Schooner adds asynch replication to MySQL appliance
Remote disaster recovery
Schooner Information Technology, which makes a line of virtual machine appliances that goose the performance of MySQL databases and Memcached web cache programs by using CPU threads, cores and solid state disks in modern x86-based servers, has tweaked its MySQL appliance to give it better disaster recovery.
Schooner came out of stealth back in April 2009 with a set of hardware appliances based on IBM's System X servers equipped with Intel flash storage, which delivers about a six to eight factro performance boost over servers running off-the-shelf versions of MySQL and Memcached (and not using flash storage).
But the hardware was almost irrelevant. What was key is a set of software originally called the Data Fabric API and now called the Schooner Operating Environment, which creates very efficient and thread-aware userspace for Schooner's own blackbox, reverse-engineered, Memecached clone or Oracle's MySQL Enterprise Edition database (which it licenses from Oracle) on top of Red Hat Enterprise Linux 5.4 and 5.5.
In February this year, Schooner ditched the IBM hardware and packaged up these two products as virtual machine appliances, because flash memory was now common and customers wanted to use their servers, not the single box sold by IBM. They are now called SchoonerSQL and Membrain.
In April of this year, Schooner beefed up its version of the MySQL database (which includes a modified version of the InnoDB database engine that Oracle also controls) with a feature called Active Cluster, which offers synchronous replication across clustered MySQL nodes. As Jerry Rudisin, Schooner's CEO, told El Reg back in the spring, this is better than the kludgey asynchronous replication, which has a production database backing up to one master and then propagating changes to slave nodes, that is part of Oracle's MySQL stack. Oracle also sells MySQL Cluster, which does synchronous replication, but that feature does not work with the InnoDB engine, which is popular among MySQL shops.
After taking some shots at Oracle for its asynchronous replication for MySQL earlier this year, Rudisin tells El Reg that customers said they needed this capability not within the data center or the metro distances (60 miles or less) that the Active Cluster feature can work over, but for longer hauls to provide true geographically remote disaster recovery. You can't fight the speed of light, so you can't do synchronous replication beyond that 60 mile distance. But, says Rudisin, you can do a better job speeding up asynchronous replication.
With the new features in the MySQL variant, which is now called SchoonerSQL v5.1, the updating of remote master and slave databases has been parallelized to make use of all the cores and threads in the machine, rather than running in series on one thread. This radically reduces the time it takes to do a failover from a production database in one location, and a backup database over a wide area network. This seemingly minor change means that a failover and recovery from a remote data center can take place in seconds to minutes, instead of minutes to hours.
In addition to parallelizing the updates to the MySQL 5.1 Enterprise Edition database, the asynchronous replication feature of the SchoonerSQL database knows that it can do reads almost instantly, and queues up writes until the database is updated with transactions that were in flight when the spit hit the fan. This also helps speed up recovery times.
Schooner is offering SchoonerSQL 5.1 under a free trial, which you can participate in here. SchoonerSQL 5.1 will be available in mid-November and will include both the asynch and synch replication features bundled into the price. The database costs $9,500 per two-sockets in a server and will be certified on recent RHEL and CentOS releases of Linux as well as on mainstream IBM, Dell, and Hewlett-Packard x86 servers with popular flash storage drives. ®