Syndicate

Syndicate is a scalable software-defined storage system for wide-area networks. Syndicate creates global read/write storage volumes on top of existing systems, but while preserving end-to-end app-specific storage semantics. With less than 200 lines of Python, and without having to run any servers of your own, you use a Syndicate volume to do things like:

Scale up reads on a remote server with a CDN while guaranteeing that readers always see fresh data.
Turn a collection of URLs as a copy-on-write shared filesystem which preserves your and your peers' changes.
Implement end-to-end encryption on top of your Dropbox folder, while mirroring your files to Amazon S3 and Google Drive.
Publish your code for the world to download, while guaranteeing end-to-end authenticity and integrity.

This is the core software package, which contains the Syndicate Metadata Serivice and libraries needed to create Syndicate gateways.

Syndicate is funded by the NSF #1541318. Syndicate is overseen by the NSF CC*DNI DIBBS group, part of the NSF's Advanced Cyberinfrastructure program.

Examples

Here are a few examples of how we are currently using Syndicate:

Augmenting scientific storage systems (like iRODS) and public datasets (like GenBank) with ingress Web caches in order to automatically stage large-scale datasets for local compute clusters to process.
Creating a secure DropBox-like "shared folder" system for OpenCloud that augments VMs, external scientific datasets, and personal computers with a private CDN, allowing users to share large amounts of data with their VMs while minimizing redundant transfers.
Scalably packaging up and deploying applications across the Internet.
Creating webmail with transparent end-to-end encryption, automatic key management, and backwards compatibility with email. Email data gets stored encrypted to user-chosen storage service(s), so webmail providers like Gmail can't snoop. See the SyndicateMail project for details.

The "secret sauce" is a novel programming model that lets the application break down storage I/O logic into a set of small, mostly-orthogonal but composible I/O steps. By combining these steps into a networked pipeline and controlling when and where each step can execute in the network, the application can preserve domain-specific invariants end-to-end without having to build a whole storage abstraction layer from scratch.

Where can I learn more?

Please see a look at our whitepaper, published in the proceedings of the 1st International Workshop on Software-defined Ecosystems (colocated with HPDC 2014).

Also, please see our NSF grant for our ongoing work.

Building

To build, type:

    $ make MS_APP_ADMIN_EMAIL=(your admin account email)

To build Syndicate, you will need the following tools, libraries, and header files:

Quick Setup

Syndicate's end-point components (gateways) coordinate via an untrusted logically central metadata service (MS). The MS implementation currently runs in Google AppEngine, or in AppScale. You can test it locally with the Python GAE development environment. Please see the relevant documentation for GAE, AppScale, and the development environment for deployment instructions. You should be able to run the MS on the free tier in GAE.

The MS code is built to ./build/out/ms. You can deploy it to GAE from there, or run it with the development environment with:

    $ dev_appserver.py ./build/out/ms

Now, you need to set up your ~/.syndicate directory. An admin account and keypair are automatically generated by the build process (in ./build/out/ms). To use them, type:

    $ cd ./build/out/bin/
    $ ./syndicate setup $MS_APP_ADMIN_EMAIL ../ms/admin.pem http://localhost:8080

Replace $MS_APP_ADMIN_EMAIL from the value you passed to make earlier, and replace http://localhost:8080 with the URL to your MS deployment (if you're not running the dev_appserver.py).

First, you must create a volume. Here's an example that will create a volume called test-volume, owned by the administrator, with a 64k block size:

    $ ./syndicate create_volume name="test-volume" description="This is a description of the volume" blocksize=65536 email="$MS_APP_ADMIN_EMAIL"

Now, you can create gateways for the volume--the network-addressed processes that bind existing storage systems together and overlay your application's domain-specific I/O coordination logic over them. To create a user gateway that will allow you to interact with the test-volume volume's data via the admin account, type:

    $ ./syndicate create_gateway email="$MS_APP_ADMIN_EMAIL" volume="test-volume" name="test-volume-UG-01" private_key=auto type=UG
    $ ./syndicate update_gateway test-volume-UG-01 caps=ALL

This will create a user gateway called test-volume-UG-01, generate and store a key pair for it automatically, and enable all capabilities for it. Similarly, you can do so for replica gateways (using type=RG) and acquisition gateways (using type=AG). Replica gateways take data from user gateways and forward it to persistent storage. Acquisition gateways consume data from existing storage and make it available as files and directories in the volume.

