(Page 0 of 1)
Cost and complexity barriers to hyperscale datacenter deployment
Growing deployments of Web 2.0, social media, Internet commerce, cloud computing, big data, Hadoop® and online game hosting – they are all driving rising demand for new hyperscale datacenter architectures that increase application performance, energy efficiency and server density to help reduce total cost of ownership (TCO). The sheer scale of these applications makes traditional enterprise server features such as extensive management tools and traditional enterprise equipment including blade systems, network switches, hot-swappable hardware and dedicated computing, management and storage networks ill-suited to hyperscale deployments. The reason is that their cost and complexity are barriers to scaling the infrastructure and make it difficult for organizations to meet hyperscale datacenter TCO, performance and efficiency requirements.
These massive datacenters are kings of operating efficiency. Because of their vast scale, hyperscale datacenters – many with at least 200,000 servers and anywhere from 1.5 million to 7 million hard drives – avoid even the smallest capital and operating expenditure excesses. For example, adding one LED to 200,000 servers adds up to 26,000 watts of power consumption and $10,000 for LEDs.
The hyperscale segment, served jointly by the OEM and self-built/ODM category, is the fastest-growing server segment, accounting for 11% of total server shipments in 2010 and, by 2015, is expected to account for 17% of x86 servers shipped, according to research firm Gartner.*
Hyperscale datacenters are usually self-built and based on four to six unique platforms for clustered database, Web interface, media serving, bulk data storage, data analytics and third-party hosting applications. Google, Facebook, Amazon, Baidu and Alibaba are among companies that have deployed hyperscale infrastructures.
Simple, easy-to-scale servers key to hyperscale deployments
Hyperscale datacenters are built with minimalistic, easy-to-scale servers comprising off-the-shelf computing components. Purchased by the rack or the pod, hyperscale servers are custom-built and configured for the customer. These inexpensive systems efficiently scale to thousands of servers while achieving very low TCO. To help customers reduce costs, hyperscale servers use homegrown or open-source software, simplifying management at scale, and eliminate hardware component redundancy found in traditional enterprise servers.
Hyperscale architectures enable organizations to easily add computing, networking, memory and storage resources to servers in larger cluster computing, distributed computing or grid computing environments. Key hyperscale architecture features include horizontal scalability for higher performance, throughput and capacity, and redundancy for high availability.
Hyperscale datacenters require distinct storage platform architectures for low-latency database transactions, high-capacity data storage, bandwidth-consuming analytics, protected database storage and replicated bulk storage. Storage architectures can drive up management costs at scale. At hyperscale, even simple questions – such as how to connect millions of hard drives and how to accelerate the performance of critical workloads – become sizable challenges.
LSI delivers bedrock technologies for hyperscale datacenters
LSI plays a key role in the hyperscale datacenter market, providing the foundational technologies that make hyperscale datacenters possible. LSI is the leading silicon supplier for datacenter hard disk drive controllers and SSD controllers. LSI® MegaRAID® controller cards and Serial Attached SCSI (SAS) fabric technologies enable reliable, economic and highly available storage. Buttressed by LSI MegaRAID performance and reliability, even Serial ATA (SATA) drives – traditionally suited only to secondary storage in datacenters because of their reliability profile – can be deployed in high-availability environments. The top six hyperscale datacenters use LSI host bus adapters (HBAs), SAS infrastructure and RAID storage solutions.
LSI Nytro™ flash accelerator cards feature on-board flash technology with LSI market-leading RAID-on-Chip (ROC) technology to greatly reduce transaction latency, dramatically improving utilization and workload efficiency in critical database applications. Most of the top six hyperscale datacenters use LSI flash solutions for performance-sensitive applications.
- Nytro WarpDrive® products offer high performance with low latency, consistent performance, low CPU burden for efficient application acceleration, and operational improvements.
- Nytro MegaRAID products eliminate the need for battery backup, dramatically improve server performance during disk failure, reduce rebuild times from days to hours, and deliver other operational improvements.
To simplify management and reduce related costs at scale, LSI offers a single software and management model across the hyperscale datacenter. The model features the same drivers, software platform, APIs and management infrastructure.
*Gartner - “Market Trends: Original Design Manufacturers' Influence on Server Market Dynamics, Outlook and Forecast, 2012-2016,” January 2013