===================================================================

README for reproducibility submission

• Paper ID: 457
• Paper title: Hybrid Deterministic and Nondeterministic Execution of Transactions in Actor Systems

Snapper-Orleans (SIGMOD 2022)

Snapper is a transaction library currently built on top of Orleans. Snapper exploits the characteristics of actor-oriented programming to improve the performance of multi-actor transactions by employing deterministic transaction execution. where pre-declared actor access information is used to generate deterministic execution schedules.

Snapper can also execute actor transactions using conventional nondeterministic strategies, including S2PL, to account for scenarios where actor access information cannot be pre-declared.

A salient feature of Snapper is the ability to execute concurrent hybrid workloads, where some transactions are executed deterministically while the others are executed nondeterministically.

Source Code

The source code of Snapper can be downloaded here: Snapper source code.

Concurrency.Implementation contains the implementation of key components of Snapper, including Coordinator and Transactional actor discussed in the paper. The code files are named as GlobalTransactionCoordinatorGrain.cs and TransactionExecutionGrain.cs respectively. And Persist.Grains\PersistSingleton.cs implements the Logger.

SmallBank and TPCC are two benchmarks we used. SmallBank.Grains and TPCC.Grains contain the transaction logic of user-defined actors. For example, the SmallBank.Grains\CustomerAccountGroupGrain.cs extends the base class TransactionExecutionGrain in order to get transactional guarantees provided by Snapper.

SnapperSiloHost is used to configure and start the server process. SnapperExperimentProcess contains the client program that sends transaction requests to the server.

Software Dependency

• SDK: Snapper is developed with C# on .NET platform. To compile and run the source code, .NET Core SDK 3.1.301 should be installed.
• NuGet packages: The NuGet packages and corresponding versions are listed in each *.csproj file. Orleans 3.4.3 is the one we used for running experiments.
• Matlab: used for generating figures with given experimental data.

Cloud Deployment

• Server: The server side is an Orleans Cluster which contains 1 silo. The server is deployed on an AWS EC2 instance.
• Cluster membership table: the Orleans Cluster membership information is stored in a DynamoDB table. To get access to DynamoDB, the ServiceRegion, AccessKey and SecretKey should be provided.
• Client: We use Orleans Clients to send transaction requests to the server. In our experiments, SnapperExperimentProcess is deployed on another AWS EC2 instance.
• Both EC2 instances should co-locate in the same region and availability zone.

Datasets

This section explains how the transaction workloads are generated and used.

• Uniform & Hotspot distribution: Both types of workloads are generated on-the-fly when the SnapperExperimentProcess is started.
• Zipfian distribution: This workload needs to be generated in advance. In our client program, the zipfian distributed workloads are read from pre-generated data files. The data generator can be found in SmallBank.DataGenerator.
  • Run the data generator: dotnet run --project SmallBank.DataGenerator\SmallBank.DataGenerator.csproj
  • The generated dataset can be found in the folder data\zipfian_workload.

How To Run Experiments

Follow the steps below to re-produce all the experimental results presented in the paper.

Prepare the virtual machines

1. Start a Server machine and a Client machine on AWS Cloud Platform.
   • Both VMs should choose AMI Microsoft Windows Server 2019 Base.
   • Specification for Server VM: c5n.9xlarge (36-core, 96GB memory), 128GB io2 SSD, 63999 IOPS
   • Specification for Client VM: c5n.9xlarge (36-core, 96GB memory), 40GB gp2 SSD
   • Both VMs should locate in the same region and same subnet (eg. us-east-2a), and put in the same placement group.
   • Configure the security group and make sure all network traffic is allowed between these two VMs.
2. Connect to and configure both VMs.
   • Turn off IE Enhanced Security Configuration in Server Manager ==> Local Server.
   • Turn off firewall.
3. Install git on both VMs.
4. Install .NET Core SDK 3.1.301 on both VMs.

Prepare the source code and dataset

1. Download the Snapper source code (main branch) to both Server machine and Client machine.
   • git clone -b main https://github.com/diku-dk/Snapper-Orleans.git.
2. Create the file AWS_credential.txt, which should contain 3 lines: ServiceRegion (eg. us-east-2), AccessKey, SecretKey.
3. Put the AWS_credential.txt file into the Snapper-Orleans folder on both VMs.
4. Run the data generator on the Client VM:
   • dotnet run --project SmallBank.DataGenerator\SmallBank.DataGenerator.csproj.
   • It will take around 5min to complete.

Run the experiments and print figures (Fig.12, 13, 14, 16, 17)

1. Run the powershell script on the Server VM to start the server: .\RunSilo.ps1.
2. Run the powershell script on the Client VM to start the client: .\RunClient.ps1.
   • Both powershell scripts are included in the source code folder.
   • The client should be started without waiting for the server side script to complete.
3. It will take around 4 hours to complete.
4. Check all the results in data\result.txt stored on the Client machine.
5. To print the same figures as presentecd in the paper, you need to do the following steps in order:
   • Run the project to prepare data: dotnet run --project SnapperPrepareDataForFigures\SnapperPrepareDataForFigures.csproj.
   • Run the matlab script data\LoadDataForFigures.m to load data to matlab workspace.
   • Run the matlab script data\PrintFigures.m to print figures.
   • The data we used for figures in the paper can be found in data\LoadDataInPaper.m.

Run the experiments and print figures (Fig.15)

0. Download the Snapper source code (InvOrleansTxn branch) to both Server machine and Client machine.
   • git clone -b InvOrleansTxn https://github.com/diku-dk/Snapper-Orleans.git.
   • The difference between the main and InvOrleansTxn branches is that, in InvOrleansTxn branch, we made some changes to the Orleans source code so to collect some time intervals while executing an Orleans transaction.
1. Run the powershell script on the Server VM to start the server: .\RunSilo.ps1.
2. Run the powershell script on the Client VM to start the client: .\RunClient.ps1.
   • Both powershell scripts are included in the source code folder.
   • The client should be started without waiting for the server side script to complete.
3. It will take around 10min to complete.
4. Check all the results in data\breakdown_latency.txt stored on the Client machine.
   • In the result file, each row contains the breakdown latency of a single transaction. Each row has 9 values that represent 9 time intervals as described in the paper (Fig.15). To get the results shown in Fig.15, we need to first filter out abnormal data points (outliers) such that, for each time interval, standard_deviation < 25% * average. Then we should calculate the average value for each of the time intervals.
5. To print the same figures as presentecd in the paper, you need to do the following steps in order:
   • Run the project to prepare data: dotnet run --project SnapperPrepareDataForFigures\SnapperPrepareDataForFigures.csproj.
   • Run the matlab script data\LoadDataForFigures.m to load data to matlab workspace.
   • Run the matlab script data\PrintFigures.m to print figures.
   • The data we used for figures in the paper can be found in data\LoadDataInPaper.m.

===================================================================