Skip to content

bft-smart/library

Repository files navigation

Byzantine Fault-Tolerant (BFT) State Machine Replication (SMaRt) v2.0

This is a Byzantine fault-tolerant state machine replication project named BFT-SMaRt, a Java open source library maintained by the LASIGE Computer Science and Engineering Research Centre at the University of Lisbon.

This package contains the source code (src/), dependencies (lib/), documentation (docs/), running scripts (runscripts/), and configuration files (config/) for version 2.0 of the project.

Quick start

To run any demonstration you first need to configure BFT-SMaRt to define the protocol behavior and the location of each replica.

The servers must be specified in the configuration file (see config/hosts.config):

#server id, address and port (the ids from 0 to n-1 are the service replicas) 
0 127.0.0.1 11000 11001
1 127.0.0.1 11010 11011
2 127.0.0.1 11020 11021
3 127.0.0.1 11030 11031

Important tip #1: Always provide IP addresses instead of hostnames. If a machine running a replica is not correctly configured, BFT-SMaRt may fail to bind to the appropriate IP address and use the loopback address instead (127.0.0.1). This phenomenon may prevent clients and/or replicas from successfully establishing a connection among them.

Important tip #2: Clients requests should not be issued before all replicas have been properly initialized. Replicas are ready to process client requests when each one outputs Ready to process operations in the console.

The system configurations also have to be specified (seeconfig/system.config). Most of the parameters are self-explanatory.

Important tip #3: When using the library in real systems, always make sure to set system.communication.defaultkeys to false and system.communication.useSignatures to 1. Also make sure that only the config/keys directory has the private key for the respective replica/client.

Compiling

Type ./gradlew installDist in the main directory. The required jar files and default configuration files will be available in the build/install/library directory.

WARNING: You might need to give execution permission to the gradlew script.

Copy content of build/install/library into multiple folders for local testing or machines for distributed testing.

Running the counter demonstration

You can run the counter demonstration by executing the following commands, from within the folders containing compiled code across four different consoles (4 replicas, to tolerate 1 fault):

./smartrun.sh bftsmart.demo.counter.CounterServer 0
./smartrun.sh bftsmart.demo.counter.CounterServer 1
./smartrun.sh bftsmart.demo.counter.CounterServer 2
./smartrun.sh bftsmart.demo.counter.CounterServer 3

Important tip #4: If you are getting timeout messages, it is possible that the application you are running takes too long to process the requests or the network delay is too high and PROPOSE messages from the leader does not arrive in time, so replicas may start the leader change protocol. To prevent that, try to increase the system.totalordermulticast.timeout parameter in config/system.config.

Important tip #5: Never forget to delete the config/currentView file after you modify config/hosts.config or config/system.config. If config/currentView exists, BFT-SMaRt always fetches the group configuration from this file first. Otherwise, BFT-SMaRt fetches information from the other files and creates config/currentView from scratch. Note that config/currentView only stores information related to the group of replicas. You do not need to delete this file if, for instance, you want to change the value of the request timeout.

Once all replicas are ready, the client can be launched as follows:

./smartrun.sh bftsmart.demo.counter.CounterClient 1001 <increment> [<number of operations>]

If <increment> equals 0 the request will be read-only. Default <number of operations> equals 1000.

Important tip #6: Always make sure that each client uses a unique ID. Otherwise, clients may not be able to complete their operations.

Read-only optimization

BFT-SMaRt implements a read-only optimization that allows replicas to process read-only requests without executing consensus protocol. Recent work (see section Additional information and publications) has shown that this optimization could violate the liveness property of the system.

The recent BFT-SMaRt version implements the proposed solution, which guarantees that the system will not violate the live property when using the read-only optimization. However, due to the high memory consumption of the current implementation, this optimization is turned off by default but can be enabled by setting the system.optimizations.readonly_requests parameter to true in the config/system.config file.

State transfer protocol(s)

BFT-SMaRt offers two state transfer protocols. The first is a basic protocol that can be used by extending the classes bftsmart.tom.server.defaultservices.DefaultRecoverable and bftsmart.tom.server.defaultservices.DefaultSingleRecoverable. These classes log requests into memory and periodically take snapshots of the application state.

The second, more advanced protocol can be used by extending the class bftsmart.tom.server.defaultservices.durability.DurabilityCoordinator. This protocol stores its logs to disk. To mitigate the latency of writing to disk, such tasks are done in batches and in parallel with the requests' execution. Additionally, the snapshots are taken at different points of the execution in different replicas.

Important tip #7: We recommend developers to use bftsmart.tom.server.defaultservices.DefaultRecoverable, since it is the most stable of the three classes.

Important tip #8: Regardless of the chosen protocol, developers must avoid using Java API objects like HashSet or HashMap, and use TreeSet or TreeMap instead. This is because serialization of Hash* objects is not deterministic, i.e, it generates different byte arrays for equal objects. This will lead to problems after more than f replicas used the state transfer protocol to recover from failures.

Group reconfiguration

The library also implements a reconfiguration protocol that can be used to add/remove replicas from the initial group.

You can add a replica to the group on-the-fly by executing the following command:

./smartrun.sh bftsmart.reconfiguration.util.DefaultVMServices <smart id> <ip address> <port client-to-replica> <port replica-to-replica>

You can remove a replica from the group on-the-fly by executing the following command:

./smartrun.sh bftsmart.reconfiguration.util.DefaultVMServices <smart id>

Important tip #9: Everytime you use the reconfiguration protocol, you must make sure that all replicas and the host where you invoke the above commands have the latest config/currentView file. The current implementation of BFT-SMaRt does not provide any mechanism to distribute this file, so you will need to distribute it on your own (e.g., using the scp command). You also need to make sure that any client that starts executing can read from the latest config/currentView file.

BFT-SMaRt under crash faults

You can run BFT-SMaRt in crash-faults only mode by setting the system.bft parameter in the configuration file to false. This mode requires fewer replicas to execute, but will not withstand full Byzantine behavior from compromised replicas.

Generating public/private key pairs

If you need to generate public/private keys for more replicas or clients, you can use the following command.

To generate RSA key pairs, execute the following command:

./smartrun.sh bftsmart.tom.util.RSAKeyPairGenerator <id> <key length> [config dir]

To generate ECDSA key pairs, execute the following command:

./smartrun.sh bftsmart.tom.util.ECDSAKeyPairGenerator <id> <domain parameter> [config dir]

Default config dir are config/keysRSA and config/keysECDSA, respectively. The commands above create key pairs both for clients and replicas. Alternatively, you can set the system.communication.defaultkeys to true in the config/system.config file to forces all processes to use the same public/private keys pair and secret key. This is useful when deploying experiments and benchmarks, because it enables the programmer to avoid generating keys for all principals involved in the system. However, this must not be used in a real deployments.

Additional information and publications

If you are interested in learning more about BFT-SMaRt, you can read:

  • The paper about its state machine protocol published in EDCC'12;
  • The paper about its advanced state transfer protocol published in Usenix'13;
  • The tool description published in DSN'14;
  • The paper about read-only optimization published in SRDS'21.

Feel free to contact us if you have any questions!