By the time the first player stepped into the batter’s box for the official start of the 2008-2009 Major League Baseball season, MLB Advanced Media had already hit a home run of its own.
For the company, spring training meant rearchitecting the MLBlogs Network, the official affiliate blogging site for Major League Baseball. Up to the plate stepped Logicworks, but instead of a bat, the company brought a VMware-supported platform and managed database services.
The idea was to create an architecture that could scale and deliver a 100 percent uptime for the 3,000 users of MLBlogs. Logicworks’ services included the creation of a redundant database architecture to give the MLBlogs Network end-to-end database monitoring, fulfillment of backups, and management of failover and failback processes.
The database comprises two physical servers in a master/slave relationship-the primary one records all data while transactions are shipped to the secondary, or slave, database.
For MLB Advanced Media Chief Technology Officer Joe Choti, uptime and scalability are key. Traffic spikes at different points throughout the day; around 9 a.m., for example, the traffic is high as people log on to share their thoughts on the past day’s events, he said. Traffic spikes again when the first game of the day begins, he said.
“A 1 o’clock game between the Yankees and the Red Sox causes traffic to go through the roof in nanoseconds,” Choti said. “It’s our job to predict traffic load, and we communicate to them and they right-size our platform.”
Logicworks went with a VMware Virtual Infrastructure cluster consisting of three VMware ESX servers and 12 Red Hat Linux virtual machines. The Linux machines are charged with running the Web application servers required by the social media platform MLBlogs uses, Six Apart’s Movable Type.
The flexibility of the virtualized environment allows virtual machine resource allocations to be configured to help provide high availability and scale as necessary, said Gregg Kitaeff, vice president of sales at Logicworks.
“Scalability can be addressed in a couple of ways … either by allocating additional computing power to specific VMs [virtual machines]; if any one ESX [physical] server becomes too taxed, usage-heavy VMs can be moved to an ESX server that is not as heavily used,” Kitaeff said. “Of course, capacity can be added by either adding additional VMs or additional ESX servers. The specific action will be determined by balancing the short- and long-term needs.”
The final tier of the infrastructure is a storage layer where the data from the VMs lives. The company created a DRBD (Distributed Replicated Block Device) storage cluster, which provides automatic failover in a shared-nothing configuration. Writes to the primary node are simultaneously copied to the secondary node, so that if the primary server fails, the secondary server remounts the block device to ensure continued availability.
“So we have two storage servers that replicate data at the block level, and the two servers share a virtual IP address using another open-source tool called UCARP,” Kitaeff said. “In the event of failure … the site the bloggers are on is not affected, the sessions are not interrupted, and I think that was a real critical piece for MLB.”
Choti said that the site is designed with the three principles in mind.
“First and foremost [is] the enjoyment of the baseball fan, secondly for the ability to handle the spiky loads that we have here … and last but not least availability and high availability is paramount to what we do,” he said.