Home

Awesome

Buttwoo feed format

Status: Ready for use

Buttwoo is a new binary feed format for SSB. It draws inspiration from bamboo and meta feeds.

The format is designed to be simple to implement, be easy to integrate into existing implementations, be backwards compatible with existing applications and lastly to be performant.

Buttwoo uses bipf for encoding since its a good foundation for an append-only-log system (write once, many reads). It uses ssb-bfe for binary encodings of feed IDs and message IDs.

Format

A buttwoo message consists of a bipf-encoded array of 3 fields:

The message ID is the BFE encoding of the blake3 hash of the concatenation of metadata bytes with signature bytes.

Metadata

The metadata field is a bipf-encoded array with 8 fields:

Signature

The signature uses the same HMAC signing capability (sodium.crypto_auth) and sodium.crypto_sign_detached as in the classic SSB format (ed25519).

It is important to note that one author can have multiple feeds, each feed defined as author + parent. sequence and previous relates to the feed. Also note that unless parent is used, this behaves exactly like an ordinary classic SSB feed.

Content

The content is a free form field. When unencrypted, it MUST be a bipf-encoded object. If encrypted, content MUST be an ssb-bfe encrypted data format.

Subfeeds

As noted above it is possible to have multiple feeds with the same author. To initiate a subfeed one create a special message on the feed. This message must use the 0x01 tag and content can include extra information about what is contained in the subfeed, such as the feed purpose. The id of this message serves as the parent id of each message on the subfeed.

In contrast with bendy butt, subfeeds maintain the same feed identitier. This makes it easier to work with in situations where, what in classic SSB would be single feed, is split into multiple parts. As an example, a feed could be split into: about messages, the social graph and ordinary messages. While on the other hand bendy butt had a clear separation between what are meta feeds and what are normal feeds, allowing normal feeds to use different feed formats. In this way, they can be seen as complementary.

Performance

A benchmark of a prototype shows the time it takes to validate and convert for storing in a database to be reduced in half for single message validation. Similar to classic it is possible for lite clients to queue up messages for validation and only check the signature of a random or the latest message. This can improve the bulk validation substantially in onboarding situations.

Size

There is roughly a 20% size reduction in network traffic compared to classic format.

Validation

A buttwoo message MUST conform to the following rules:

Content, if available MUST conform to the following rules:

Integration with existing SSB stack

EBT

Data sent over the wire should be bipf encoded as:

transport:  [metadata, signature, content]
metadata:   [author, parent, sequence, timestamp, previous, tag, contentLen, contentHash]

If content is not encrypted, then this value will be a bipf encoded buffer. If encrypted, this will be a base64 encoded BFE string representation.

The feedId should be author + parent.

Design choices

Keeping timestamps

One difference between buttwoo and bamboo is that bamboo does not have timestamps in the format, instead leaving those to be part of the content. This is important for private messages. This choice was mostly formed from an backwards compatible perspective. It should be noted that with meta feeds it becomes possible to store the messages of a private group in a feed that is only exchanged with members of the group, thus leaving the potential metadata leak problem void.

Lipmaa links

Another difference between buttwoo and bamboo is that lipmaa links are not included. Lipmaa links allows partial replication in cases where the specific subset of messages are not important, only that they form a valid chain back to the root message. This comes at the cost that validation is now more expensive because for roughly every second message an additional link needs to be checked.

Furthermore with meta feeds we can now partition the data of a feed into subsets (such as the friends graph and about messages in separate feeds). This leaves public messages where lipmaa links based partial replication could be useful. Also note for this to really work, the friend graph needs to include the root hash besides the feed, otherwise an adversary (given the private key) could still create a fake feed.

Lastly we already have another mechanism for doing partial replication, namely tangles where messages from multiple feeds are linked together and form their own chain.

To sum up, the advantages does not outweight the disadvantages for SSB.

Sign hash of the content

Similar to bamboo the signature is over the hash of the content and not the actual content. This allows validation of a log without the actual content. It should be noted that deletion is a whole topic in itself, this is just to note that this format also supports this case.

Bipf encoding

While many encodings could be used for encoding of especially the metadata part, bipf is a relatively simple format. The JavaScript implementation is roughly 250 lines for encode and decode. Bipf allows the content to be reused when encoding for the database in ssb-db2 resultating in roughly half the time used compared to existing feed format.