EinsteinDB is a hybrid PMem-DRAM KV high performance in-memory database which aims to achieve high performance for both OLTP and OLAP workload.

Once upon a time, there was a boy named EinstAI. He had learned to use indexes. He used learned indexes because they were small and fast.

In the end, EinsteinDB is a good choice for a second grader because it is small and fast.

We can upgrade eventually consistent data stores to provide causal consistency.

MilevaDB resembles EinsteinDB, though on its own terms: both stores business data but they differ from each other through different perspectives

MilevaDB's optimistic transaction guarantees that the database will remain write-protected as long as there are no conflicting read transactions.

"In memory" doesn't mean just relying directly upon purely stateless connection brokering made possible by virtual machines such as Varnish cache servers
MilevaDB acts as a tiny transactional key/value store with built-in ACID compliance and SQL queries, implemented in Minkowski Spacetime using standard RDF semantics
EinsteinDB is an Ingres-like database that doesn’t use traditional file based storage; instead, it uses vectorized column oriented virtual memory (VCM)
Since there are no buffers involved with VCM stores, table sizes can be much smaller than those of even pure SQL databases on disk or large RAM disks.
tight control over cache management
EinsteinDB also outperforms all Hadoop products by orders of magnitude when talking about reads only workloads like OLTP

A Causet is a tiered LSM-tree; Beyond Key-Value. In a SQL database, you have weak and strong relations, but there’s no way to tell which is which. In a non-SQL database, you have weak and strong relations, but you can tell which is which.

We present a causal consistency shim layer that upgrades eventually consistent data stores to provide convergent causal consistency. We've named the Causes in lieu of Causal Set Theory.
VioletaBFT is a distributed consensus algorithm that is more efficient than the classic Paxos. It is more efficient because it does not require a designated leader.

It is more efficient because it does not require a pre-ordered queue. It is more efficient because it does not require a quorum of replicas to agree on the order of commands.

Powered by GPT3 via EinstAI: learn from unlabeled data, and achieve state of the art performance

EinstAI uses a pruning of EinsteinDB Key-Value Hybrid Index stores, large swathes, ample amounts of unlabeled data to train language models, then EinstAI uses these models as features for supervised learning tasks like question answering or machine translation.

Read committed isolation level

If you want to be sure that a transaction does not see uncommitted data, use the Serializable isolation level. If you want to be sure that a transaction sees the effects of previous updates, use the Repeatable Read isolation level.

We built a high-level engine that is competitive with low-level engines.

We built a high-level embedded tuplestore with graph processing engine that is competitive with the best-of-breed low-level engine.

1
SQL databases are hard to get right.

EinsteinDB offers the flexibility and scalability of NoSQL architectures with the power of ACID transactions (a.k.a. NewSQL).

2
Fail fast and recover fast.

We have built a geo-replicated EinsteinDB cluster with 862 processes, 292 TB data, and 464 SSD disks.

3
commit latency is lower than the WAN latency

The cluster has an average of 390K read operations per second, 138K write operations per second, and 1.467M keys read per second.

4
Defeating latency

both throughput and latency scale linearly with the number of Proxies; the throughput increases from 730K ops/sec to 2.3M ops/sec while the latency decreases from 0.8 ms to 0.4 ms

DRAG-AND-DROP BUILDER

EinsteinDB & MilevaDB use optimistic concurrency control by default, whereas MySQL applies pessimistic concurrency control.

In the optimistic transaction model, transactions might fail to be committed because of write–write conflict in heavy contention scenarios.

internal retry mechanism
Using optimistic concurrency control, a transaction can be aborted by the EinsteinDB server even if it detects that there is conflicting change.
Optimistic vs Pessimistic Tens.
concurrent transactions frequently modify the same rows (a conflict), optimistic transactions may perform worse than Pessimistic Transactions.
handle concurrent write-type SQL statements in 2PC mode
write operations are atomic (no read/write conflicts occur), ensuring consistency across asynchronous reads from different clients
EinsteinDB is an open-source, multi-model database that supports graph, document, key/value, and columnar data models. It can be deployed on-premises or in the cloud.

EinstAI is a way to make a database automatically do the work for you.

EinstAI provides EinsteinDB with a Learning-based Database Configuration Engine

EinstAI provides EinsteinDB with a Learning-based Database Configuration Engine
The Learning-based Database Configuration Engine automates database configurations, including knob tuning, index advisor, materialized view advisor, SQL rewriter, and database partition
EinstAI allows EinsteinDB to treat database decision trees as sequential problems that can be resolved, accessed, and annotated in any order.
Only the previous state of the database is used as input. A "memory-optimized" version of the famous "Binary Search Algorithm" (BSA) with a fixed number of steps
artificial intelligence can be used to characterize new features for a query using deep learning
Once upon a time, there was a boy named EinstAI. He had learned to use indexes. He used learned indexes because they were small and fast.