f525f99383
|
||
|---|---|---|
| .. | ||
| README.md | ||
README.md
| slug | title | name | status | category | tags | editor | contributors |
|---|---|---|---|---|---|---|---|
| 51 | 51/WAKU2-RELAY-SHARDING | Waku v2 Relay Sharding | raw | Standards Track | Daniel Kaiser <danielkaiser@status.im> |
Abstract
This document describes ways of sharding the Waku relay topic, allowing Waku networks to scale in the number of content topics.
Note: Scaling in the size of a single content topic is out of scope for this document.
Background and Motivation
Unstructured P2P networks are more robust and resilient against DoS attacks compared to structured P2P networks). However, they do not scale to large traffic loads. A single libp2p gossipsub mesh, which carries messages associated with a single pubsub topic, can be seen as a separate unstructured P2P network (control messages go beyond these boundaries, but at its core, it is a separate P2P network). With this, the number of Waku relay content topics that can be carried over a pubsub topic is limited. This prevents app protocols that aim to span many multicast groups (realized by content topics) from scaling.
This document specifies three pubsub topic sharding methods (with varying degrees of automation), which allow application protocols to scale in the number of content topics. This document also covers discovery of topic shards.
Named Sharding
Named sharding offers apps to freely choose pubsub topic names. App protocols SHOULD follow the naming structure detailed in 23/WAKU2-TOPICS. With named sharding, managing discovery falls into the responsibility of apps.
The default Waku pubsub topic /waku/2/default-waku/proto can be seen as a named shard available to all app protocols.
Note: Future versions of this document are planned to give more guidance with respect to discovery via 33/WAKU2-DISCV5, DNS discovery, and inter-mesh discovery via gossipsub control messages (also using circuit relay). It might make sense to deprecate 23/WAKU2-TOPICS as a separate spec and merge it here.
From an app protocol point of view, a subscription to a content topic waku2/xxx on a shard named /mesh/v1.1.1/xxx would look like:
subscribe("/waku2/xxx", "/mesh/v1.1.1/xxx")
Static Sharding
Static sharding offers a set of shards with fixed names. Assigning content topics to specific shards is up to app protocols, but the discovery of these shards is managed by Waku.
Static shards are managed in shard clusters of 1024 shards per cluster.
Waku static sharding can manage 2^16 shard clusters.
Each shard cluster is identified by its index (between 0 and 2^16-1).
A specific shard cluster is either globally available to all apps (like the default pubsub topic), specific for an app protocol, or reserved for automatic sharding (see next section).
Note: This leads to
2^16 * 1024 = 2^26shards for which Waku manages discovery.
App protocols can either choose to use global shards, or app specific shards. (In future versions of this document, automatic sharding, described in the next section, will become the default.)
Like the IANA ports, shard clusters are divided into ranges:
| index (range) | usage |
|---|---|
| 0 | global |
| 1 - 15 | reserved |
| 16 - 1023 | specific app protocols |
| 1024 - 49152 | all app protocols |
| 49152 - 65535 | automatic sharding |
The informational RFC 52/WAKU2-RELAY-STATIC-SHARD-ALLOC lists the current index allocations.
The global shard with index 0 and the "all app protocols" range are treated in the same way, but choosing shards in the global cluster has a higher probability of sharing the shard with other apps. This offers k-anonymity and better connectivity, but comes at a higher bandwidth cost.
The name of the pubsub topic corresponding to a given static shard is specified as
/waku/2/static-rshard/<shard_cluster_index>/<shard_number>,
an example for the 2nd shard in the global shard cluster:
/waku/2/static-rshard/0/2.
Note: Because all shards distribute payload defined in 14/WAKU2-MESSAGE via protocol buffers, the pubsub topic name does not explicitly add
/prototo indicate protocol buffer encoding. We usershard(as well asrsas the ENR key) to indicate these are relay shard clusters; further shard types might follow in the future.
From an app point of view, a subscription to a content topic waku2/xxx on a static shard would look like:
subscribe("/waku2/xxx", 43)
for global shard 43. And for shard 43 of the Status app (which has allocated index 16):
subscribe("/waku2/xxx", 16, 43)
Discovery
Waku v2 supports the discovery of peers within static shards, so app protocols do not have to implement their own discovery method.
Nodes add information about their shard participation in their 31/WAKU2-ENR. Having a static shard participation indication as part of the ENR allows nodes to discover peers that are part of shards via 33/WAKU2-DISCV5 as well as via DNS.
