Allmydata, Inc. [1], provider of the "Allmydata" consumer backup product, is pleased to announce the first public release of "Tahoe", a secure, distributed storage grid with a free-software licence. The source code that we are releasing is the current working prototype for Allmydata's next-generation product. This release is targeted at hackers who are willing to use a minimal, text-oriented web user interface. This prototype is not recommended for storage of confidential data nor for data which is not otherwise backed up, but it already implements a functional distributed storage grid and is useful for experimentation, prototyping, and extension. LICENCE Tahoe is offered under the GNU General Public License (v2 or later), with the added permission that, if you become obligated to release a derived work under this licence (as per section 2.b), you may delay the fulfillment of this obligation for up to 12 months. INSTALLATION This release of Tahoe works and passes all unit tests on Linux/x86, Linux/amd64, Mac/Intel, Mac/PPC, Windows-native, and Cygwin. To install, download the tarball [2], untar it, go into the resulting directory, and follow the directions in the README [3]. USAGE Once installed, create a "client node". Instruct this client node to connect to a specific "introducer node" by means of config files in the client node's working directory. To join a public grid, copy in the .furl files for that grid. To create a private grid, run your own introducer, and copy its .furl files. See the README for step-by-step instructions. Each client node runs a local webserver (enabled by writing the desired port number into a file called 'webport'). The front page of this webserver shows the node's status, including which introducer is being used and which other nodes are connected. Links from the status page lead to others that give access to a shared virtual filesystem, in which each directory is represented by a separate page. Each directory page shows a list of the files available there, with download links, and forms to upload new files. Other ways to access the filesystem are planned, as well as other structures than the single globally-shared namespace implemented by this release: please see the roadmap.txt [5] for some rough details. HACKING AND COMMUNITY Please join the mailing list [4] to discuss the ideas behind Tahoe and extensions of and uses of Tahoe. Patches that extend and improve Tahoe are gratefully accepted -- roadmap.txt shows the next improvements that we plan to make. You can browse the revision control history, source code, and issue tracking at the Trac instance [6]. Please see the buildbot [7], which shows how Tahoe builds and passes unit tests on each checkin, and the code coverage results [8] and percentage-covered graph [9], which show how much of the Tahoe source code is currently exercised by the test suite. NETWORK ARCHITECTURE Each peer maintains a connection to each other peer. A single distinct server called an "introducer" is used to discover other peers with which to connect. To store a file, the file is encrypted and erasure coded, and each resulting share is uploaded to a different peer. The secure hash of the encrypted file and the encryption key are packed into a URI, knowledge of which is necessary and sufficient to recover the file. To fetch a file, starting with the URI, a subset of shares is downloaded from peers, the file is reconstructed from the shares, and then decrypted. A single distinct server called a "vdrive server" maintains a global mapping from pathnames/filenames to URIs. We are well aware of the limitations of decentralization and scalability inherent in this prototype. In particular, the completely-connected property of the grid and the requirement of a single distinct introducer and vdrive server limits the possible size of the grid. We have plans to loosen these limitations (see roadmap.txt). Currently it should be noted that the grid already depends as little as possible on the accessibility and correctness of the introduction server and the vdrive server. Also note that the choice of which servers to use is easily configured -- you should be able to set up a private grid for you and your friends almost as easily as to connect to our public test grid. SOFTWARE ARCHITECTURE Tahoe is a "from the ground-up" rewrite, inspired by Allmydata's existing consumer backup service. It is primarily written in Python. Tahoe is based on the Foolscap library [10] which provides a remote object protocol inspired by the capability-secure "E" programming language [11]. Foolscap allows us to express the intended behavior of the distributed grid directly in object-oriented terms while relying on a well-engineered, secure transport layer. The underlying networking is provided by the Twisted library [12]. Computationally intensive operations are performed in native compiled code, such as the "zfec" library for fast erasure coding (also available separately: [13]). [1] http://allmydata.com [2] http://allmydata.org/source/tahoe/tahoe-0.2.0-0-UNSTABLE.tar.gz [3] http://allmydata.org/trac/tahoe/browser/README [4] http://allmydata.org/cgi-bin/mailman/listinfo/tahoe-dev [5] http://allmydata.org/trac/tahoe/browser/roadmap.txt [6] http://allmydata.org/trac/tahoe [7] http://allmydata.org/buildbot [8] http://allmydata.org/tahoe-figleaf/figleaf/ [9] http://allmydata.org/tahoe-figleaf-graph/hanford.allmydata.com-tahoe_figleaf.html [10] http://twistedmatrix.com/trac/wiki/FoolsCap [11] http://erights.org/ [12] http://twistedmatrix.com/ [13] http://allmydata.org/trac/tahoe/browser/src/zfec