The distributed shared objects (DSO) datastore allows to share, call and persist objects remotely.
git clone https://github.com/crucial-project/dso
./dso/client/src/test/bin/local/test.sh -create // create a server using docker
./dso/client/src/test/bin/local/test.sh -[blobs|counters|countdownlatch|barrier|sbarrier]
./dso/client/src/test/bin/local/test.sh -delete // delete the server
Cloud applications are commonly split into three distinct tiers: presentation, application and storage. The presentation tier displays information related to the application services, such as web pages. The application tier contains the business logic. The storage tier is in charge of serving and persisting data.
Traditionally, an object-relational mapping (ORM) converts the data between the application and storage tiers. The ORM materializes the frontier between the two tiers, and it reduces the coupling. However, it also forces to repeatedly convert the objects between their in-memory and their serialized representations back and forth. This negatively impacts performance and increases execution costs.
DSO is a general-purpose synchronization and data sharing framework. With DSO, instead of fetching objects from storage, the application directly calls them. DSO ensures that the objects are persisted and shared consistently among several client machines.
DSO offers several client-side programming libraries.
The most complete one is for the Java language.
To declare a DSO object in Java, the programmer uses the keyword @Shared on the field of an object.
As an example, consider the following two classes.
class Hero{@Shared Room location;}
class Room{Treasure loot();}
The Hero class contains a location field annotated with @shared.
This tells DSO to push the location to the storage tier, allowing several instances of Hero on several application machines to access the same location object transparently.
DSO ensures that the object are strongly consistent over time.
In the example above, this means for instance that if two heroes stand in the same room, only one of them may loot the treasure.
More precisely, the synchronization contract of every DSO object o is that o is atomic, aka. linearizable.
In Java, this is equivalent to guarding every method m of some object o with synchronized(o){o.m}.
DSO includes a library of shared objects (counter, integer, list, maps, barrier, etc.). The provided objects are listed in the client module.
DSO follows a standard client-server architecture. The current server implementation is layered above Infinispan.
The project includes libraries for Python as well as the Shell language.
Below, we explain how to deploy a DSO server and access it in different contexts.
To run the server locally, one may type:
docker run --net host --rm --env EXTRA="-rf 2" --env CLOUD=local --env PORT=11222 0track/dso-server:latest
Local (unit) tests are available under ./dso/client/src/test/bin/local/test.sh.
To use DSO in a distributed context, the simplest approach is to rely on a container orchestrator. The project includes a set of scripts for Kubernetes (k8s). To run the tests in a k8s cluster, one may use the following commands:
./client/src/test/bin/k8s/[aws,gcp]/bootstrap.sh
./client/src/test/bin/k8s/test.sh -create
./client/src/test/bin/k8s/test.sh -[blobs|counters|countdownlatch|barrier|sbarrier]
./client/src/test/bin/k8s/test.sh -delete
As pointed above DSO uses internally Infinispan which itself relies on the JGroups stack for discovery and communication. To deploy one or more servers in EC2, DSO uses the S3 ping facility of JGroups.
The configuration file for the server is jgroups-dso-ec2.xml.
To deploy the server in your own EC2 instances, you need to fix the following 3 parameters in this XML file.
<S3_PING
location="your_bucket"
access_key="your_key"
secret_access_key="your_secret"
/>
The access key and the corresponding secret are credentials to write in the bucket. We advice you to create an IAM in EC2 for that purpose.
It is also possible to build your own server, e.g., to deploy a specific library of shared objects.
First, move to version ispn13-standalone, using git ispn13-standalone.
Then, to build an archive containing the server, use the mvn install -DskipTests at the root of the project.
The server archive, named dso-server-*.tar.gz, is located in server/target.
To launch the server run the script server.sh from the root of the archive.
Every class used by a client, e.g., the Hero class above, should be known at the server.
This requires to add the appropriates .class or jar files to the classpath of the server.
Alternatively, the server can dynamically load new jars (by default, they should be located in /tmp).
More uses cases are provided in the examples repository of the Crucial project.
The serverless shell uses DSO to synchronize distributed shell instances executed atop a FaaS infrastructure (e.g., AWS Lambda).
We used DSO to port the Random Forest classification algorithm in the Smile library.