In the world of relational databases, MySQL and Oracle RAC (Real Application Cluster) own a significant segment of the market. Oracle RAC owns the majority of the enterprise market, while MySQL gained significant popularity amongst many of the Web 2.0 sites.
Today, both of those databases are owned by Oracle (MySQL was acquired by Sun which was later acquired by Oracle).
The following diagrams show the enterprise database marketshare covered by Gartner and Cloud Database market share covered by Jalistic - a Java PaaS provider.
The acquisition of MySQL by Oracle raised concern over the future of the project due to the inherent conflict of interest between its two database products. Oracle RAC is the company’s main “cash cow” product, while MySQL competes for the same audience.
Shortly after Oracle’s acquisition of MySQL, the open source database was forked by one of its original founders into a new project named MariaDB. MariaDB was established to provide an alternative development and support option to MySQL and is now becoming the default database of choice of RedHat.
MySQL vs Oracle RAC Clustering Models
The two databases take a fairly different approach to scalability.
Oracle RAC is based on a shared storage model. With Oracle, the data is broken into strips that are spread across multiple devices, and multiple database servers operate concurrently and in sync over the (shared) data.
MySQL, on the other hand, does not use a shared data model. With MySQL, a given data set is managed by a single server (a model called shared nothing). With MySQL, scaling and High Availability (HA) is achieved by managing copies of the data. As only a single machine can update the data, this mode can only scale-up by adding more capacity to the machine that owns the database. As machines have limits to capacity yet must keep up with large amounts of data or many users, the database needs to be broken into several independent databases (a process called sharding). However, sharding is a complex process, is not dynamic and requires assumptions on data distribution, data growth and the queries. For these reasons, sharding is not possible with many applications. In particular, a cloud deployment is expected to be elastic to dynamically support changing needs and user behaviors.
For High Availability, the MySQL shared nothing approach uses Primary/Backup model with a single master and a backup node (called slave) that manage a copy of the data. Each update to the master requires that the same update will be executed on the slave. The primary and slave nodes require some handshake protocol to determine who is the master and sync the changes of the data. The master node performs updates/writes to the persistent file-system and the level of High Availability is set by the DBA, who decides if the slave can lag behind or needs to confirm the duplicate updates within each transaction.
For scaling, this model can use the slave as a read replica (or make additional copies of the data), a method called horizontal scaling in which read requests are spread across the different copies of the data. (However, all the writes need to go to the master and then be reflected on the replicas/slaves.)
Relational Databases on the Cloud
The high-end position of Oracle RAC, the low-cost and open source nature of MySQL, along with the adoption of the cloud as the infrastructure platform led to a vastly different method of deployment of databases in the cloud.
Oracle RAC took an approach similar to the old mainframe: to produce a pre-engineered appliance (called Exadata) that comes with the software and hardware integrated. That approach was specifically aimed at existing customers of Oracle RAC who needed a quick resolution to their scalability needs without redesigning their applications. Plugging a more optimized stack helped to push the scalability bar without changing the existing applications that rely on the database.
Amazon launched RDS, which is basically an on-demand version of the MySQL database. This approach fits nicely with the existing users of MySQL who are looking for a more affordable way to run their database in the cloud.
The main limitation of the Amazon RDS approach is that it inherits the limitations of MySQL and thus provides a limited and relatively complex read-scalability and doesn't provide a good solution for write-scalability other than the scale-up approach.
ScaleDB - a NewSQL Alternative to Amazon RDS and Oracle RAC
ScaleDB is a NewSQL database that takes an approach similar to the read/write scalability model of Oracle RAC and implements it as a platform that transparently supports the MySQL (or MariaDB) database. As a result, existing MySQL/MariaDB applications can leverage the platform without any changes - they use MySQL or MariaDB as the front end engine which connects to the ScaleDB platform that provides a new and more concurrent database and storage services. This approach makes it possible to run multiple MySQL or MariaDB server instances against a distributed storage in a shared data model (similar to Oracle RAC).
The diagram below shows in high level how ScaleDB works.
Each Database Node runs a standard MariaDB database instance with ScaleDB plugged as the database engine and as an alternative storage and index device.
Scaling of the data and the index is done by processing requests concurrently from multiple database instances and leveraging the ScaleDB distributed storage tier (at the storage/file-system level) where data and index elements are spread evenly across multiple storage nodes of the cluster. Read more on how ScaleDB works.
ScaleDB vs Other NewSQL Databases
Most of the NewSQL databases are based on MySQL as the engine behind their implementation. The main difference between many NewSQLs and ScaleDB is that NewSQL databases brings the NoSQL approach into the SQL world, where ScaleDB takes the Oracle RAC shared storage approach to scale the database.
ScaleDB can deliver write/read scale while offering close to 100% compatibility, whereas in many of the alternative NewSQL approaches, scaling would often involve significant changes in the data model and queries semantics.
Benchmark - ScaleDB vs Amazon RDS
To demonstrate the difference between a standard cloud deployment of MySQL and a ScaleDB deployment and in order to find whether the ScaleDB approach can live up to its promise we conducted a benchmark comparing Amazon RDS scalability for write/read workloads with that of ScaleDB. We tested a dataset that does not fit to the memory (RAM) of a single machine and used the largest machines offered by Amazon. We required that scaling would be dynamic and that all types of queries would be supported. These requirements made sharding a no-go option.
The benchmark is based on the popular YCSB - Yahoo Benchmark as the benchmarking framework.
The results of that benchmark are shown in the graphs below.
Both illustrate a relatively flat scalability with Amazon RDS and a close to linear scalability on the ScaleDB front.
Benchmark Framework - YCSB - Yahoo Benchmark
Cloud environment: Amazon
Machine Type: Extra Large
Data Element Size (per row) - 1k
Data Capacity: 50GB
Zones - Single zone.
RDS was set with 1000 provisional ios
ScaleDB cluster setup - 2 database nodes, 2 data volumes (4 machines -data is striped over 2 machines and each machine had a mirror).
ScaleDB MySQL engine - MariaDB
Running ScaleDB on OpenStack and Other Clouds with Cloudify
The thing that got me excited about the project is that it serves as a perfect fit for many of our enterprise customers who are using Cloudify for moving their existing applications to their private cloud without code change. Those customers are looking for a simple way to scale their applications and many of them run today on Oracle RAC for that purpose.
The move of enterprises to a cloud-based environment also drives the demand for a more affordable approach to handle their database scaling, which is provided by ScaleDB.
On the other hand, setting up a database cluster can be a fairly complex task.
By creating a Cloudify recipe for ScaleDB, we could remove a large part of that complexity and set up an entire ScaleDB database and storage environment through a single click.
In this way we can run ScaleDB on demand as we would with RDS and scale on both read and write as with Oracle Exadata, only in a more affordable fashion.
1. Cloudify Recipe for running Scale DB