What's BangDB?
BangDB is a converged multi-model and multi-flavor database platform for building modern applications which require to handle different kinds of data and requirements in a single space. BangDB have been created from ground up, written from scratch to align with the data trend fully and make it ready for the future needs as well. Some of the key design goals of the database were following before the development phase.
Key design goals
- Align with the upcoming data trend and build a future ready database for creating solutions.
- Store, query, link, join different kinds of data in the same database in a very simple manner.
- Stream and event processing as part of the ingestion construct for continuous intelligence.
- Graph should be linked with the different aspect of the data processing to leverage context.
- AI should be integrated with data such that the implicit and autoML is part of the operation.
- High performance, full read/write concurrency, best hardware utilization, avoid resource bloating.
- Break in-memory only concept - Durability and overflow to disk with predictable performance.
- Event driven design for handling tens of thousands of concurrent connections at the single server.
- Database should be deployed within micro-processors, commodity hardware or in the cloud.
- Provide a no-code platform to the users on cloud to let them focus on use cases ad their business.
And need for several core database level Features and more…
- Need to have buffer pool and page cache - to get total control on every single byte.
- Should leverage the servers infra fully - read and write both fully concurrent.
- Transaction (OCC) for data consistency - for selected data set
- Write ahead logging for data durability, crash recovery and backup.
- Auto balancing and managed cluster with cluster aware replica
- Every instance to be TCP and HTTP(S) server - no proxy or another layer in front.
- Consistent access model across different deployment and all connected - embedded, cluster
- Slotted aggregations, stats for ZERO post processing for stream related processing and queries
- All the different elements should be able to run in a single process - This is a must have.
- Connect to database by any means - clients, cli, REST api
The above goals quite comprehensive and required a fresh thought, innovative approach and shift in the way NoSQL's were designed and implemented.
Need for convergence
It becomes obvious from the problem descriptions and challenges that the problem space, due to the current data trend, is converging too many requirements in the singular space. For example, dealing with multiple kinds of data, stream processing with graph linking and storage, AI as part of query and etl etc. Therefore, to counter this convergence at the problem space, we must also converge at the solution level. And if we can do this, then it will straight forward task to deal with the emerging and current use cases properly.
Problem space has a convergence of the varied requirements at the core.
We must converge the solution space to counter the same.
Converge at system level, data level, deployment level. Converge to create a unit of compute which scales linearly.
![Converged Database - BangDB](/_next/image?url=%2Fassets%2Fabout%2F14.png&w=750&q=75)
Benefits of convergence
![Benefits of convergence - BangDB](/_next/image?url=%2Fassets%2Fabout%2F15.png&w=1080&q=75)
Architectural shift
The most important decision was to move away from traditional 3-tier or n-tier model towards creating a unit of compute by converging the elements and let that unit of compute scale linearly. One of the biggest issues with the model where we split the system into multiple parts and distribute those parts over the network. Finally combine the results to throw the output to the client.
This is typical micro-service-oriented approach, which doesn't fit well with a system like this. This has several issues which are discussed above in detail. But apart from the mentioned issues like it's a costly affair, resource intensive approach, takes too much time to build and ship, app layer becomes brittle leading to huge loss of team's productivity etc., there are core issues as well with this approach.
Siloed (semi-siloed) architecture forces too many network hops along with too many copies of data.
- Same data gets copied multiple times within and across silos.
- Several network hops across the layers to process a single event/data.
![Architectural shift - BangDB](/_next/image?url=%2Fassets%2Fabout%2F16.png&w=1080&q=75)
With convergence, we avoid these issues as well. Once data reaches a node, all computations happen there. This also allows us to handle way more client's request for the same configuration, which reduces cost, complexity and improves performance. Managing such converged system is easy, and it doesn't even require a dedicated set of resources for the same.
Event Driven (SEDA)
Staged Event Driven Architecture (SEDA) is at the core to meet big part of the architectural constraint. Basically, we deal with FSM (finite state machine), within the unit of compute for managing the sub systems with boundaries. The entire structuring of the stages within the node is also configurable. We can create different structures for stages and how they would work together base don requirements. Even the worker level details are configurable. Which means each node, types could be configured for our need. BangDB pre-configures all these which works best in most of the cases.
![Event Driven (SEDA) - BangDB](/_next/image?url=%2Fassets%2Fabout%2F17.png&w=1080&q=75)
There are more than 100 configurations that could be tweaked as needed. But 80-90% are not required to be changed most of the times as they work best in most of regular cases.
Deployment flavors
BangDB comes in different flavors in which it could be deployed or used. BangDB can be deployed within a micro-processor. This is required from IOT perspective. To truly manage edge computing, the system should be within devices as well.
Then BangDB cluster can be deployed in client/server models for majority of the use case with replicas attached. BangDB can be distributed further by creating a unit cluster consisting of Core database, training, resource server and user service. This suits most of the use cases and requirements.
Finally, BangDB can be deployed in P2P mode as well, where there is no master or slave. This is auto managed distributed ring where node joins or leaves at will, where distribution is managed on its own. BangDB is going to launch this version by Sep 2024. More on this when we launch.
![Deployment flavors - BangDB](/_next/image?url=%2Fassets%2Fabout%2F18.png&w=1080&q=75)