Newsql: finding the balance between nosql and SQL)

[It168]Nosql was originally abbreviated as not SQL, and has now evolved into not only SQLBecause developers have realized that, in fact, those who prefer SQL and those who have a lot of SQL experience are moreProgramThe cost of employee employment is lower, and such a person is more likely to be recruited.

In fact, it is an error that nosql does not use SQL. SQL is not used because of poor performance, but because the relational product architecture is not suitable for some types of tasks. However, without these types of tasks, it is easy to fall into the trap of thinking that SQL is equivalent to relational technology. The key aspect of relational architecture is the separation of physical instances from logical implementation. However, this is true for most relational products that contain relational storage and relational access layers. However, this example does not need to demonstrate the differences between the relational and XML-based data storage engines of DB2.

In any case,Nosql developers have realized their errors and added SQL Functions. But what I really want to discuss is the so-calledNewsql Database. These databases have been designed to contain SQL interfaces (instead of those added later) from the very beginning, and do not require the relational storage engine at the lower layer.

The latest newsql database is voltdb, another file of Michael Stonebraker. It is a standard relational database, but removes unnecessary small components that have accumulated all links for more than 40 years, making it simpler and more effective than traditional versions, because of this, it is more efficient than commercial databases and reduces its coverage, as is true for all newsql databases. Another newsql database is from xeround, a cloud-based database company that provides flexible scaling and nosql roots. Like all other newsql databases, the company's newsql database focuses mainly on transaction processing. In addition to its inherent capabilities, the database of xeround is different from the other, and it looks like MySQL, this makes it easy to migrate existing mysql users to the cloud.

There are also two interesting newsql databases: nuodb (formerly nimbusdb) and justonedb. At present, I have not yet obtained the test procedures for both, because nuodb is currently in beta stage, and the justonedb project has just started.

Nuodb uses a distributed object architecture (like many nosql databases) in an end-to-end environment similar to BitTorrent ). When a record is updated, it is appended to the existing data instead of replacing it. Therefore, all historical data in the database can be seen. This architecture involves the use of transaction nodes and archive nodes. The former uses memory, and the latter uses key/value storage to maintain data. Because multiple archiving nodes can keep the same data that has not been backed up, it is unnecessary to copy data for high availability and block data. This sounds like a standard nosql database, but the biggest difference is that the nuodb database is specially designed to support SQL, it fully supports acid (Atomicity-atomicity, consistency-consistency, independence-isolation, and persistence-Durability ).

Justonedb is another matter. It also fully supports acid (such as the hybrid l nosql database) and is designed to support SQL. Its Environment looks like PostgreSQL and runs on Heroku's cloud. However, it uses a different storage architecture called tunnel storage. I cannot explain what tunnel storage means (I can only say that it is neither column-based nor key-value-based), but I can be sure that this technology has some benefits, its features include full support for join and no need for indexes or chunks. Interestingly, it may support Query Processing (for example, you may never need to scan the entire table) and OLTP, although its company focuses on the latter. In addition, like nuodb, justonedb always appends data and never deletes their source data. However, it differs from its competitors in that it focuses on scalling up rather than scalling out, at least for the moment, that is, how to scale on a server rather than cross-server. The company's point of view (A not unreasonable point of view) is that under ideal circumstances, people generally want to scale the better, that is, why the company focuses on scalling up rather than scalling out. The internal operation of the TPC-H benchmark demonstrates a significant improvement in performance compared to the standard PostgreSQL orders (100x +.

If you like MongoDB, Cassandra, or hadoop, nosql is very interesting, but these databases are not experienced by most people. On the other hand, newsql is much more experienced for most people. Many people spend a lot of time on it, And newsql database is more interesting and worth testing.