Gateways are dynamically programmable via drivers. There are some sample drivers in python/syndicate/drivers/ag/disk (an AG driver that imports local files into a volume) and python/syndicate/drivers/rg/s3 (an RG driver that replicates data to Amazon S3). Drivers are specific to the type of gateway--you cannot use an RG driver for an AG, for example.

To set a driver, you should use the driver= flag in the update_gateway directive. For example:

    $ ./syndicate update_gateway sample-AG driver=../python/syndicate/drivers/ag/disk

The gateway should automatically fetch and instantiate the driver before the command completes.

Finally, you can proceed to run the gateways as follows. Each gateway takes -v for the volume name, -g for the gateway name, and -u for the user name. For example:

    $ # Start a UG
    $ /path/to/user/gateway -u "$MS_APP_ADMIN_EMAIL" -v "test-volume" -g "test-volume-UG-01"

Other options include -f for foreground, -d for debug level, and -c for path to the configuration file.

Here's a more complete example that sets up a user gateway and a replica gateway for the user bob@example.com. Both gateways will run locally. The user gateway will be able to read, write, and coordinate writes for files, and the replica gateway will have all such capabilities disabled (meaning it can only get, put, and delete blocks). The replica gateway will listen on 31112, and the user gateway will listen on 31111.

In this example, we will put, and then cat a file from a replica gateway. This is taken from a test scenario in the test suite.

    $ # set up the admin account
    $ ./syndicate --trust_public_key -c /tmp/syndicate-test-ezOKLS/syndicate.conf setup "$MS_APP_ADMIN_EMAIL" ./ms_src/admin.pem http://localhost:8080
    $
    $ # create an unprivileged user, and automatically generate a keypair
    $ ./syndicate -c /tmp/syndicate-test-ezOKLS/syndicate.conf create_user email=jcnelson@cs.princeton.edu private_key=auto
    $ 
    $ # create a volume with 4K blocks
    $ ./syndicate -c /tmp/syndicate-test-ezOKLS/syndicate.conf create_volume name=testvolume-f0170356 description=test volume blocksize=4096 email=jcnelson@cs.princeton.edu
    $
    $ # make a replica gateway, and automatically generate a keypair
    $ ./syndicate -c /tmp/syndicate-test-ezOKLS/syndicate.conf create_gateway email=jcnelson@cs.princeton.edu volume=testvolume-f0170356 name=testgateway-RG-c65925b4 private_key=auto type=RG
    $ 
    $ # drop RG caps
    $ ./syndicate -c /tmp/syndicate-test-ezOKLS/syndicate.conf update_gateway testgateway-RG-c65925b4 caps=NONE
    $ 
    $ # update driver and port number
    $ ./syndicate -c /tmp/syndicate-test-ezOKLS/syndicate.conf update_gateway testgateway-RG-c65925b4 port=31112 driver=./python/syndicate/rg/drivers/disk
    $
    $ # start the RG
    $ /usr/local/bin/syndicate-rg -c /tmp/syndicate-test-ezOKLS/syndicate.conf -d2 -f -u jcnelson@cs.princeton.edu -v testvolume-f0170356 -g testgateway-RG-c65925b4
    $
    $ # make a UG
    $ ./syndicate -c /tmp/syndicate-test-ezOKLS/syndicate.conf create_gateway email=jcnelson@cs.princeton.edu volume=testvolume-f0170356 name=testgateway-UG-307cff4b private_key=auto type=UG
    $ 
    $ # give the UG all capabilities, and set its port number
    $ ./syndicate -c /tmp/syndicate-test-ezOKLS/syndicate.conf update_gateway testgateway-UG-307cff4b caps=ALL port=31111
    $
    $ # put the file /tmp/tmpsZigf8 to /put-e02dfe9a
    $ /usr/local/bin/syndicate-put -c /tmp/syndicate-test-ezOKLS/syndicate.conf -u jcnelson@cs.princeton.edu -v testvolume-f0170356 -g testgateway-UG-307cff4b /tmp/tmpsZigf8 /put-e02dfe9a
    $ 
    $ # cat it again, to stdout
    $ /usr/local/bin/syndicate-cat -c /tmp/syndicate-test-ezOKLS/syndicate.conf -u jcnelson@cs.princeton.edu -v testvolume-f0170356 -g testgateway-UG-557992d7 /put-e02dfe9a

syndicate-storage / syndicate-core

readme

Syndicate

Examples

Where can I learn more?

Building

Quick Setup