the script, described later):

1. Send a query to each running node, asking for its enode, using parity_enode method
2. Gather all enode values, send each value to every node, using parity_addReservedPeer method

At the end, the nodes should learn about each other's presence.

Launch a client:

Either YCSB or Smallbank comes with a driver for Parity, and the launch command is the same as in Ethereum.

  `.../driver -db parity -threads <#threads> -P <workload> -endpoint <host:port> -txrate <rate>`

---

Deployment using a script

The steps above are tedious and error prone. The following scripts help automate the whole thing

Config file

The file config.py is very important, as it defines global values, namely:

• NS: number of nodes (servers)
• NODES, CLIENTS: list of IP addresses of the servers and of the clients
• THREADS, RATES, [EXPS]: number of threads and tx rate. These 2 can be replaced by a list of tuples representing (#threads, rate) to be run.
• HOME_DIR: the current directory containing this scripts and other
• DATA_DIR, LOG_DIR: data and log directory for the servers
• PARITY_EXE: where the Parity executable is
• CLIENT_LOG: directory where the client's logs go
• enode_command, add_peer, partition_cmd, kill_cmd: string formatted OS commands
• TIMEOUT: applicable to the security benchmark

Run scripts

The file run.py drives all experiments.

  `python run.py <start/kill> [-security/-fixed/-drop/-smallbank/-donothing]`

It consists of the following important functions:

• start_parity(.): ssh to every node in NODES and execute the start_parity.sh command with the appropriate parameters.

• get_enodes(.): send parity_enode RPC request to every node, return a list of enode:// strings

• add_peers(.): send parity_addReservedPeer request to every node, for every enode:// string in the list

• start_clients(.): loop through the client-server tuple and invoke start_xxx_client.sh with the appropriate workload (default is YCSB) and parameters.

• partition(.): imported from partition.py, which partitions NODES in half for a given duration.

• run_exps(.): main driver, which start servers and client. For the security benchmark, it then waits for 100 seconds before invoking partitioning. Finally, it waits for a fixed duration (500 seconds in total) before killing clients and server processes.

---

Run Parity benchmarks

Each benchmark has it own configuration file:

• config_[ycsb/smallbank/donothing].py: for [YCSB/Smallbank/DoNothing] benchmark, fixed number of clients and server
• config_saturation.py: for the saturation benchmark, using YCSB
• config_security.py: for the security benchmark, using YCSB
• config_scale.py: for the scale benchmark with the same number of servers and clients, using YCSB
• config_scale_fix_8.py: for the scale benchmark with fixed number of clients (8), using YCSB
• config_drop.py: for the failure benchmark, using YCSB

It is important to change settings in these files before running the corresponding benchmarks.

Run all benchmarks in one go

Having set the config files, we can run all benchmarks in one go:

  `python run_all.py`

Run individual benchmark

To run benchmark XXX, there are 2 steps:

1. Copy the benchmark's config file to config.py. That is: cp config_XXX.py config.py

2. Start the benchmark: python run.py start [opts]

Without [opts], this will execute YCSB workload for whatever specified in the config.py file (by default, it is running with fixed/8 servers and clients).