Let's talk about FoundationDB and foundation

Let's talk about FoundationDB and foundation
FoundationDB

Apple acquired FoundationDB1, and the latter immediately changed from open source to closed source. To be honest, I have never noticed such a DBMS product 2 before. From the official website, some of its design concepts are more practical. After all, Distributed DBMS has many problems that are hard to solve, it is unrealistic to achieve the same level as traditional RDBMS such as Oracle in the short term. For example, it focuses on core features and makes a trade-off between 3 and 4:

1. Scalability: adopts the Shared-nothing architecture.
2. ACID, supporting serializable isolation levels.
3. Fault Tolerance is not just to avoid spof.
4. Copy, multi-copy storage.
5. Data Model, ordered KV storage.
6. Change notification. The client can subscribe to changes to the record.
7. Atomic operation to avoid Read-modify-write?
8. OLTP, suitable for small reads and writes. OLTP is restricted. For example, it does not support modifying too many transactions.
9. OLAP, supports range scanning of a large number of data. OLAP is not equal to a large amount of data scanning.
10. The core does not support query languages, but only APIs.
11. Concurrency Control adopts MVCC and Optimistic Concurrency Control.
12. The thread model is not a connection or thread.
13. Applications can be interactive and independent from stored procedures.
14. Snapshots. Are you sure you want to maintain consistency.

However, the actual practice 5 and the complete limit 6 cannot be accurately determined from these materials. In Quora, I found a question and answer 7 in 2013. Let's take a look at some of the trade-off about this system. There are other materials:

1. Http://2014.nosql-matters.org/cgn/wp-content/uploads/2014/05/Jennifer-Rullmann-NoSQL-and-ACID.pdf
2. Http://www.methodsandtools.com/archive/acidnosqldatabase.php
   As Gilbert and Lynch note, their proof relies on an asynchronous model in which "there is no clock, and nodes must make decisions based only on the messages received Ed and local computation. "Mahajan revisit Gilbert and Lynch's work and strengthen its result by employing an asynchronous model allowing local clocks, but no clock globally visible to all nodes.... Real Time Causal (RTC) consistency that is strictly stronger than eventual consistency. Percolator, External Store, and, most recently, Spanner.

Many people prefer NoSQL and NewSQL compared with traditional RDBMS. To a large extent, this comparison is not easy, because there are huge differences between the requirements they target and the design trade-offs. For example:

1. The amount of data processed. The traditional OLTP data is not too big, and the centralized system can handle it well. However, once the data is distributed in different IDCs, the situation is very different.
2. Reliability requirements. The processing methods and capabilities of centralized and distributed systems are obviously not the same. Paxos came into being after RDBMS became mature.
3. Data Model. New databases rarely support relational models, such as various constraints and derived indexes and sequences.
4. Query Language. API, SQL, and even XQuery?
5. Data consistency. The ACID of the centralized system is obviously superior.
6. Important features. Stored Procedures, views, triggers, and data warehouses can easily list features that are hard to achieve by a bunch of new systems.

One size does not fit all.

Footnotes: 1

Https://foundationdb.com/

2

Https://en.wikipedia.org/wiki/FoundationDB

3

Https://foundationdb.com/key-value-store/features

4

Https://foundationdb.com/key-value-store/white-papers/anti-features

5

Https://foundationdb.com/files/Architecture.pdf

6

Https://foundationdb.com/files/FoundationDB-KeyValueStore-Spec-Sheet-3.0.x.pdf

7

Https://www.quora.com/What-is-FoundationDBs-secret-sauce