A load-balancing startup is taking the balancing act to databases, and its managed to boot MySQL out of another startup as its first bragging rights.
The startup, Xkoto, launched at the end of December with its Database Load Balancer. It currently works with IBMs recently launched freebie version of its database, DB2 Express-C.
Its launch coincided with that of another startup, Savvica, with its free e-learning service, which was running on a MySQL database when it first rolled out.
According to Xkoto CEO Albert Lee, MySQL worked fine for the e-learning startups first two months. But after Savvica got into the marketing game, including a write-up in an online thread called Tech Crunch that could be considered the equivalent of the Super Bowl in the companys market space, usage shot through the roof.
Because Savvicas e-learning service entails heavy read/write on the database—as it allows users to create e-learning content, upload the content for delivery to students, grade papers and the like—the servers were slammed.
“Usage spiked, and the server fell over,” Lee said.
Hence, as Xkoto announced on Feb. 22, Savvica embarked on and has just now completed a migration off MySQL to a configuration using Xkotos Database Load Balancer and DB2 Express-C.
Lee said the situation at Savvica looked the same as at any company where businesses must ensure that servers can handle whatever surge in traffic comes their way. To do so, businesses typically go to a big-box architecture, making sure they have server capacity for peak load.
Alternatively, they go the utility route, paying for capacity on the fly. Such a model is constrained by the type of architecture in use, however.
Xkoto, in contrast, is taking a horizontal rather than a vertical approach to architecture.
“Were saying youre running out of capacity, throw another low-cost commodity server on the cluster horizontally,” Lee said. “We can spread the traffic around.”
Web load balancers and application load balancers are already a known commodity at most data centers. What Xkoto wants to do is extend the concept in customers minds to encompass load balancing in the database.
“That analogy were extending down to the data tier and doing what the Ciscos and F5s do above us in the stack,” Lee said.
Key for the Database Load Balancer is to fit between the application and the data clusters and to control cluster load from that vantage point.
Page 2
Xkoto Chief Technology Officer Ariff Kassam explained that the Database Load Balancer works by way of having each node maintain an exact replica of the database. To do that, any write transactions have to be broadcast to all nodes. The technology performs load balancings on read transactions, spreading reads across the cluster to get the speed up.
“Thats how we get scale up—by parsing and distributing reads across the cluster,” Kassam said. “We dont reach everybody with reads. We just send to” the node with the most capacity at a given time, he said.
According to Lee, Xkotos technology is achieving 85 percent scalability each time a low-end, commodity server is added to a cluster, as the company demonstrated in a benchmark performed at IBMs request.
That compares to adding boxes to a cluster without a load balancer, wherein the database cant use the added box unless an enterprise is using software from a vendor, such as IBMs DPF (Data Partitioning Facility).
The typical cluster setup involves two nodes with software that enables failover. Oracle has its Data Guard software, while Sybase has Replication Server, each of which enables two-node replication.
One node serves up application requests, while the other node sits idle, waiting until failover occurs. But such passive waiting doesnt increase performance; its only there in case of a disaster scenario.
The differentiator with Xkotos technology is its “active-active” status, Lee said. “Every node in the cluster is actively serving transactions all the time,” he said. “Its not sitting there for the once in a blue moon when the primary node fails.”
When Savvicas MySQL server crashed, the company realized MySQL wouldnt be able to grow with it, Lee said, and that Savvica would have to go with a more heavy-duty, commercial database. Because the company had already decided it would need a commercial database sometime down the line, it didnt bother to explore the options for MySQL clusters, he said.
Those solutions include Continuent, which was named Emic Networks until October. Continuent offers clustering for MySQL databases, as well as Apache Web servers and PostgreSQL, SQL Server, Oracle and Sybase databases.
MySQL, for its part, introduced clustering capabilities in April 2004, with the company bragging about MySQL Cluster delivering five 9s—in other words, 99.999 percent—availability in testing.
The issue with Oracle Database 10g grid technology was cost, Lee said.
To use Oracle in a grid configuration, companies must use RAC (Real Application Clusters), starting with at least Standard Edition 1, which costs $15,000 per CPU.
“Right out of the gate, its not an option for a startup,” Lee said. “They want as close to zero spend as possible.”
DB2 Express-C fit the bill, with its price of zero.
Xkoto intends to supply database load-balancing technology for other commercial databases as well, with a Sybase release slated for the third quarter of this year. A SQL Server version is scheduled for the end of 2006 or early 2007. The company has also fielded requests to port to Oracle and is evaluating whether to do it this year or next, Lee said.
In a release scheduled for April or May, IBM AIX will be supported.
The current release supports Red Hat and SUSE Linux.
Database Load Balancer is priced per cluster, at $10,000 for companies with 5,000 and fewer employees and $40,000 for larger companies.
Editors Note: This story was updated to correct information about Xkotos marketing.
Check out eWEEK.coms for the latest database news, reviews and analysis.