Video: MySQL Cluster Overview by Bill Papp


At the December 2010 Boston MySQL User Group meeting a few days ago, Oracle/MySQL sales engineer Bill Papp gave a presentation on 'An Overview of MySQL Cluster'.

Slides are online at

The video is on YouTube at

Here is the description Bill gives of this presentation:

Now, more than ever, databases have to scale linearly to handle high spikes in usage, deliver 24 x 7 continuous availability and low latency. At the same time, many workloads need to maintain the data consistency delivered by transactional integrity.

MySQL Cluster is the industry's leading real-time, write-scalable transactional database, combining 99.999% availability with the low TCO of open source. With a distributed shared-nothing architecture and no single point of failure, MySQL Cluster can scale linearly to meet the unprecedented demands of next generation web, e-commerce, and telecommunications / network services.

Typical use cases in web services include:

* User profile management for service authentication and authorization

* eCommerce (customer and order data, payment processing, product inventory tracking, etc.)

* Web Session Management

* Social Networks (user profiles, authentication, real-time status updat! es, etc.)

* On-Line Content Delivery (i.e. user data, content catalogs, subscription processing & management)

* Near Real-Time Data Analysis (using Geographic Replication to update MyISAM, InnoDB, etc. tables with real-time transactional updates from MySQL Cluster) 



Answers to questions some unanswered during presentation:(that aren't long enough to understand wo/ listening to presentation)

  1. When a starting node synchronizes with it's peer, it does row-copying, but copies more than one row at a time. Each node keeps a time-stamp per record, so it knows which rows needs to be copied to starting node.The time stamp is actually the global checkpoint identifier, so there is no dependency on OS synchronized timestamps (e.g ntp).The copy procedure only utilizes only row-locks, so that application can read/write to tables while node is starting.
  2. Are selects prohibited during repartioning: No :-)
  3. Re scatter/gather (terminology from presentation): For primary key/unique lookups mysqld nodes knows about where data is.For index/table scans, mysqld nodes sometimes know about where data is, depending on table and query, but sometimes it needs to ask all data-nodes.
  4. Checkpointing: Each datanode writes a REDO log, that is syncronized to disk at configurable interval. (compare innodb_flush_log_at_trx_commit). To prevent the REDO log from growing endlessly, a "local checkpoint" is also automatically performed when needed.
  5. A ndbapi node (such as mysqld) is connected to all data-nodesA connector-j connection can be set to round-robin (with failover) over several mysqld's.

Thu, 12/16/2010 - 13:18


The content of this field is kept private and will not be shown publicly.
  • Web page addresses and e-mail addresses turn into links automatically.
  • Allowed HTML tags: <a> <em> <strong> <cite> <code> <ul> <ol> <li> <dl> <dt> <dd>
  • Lines and paragraphs break automatically.

More information about formatting options

By submitting this form, you accept the Mollom privacy policy.
Website by Digital Loom