Distributed data routing protocol for finding the best available routes across the Internet.
Organises the Internet into coherent, best possible performance connections between users & applications. DARP LIVE NETWORK
The Internet today is a massive scale network that reaches every corner of the world. It is driven by multitudes of protocols, some of which are dedicated to efficiently routing data from source to destination.
Conventional technologies in the dynamically routed Internet field primarily operate at the Network Layer (Layer 3) by selecting the best path for traffic based on the shortest path between the source and the destination node, which often does not minimize data transmission latencies or bandwidth. Conventional technologies also do not take into account network degradation or network degradation events propagate slowly.
Even though the Internet is an outstanding achievement and it is remarkably reliable for how diverse technologically it is, there are still ample opportunities to make it more efficient and more reliable. At Syntropy we have discovered such a way.
Current routing technology
There are in fact a lot of Internet routing protocols, such as RIP, BGP, IGRP/EIGRP, OSPF, etc. that serve different purposes and/or allow different subnetworks (aka Autonomous Systems) to communicate with one another. The backbone of the Internet that enables connectivity consists mainly of routers. The main task of each of the protocols is to find the shortest paths from the source to the destination.
Each of the protocols accomplishes this task differently. The common feature is that every router maintains a global routing table in one form or another and uses it to find paths for network packets passing through the router. Each protocol builds this routing table differently and uses different metrics for path discovery.
There are at least several problems with the current Internet setup:
- Different ISPs may use different routing protocols;
- Different ISPs may configure their protocols differently or even use less capable hardware;
- ISPs are optimizing for the lowest cost, thus they are more inclined to route via cheaper and less performant paths;
- Interoperability between different protocols is limited;
- The end-user has no control over how their traffic is being routed;
- The least-hops path might not be the most reliable (current routing protocols do not optimize for jitter and packet loss and rarely optimize for latency).
All this contributes to the heterogeneity of the Internet making efficient routing incredibly difficult.
The very first version of DARP allowed regular servers spread around the world to connect to an overlay network with a full mesh topology. The nodes would exchange latency information (aka Pulses) in such a way that every node in the network contained the full latency matrix. Any node could use this global matrix to find the best paths with single or multiple relays.
Proto DARP is using One Way Latency (OWL) instead of Round Trip Time. This innovation allowed us to achieve at least these benefits:
- We could reduce the traffic needed to measure the latency within the network;
- We would remove the need for clock synchronization and could still find the best paths;
- We could measure latency as well as jitter and packet loss that is derived from the packets exchanged due to the protocol.
Using this latency matrix, the protocol can find optimal paths via other overlay network nodes. This information is then used to establish secure Wireguard tunnels that allow for secure programmable paths through the DARP network. And since the latency matrix is frequently updated, these paths may be reconfigured on the fly, if a better path is discovered or an outage is detected. This allows dynamic routing of secure tunnels around congestions, outages, or otherwise slow or unreliable connections.
The average latency savings found in Proto DARP network was 18.5 ms or 13.2% as compared to the direct path. Data analysis indicates that a significant latency improvement (>10 ms) is possible in 42.8% of paths.
- Proto DARP essentially stores and maintains a global routing table on every node;
- This node essentially acts as a central source of truth for the best paths;
- The topology guarantees every packet is sent via the best possible path within the network, therefore adding the ultimate performance layer to any private and small-medium size network.
While Proto DARP is very effective at measuring inefficiencies in a small to medium size network, to enable all internet resources to be amassed by the protocol we’ve added decentralization and scalability features to the protocol.
Combining the fact that full mesh is not required to achieve optimal results in a large network and the protocol can be scaled to any number of nodes, we’re launching the largest public network on the internet under decentralized
DARP (dDARP) protocol, is:
- Decentralized – meaning that there is no single authority in the network;
- Trustless – nodes do not rely on the reported latencies for path discovery;
- Sparse and optimal – nodes establish and maintain connections with only a handful of neighbours that are enough for the protocol to find the best paths.
Unlike centralized data optimization protocols, DARP does not rely on other nodes providing parts of the latency matrix and thus isn’t susceptible to latency manipulation by malicious nodes. dDARP architecture does not require any genesis or root node that orchestrates Pulse Group formation.
We have discovered that the full mesh network is not required in a large network to find the best paths in the majority of cases due to the nature of underperforming routes.
Every node in the network operates using the same rules. For two nodes to have a stable connection they must both agree on the connection parameters. If these parameters satisfy the predetermined criteria, the connection continues measuring the network state.
If an attacker compromises or attacks a part of the network, it is naturally replaced by other healthy nodes. Therefore the network is self-healing.
In dDARP every node can also share information about their neighbours, however, this information is mostly used to determine the most optimal connection pattern as well as predict outages and detect malicious nodes.
In order for a centralized data optimization algorithm to function effectively, it must maintain the full latency matrix. This requires every node to measure the latencies of every other node. This makes the size of the latency measurement packets grow linearly with the number of nodes; whereas, the size of the latency matrix is growing quadratically with the number of nodes. This essentially imposes a practical limit on the size of the network.
With dDARP, no node relies on other nodes for network state. Rather, when a source node ‘A’ needs to find the best paths to a destination node ‘B’ , it sends packets to its neighbours (see Figure 3-1), which in turn send packets to their neighbours (Fig. 3-2). Eventually, some of those packets reach the destination node (Fig. 3-3).
Every packet stores the departure timestamp as well as timestamps with accompanying signatures from each intermediate node. The destination node only has to calculate the One Way Latency, jitter, and other metrics for each packet and sort those packets by the weighted sum of OWL, jitter, etc., thus obtaining a ranked list of best paths.
Datapoints 1: dDARP optimization
- 44% of the routes can be improved by at least 10 ms.
- 37.5% of the routes can be optimized with respect to latency or jitter or both.
This means that the best paths are determined mainly by the quality of connectivity itself. And since every intermediate node signs its data in the path packet, it is not possible for malicious actors to alter the IDs of the intermediary nodes stored in the winning path packets.
Sparse and scalable network
Thanks to intelligent dDARP algorithms every node has to maintain only a handful of connections to other nodes (around 50-100), thus greatly reducing the burden imposed on the network by the measurement traffic. This in turn significantly reduces the network costs for the DARP node operators.
Datapoints 2: DARP efficiency
- At least 80% of full-mesh paths are also discovered by the decentralized DARP version (topology of 300 nodes).
- Decentralized DARP uses at least 66% less traffic for measurements as compared to the Proto DARP version.
Additionally, since the network is sparse, traffic usage does not grow when more and more nodes are joining the network.