| commit | 34a0e056a9735557bfbdd9fb68a261d0a53167f2 | [log] [tgz] |
|---|---|---|
| author | Robin Thellend <rthellend@google.com> | Thu Jul 14 16:24:46 2016 -0700 |
| committer | Robin Thellend <rthellend@google.com> | Thu Jul 14 16:25:23 2016 -0700 |
| tree | d1eff6e84d6cb9e860c9811c8888d01dcd22c3c7 | |
| parent | d63d80b789dc99f3693868cf0f8c03b5cc494d77 [diff] |
TBR: Retry failed advertisement Change-Id: I1a3c98918ce5a58626cdc077c0399950c9f5f664
This package contains an experimental prototype for peer-to-peer messaging where there might not be a direct network connection between the sender and the receiver. Messages are relayed from peer to peer, and can tolerate long delays.
There are many mobile devices, especially in developing countries, that don't have a reliable connection with cloud service providers to exchange messages with each other and with cloud services. These devices often come in contact with other devices that might be better connected and could be used to carry messages for them.
The architecture is inspired by RFC 4838 (Delay-Tolerant Networking Architecture).
The Sender of the message gives it to a Messenger Node. The Messenger Node will store the message until it meets other nodes that can move the message towards the Recipient.
This prototype uses the Vanadium Discovery API to find nearby and global peers, and its own Messenger Interface to share messages.
Messages are signed and can be validated offline using the Vanadium Security.
The content of each message is opaque to everyone except the intended recipient. It is the sender's responsibility to encrypt the message content.
Access control is done with the Vanadium Security model. Each node can authorize messages based on the identity of the peer that is relaying it, or the identity of the original sender.
In addition, nodes can apply restrictions on the message size, lifespan, the number of nodes it has already been through, etc.
This implementation attempts to send all the messages to all the trusted peers that it encounters.
A smarter implementation will require better routing decisions. We will need a way to know approximately where the recipient is and only send the message to peers that allow it to make progress in the right direction.
There are still some potentially useful applications of the simple send-everywhere model.
In general, the number of nodes that implement the Messenger Interface has to be small. Sharing an unbounded number of messages with thousands or millions of Messenger nodes isn't a realistic objective.
Each member of the group implements the Messenger Interface, and access to each method is restricted to the members of the group.
Messages can be addressed to individual members, or the whole group, but everyone in the group can potentially end up with a copy of all messages.
The data mules implement the Messenger Interface and optionally the MessageRepository interface, and everyone in the group pulls the data they want from them.
The data mules share the messages with the other data mules. Occasionally, the data mules come in contact with cloud services to deliver outbound messages and receive inbound messages for the group.
In this model, the data mules are similar to mobile email servers. The members of the group don't share anything directly with the other members of the group. Instead, all messages are exchanged via the data mules.
The data mules control who can access the service they provide.
In this model, everyone who is interested in the content implements the Messenger Interface and authorizes Push requests using the message's SenderBlessings so that only content from trusted sources will be accepted. Each node can opportunistically push a copy of the messages to other nodes as long as the other nodes trust the original sender of the message.
When multiple devices want to send data to a central collection server, we need peers that implement the Messenger Interface and know how to relay message to the central collection server. These messages can be authorized using a combination of SenderBlessings and recipient.
The Messenger Server has 3 main components:
The Messenger Service implements the Messenger Interface to receive incoming messages of arbitrary sizes. The content of the messages is streamed from the client to the server. If the stream is interrupted, it can be resumed at a later time.
The message and its content are stored in the Messenger Storage.
When a new message is completely received, a notification is sent to the Peer Manager, and the message is added to all the existing peers' message queues.
The Messenger Storage uses the local file system to store the messages and to keep track of where the messages have been sent. Messages become visible for sharing after they are fully copied.
The Peer Manager uses the Discovery API to find peers with whom to communicate. When a new peer is found, all the existing messages are added to the send queue of this peer. Messages are removed from the queue after being successfully transmitted or when a fatal error occurs. If a transient error occurs, the message is added back to the queue to be retransmitted later.
There is a limit to how many peers a node will talk to. When there are too many peers, we use a peer selection algorithm that makes it likely that all the peers will eventually see all the messages.
For a peer limit L, the selected peers are the node's neighbors at distances 2^n, with n=[0..L].
In this diagram, each node talks to 4 peers. The first one is the first (2^0) neighbor. The second one is the second (2^1) neighbor. The third one is the fourth (2^2) neighbor, etc. So, Node A sends the message to Nodes B, C, E, and I.
After Node B received the message, it will send it to Nodes C, D, F, and J. Node C will send it to Nodes D, E, G, and K, etc. This continues until all nodes have a copy of the message.
Incoming and outgoing messages are controlled by rate limiting access lists (RateAcl). The access lists are applied to the identity of the peers, both sending and receiving the messages, and to the original sender of the message.
The demo commandline application in called vmsg.
It runs a Messenger Node to exchange text messages and files with other Messenger Nodes.
# Create credentials jiri go install v.io/x/ref/cmd/principal $JIRI_ROOT/release/go/bin/principal create $HOME/.vmsg-creds $JIRI_ROOT/release/go/bin/principal --v23.credentials=$HOME/.vmsg-creds seekblessings # Start chat app jiri go install messenger/vmsg $JIRI_ROOT/release/projects/go/bin/vmsg chat --v23.credentials=$HOME/.vmsg-creds --store-dir=/tmp/store