Web 2.0 data transactions get faster and greener

Appliances from startup Schooner do MySQL and Memcached duties with one-eighth the power and space

Cloud computing and Web 2.0 applications put a severe strain on datacenters, most of which were built for less demanding in-house and Web 1.0 apps. Faster servers and higher-capacity storage hardware, along with software innovations such as virtualization, have helped datacenter operators meet escalating service demands while saving floor space and putting a dent in overall energy consumption. However, situated squarely between your Web servers and your storage appliances are oft overlooked drivers of costly, complex hardware sprawl: caching and database appliances, charged with handling meaty data transactions.

Consider for a moment just how much more complicated and abundant today's Web application data is compared to that of yesteryear, as more companies rely on media-rich Web pages and modern e-commerce applications. Even with high-speed app servers and high-capacity storage arrays, app performance remains constrained by the limitations of the hardware that isn't well optimized to handle database and caching duties. Datacenter operators have little choice but to toss more underutilized servers into the mix to handle transactional duties.

[ Check out InfoWorld's top 10 emerging enterprise technologies for 2009. | Sensors and mapping technologies can help datacenter operators further drive down energy costs. ]

Lo, a promising solution to the problem has emerged from startup Schooner Information Technology: a forthcoming line of data-access appliances groomed to optimize application performance at the database (for MySQL) and caching (for Memcached) tiers. Specifically geared for Web 2.0 and cloud computing datacenters, the Schooner Appliances, according to the company, can provide eight times the performance of traditional servers while using just one-eighth the power and space.

Yes, those are impressive claims, and it's tempting to swallow them with a shake of salt. Then again, Schooner appears to have the brainpower to devise such appliances. Its co-founders are Dr. John Busch, previously the director of computer system architecture and analysis at Sun, and Dr. Tom McWilliams, previously a distinguished engineer at Sun, as well as founder of successful startups such as PathScale. Moreover, Schooner has backing from none other than IBM, which has signed on to manufacture, build, sell, and support the appliances.

Inside the box

The Schooner Appliances come in two flavors: Schooner Appliance for Memcached and Schooner Appliance for MySQL Enterprise. Under the hood, each is equipped with a dual quad-core Intel Xeon 5500 Nehalem processor, which delivers high performance and energy efficiency. Each also comes with optimized 512GB Intel X25-E enterprise class flash memory, employing highly parallel flash controllers. The inclusion of flash memory is indeed notable. The technology has 100-fold faster access than hard drives, as well as greater reliability and lower power consumption, thanks to the absence of moving parts. Each appliance also has a 64GB DRAM content-caching tier that provides rapid access to frequently viewed data. Finally, they provide high-speed 1/10Gb Ethernet interconnectivity.

MySQL on steroids
MySQL likely requires no introduction -- it's the most popular open source database on the market and the database of choice in the Web 2.0 world. But like any database, MySQL is ultimately hindered by the speed of the hard drives that hold its data files. MySQL shops typically boost performance by striping the data across many 15,000-rpm drives, says Busch, but "even then they must place data on the outer regions of the platter in order to minimize head movement and access time." In short, servers running MySQL are bottlenecked by hard drives, and processor utilization is very low.

Further, scaling MySQL to handle enterprise-class workloads, according to Schooner, involves extensive data partitioning and application modifications. Application query mixes, for example, must be carefully controlled due to the effect that queries such as table scans can have on performance, Busch notes. Optimizing MySQL entails the complex task of tuning data block sizes according to the workload and changing user and application patterns so that the working set fits in the buffer pool effectively.

The Schooner Appliance for MySQL Enterprise is groomed to overcome these and other debilitating obstacles by fully exploiting the Nehalem multicore processor with hyperthreading, which, along with high-capacity storage, helps boost scalability. It also reduces or eliminates the need for database sharding or horizontal partitioning. The machine's 512GB of parallel NAND flash memory speeds up transactions considerably. Moreover, the appliance is optimized for OLTP (read-write intensive workloads) and OLAP (read mostly workloads). Finally, it incorporates highly optimized InnoDB 1.0.3.

The payoff: One of these appliances, according to Schooner, can do the work of up to eight legacy servers, which translates to a dramatic reduction in energy consumption and an increase in floor space. There's also the important benefit of having fewer machines to manage.

Caching in on Memcached
As its name implies, the Schooner Appliance for Memcached is built to run Memcached, a key/value distributed cache commonly used in high-traffic Web sites. According to Busch, legacy Memcached installations are typically limited by the amount of DRAM that can be installed in a single server node. An application that requires a total of 512GB of memory is typically spread across 16 Memcached server nodes at 32GB of DRAM per node. Further, most legacy Memcached nodes use only 1Gb Ethernet links, which are sufficient because of the limited amount of DRAM and CPU throughput per node, according to Busch.

The Schooner Appliance for Memcached, by comparison, is capable of balancing 16-core hyperthreaded Nehalem processing with the network bandwidth of multiple 1Gb or 10Gb Ethernet links, and hundreds of thousands of accesses per second of flash memory.

It boils down to this: One appliance is able to handle 512GB of cached data. Thus, handling a 1TB Memcached workload would require four 2U Schooner appliances drawing 2.5kw, compared to 33 2U legacy servers, which would consume around 18.7kW. That represents a 50 percent reduction in operating costs over three years, according to Schooner, while reclaiming a significant amount of floor space.

The Schooner appliances won't officially ship until early next year, though a handful of undisclosed companies have been beta testing them. Given the backing that the technology is receiving from IBM, I believe it's a product offering well worth watching, as it addresses a critical challenge in the Web 2.0 datacenter. Costly, wasteful server sprawl and underutilization won't end overnight, but purpose-built appliances like this are going to help significantly.

This story, "Web 2.0 data transactions get faster and greener," was originally published at InfoWorld.com. Follow the latest developments in green IT and cloud computing at InfoWorld.com.