Note: In the current version of this document, sharding information is directly added to the ENR. (see Ethereum ENR sharding bit vector here Static relay sharding supports 1024 shards per cluster, leading to a flag field of 128 bytes. This already takes half (including index and key) of the ENR space of 300 bytes. For this reason, the current specification only supports a single shard cluster per node. In future versions, we will add further (hierarchical) discovery methods. We will update 31/WAKU2-ENR accordingly, once this RFC moves forward.
The representation in the ENR is specified as follows.
| key | value |
|---|---|
rs |
<2-byte index> | <128-byte flag field> |
The ENR key is rs.
The value is comprised of a two-byte shard cluster index in network byte order concatenated with a 128-byte wide bit vector.
The bit vector indicates which shards of the respective shard cluster the node is part of.
The right-most bit in the bit vector represents shard 0, the left-most bit represents shard 1023.
The representation in the ENR is inspired by Ethereum shard ENRs),
and this).
Example:
| key | value |
|---|---|
rs |
16u16 | 0x[...]0000100000003000 |
The [...] in the example indicates 120 0 bytes.
This example node is part of shards 13, 14, and 45 in the Status main-net shard cluster (index 16).
Automatic Sharding
Note: Automatic sharding is not yet part of this specification. This section merely serves as an outlook. A specification of automatic sharding will be added to this document in a future version.
Automatic sharding is a method for scaling Waku relay in the number of (smaller) content topics. It automatically maps Waku content topics to pubsub topics. Clients and protocols building on Waku relay only see content topics, while Waku relay internally manages the mapping. This provides both scaling as well as removes confusion about content and pubsub topics on the consumer side.
From an app point of view, a subscription to a content topic waku2/xxx using automatic sharding would look like:
subscribe("/waku2/xxx", auto=true)
The app is oblivious to the pubsub topic layer.
(Future versions could deprecate the default pubsub topic and remove the necessity for auto=true.)
The basic idea behind automatic sharding: Content topics are mapped using consistent hashing. Like with DHTs, the hash space is split into parts, each covered by a Pubsub topic (mesh network) that carries content topics which are mapped into the respective part of the hash space.
There are (at least) two issues that have to be solved: Hot spots and Discovery (see next subsection).
Hot spots occur (similar to DHTs), when a specific mesh network becomes responsible for (several) large multicast groups (content topics). The opposite problem occurs when a mesh only carries multicast groups with very few participants: this might cause bad connectivity within the mesh. Our research goal here is finding efficient ways of distribution. We could get inspired by the DHT literature. We also have to consider: If a node is part of many content topics which are all spread over different shards, the node will potentially be exposed to a lot of network traffic.
Discovery
For the discovery of automatic shards this document specifies two methods (the second method will be detailed in a future version of this document).
The first method uses the discovery introduced above in the context of static shards.
The index range 49152 - 65535 is reserved for automatic sharding.
Each index can be seen as a hash bucket.
Consistent hashing maps content topics in one of these buckets.
The second discovery method will be a successor to the first method, but is planned to preserve the index range allocation. Instead of adding the data to the ENR, it will treat each array index as a capability, which can be hierarchical, having each shard in the indexed shard cluster as a sub-capability. When scaling to a very large number of shards, this will avoid blowing up the ENR size, and allows efficient discovery. We currently use 33/WAKU2-DISCV5 for discovery, which is based on Ethereum's discv5. While this allows to sample nodes from a distributed set of nodes efficiently and offers good resilience, it does not allow to efficiently discover nodes with specific capabilities within this node set. Our research log post explains this in more detail. Adding efficient (but still preserving resilience) capability discovery to discv5 is ongoing research. A paper on this has been completed, but the Ethereum discv5 specification has yet to be updated. When the new capability discovery is available, this document will be updated with a specification of the second discovery method. The transition to the second method will be seamless and fully backwards compatible because nodes can still advertise and discover shard memberships in ENRs.
Security/Privacy Considerations
See 45/WAKU2-ADVERSARIAL-MODELS, especially the parts on k-anonymity. We will add more on security considerations in future versions of this document.
Receiver Anonymity
The strength of receiver anonymity, i.e. topic receiver unlinkablity,
depends on the number of content topics (k) that get mapped onto a single pubsub topic (shard).
For named and static sharding this responsibility is at the app protocol layer.
Copyright
Copyright and related rights waived via CC0.