LAS VEGAS -- Amazon can commission a virtual machine in seconds. Facebook can assign one system administrator to 24,000 servers. Google can let 1,000 servers fail in a given year without blinking an eye.
The shift [to Web-scale IT] is like moving from medieval expert guilds to industrialized manufacturing.
vice president and analyst, Gartner
While the typical enterprise data center cannot operate like this, it is time for IT teams to adopt the Web-scale IT mindset, according to several speakers at the Gartner Data Center Conference here this week.
"The shift is like moving from medieval expert guilds to industrialized manufacturing," said Carl Claunch, vice president and distinguished analyst for Gartner Inc., an IT industry analysis firm based in Stamford, Conn.
Here are some of the tips Gartner analysts offered up to achieve a Web-scale strategy:
1. Refresh quickly.
Faster technology refresh cycles can save enterprises money, not cost them more. For example, the cost of powering a server for another year could be more than the price of a server refresh. That new server will draw 20% to 30% less power than the legacy system, and host 15% more workloads. Hanging on to old systems to "get the most out of them" is backward thinking, explained David Cappuccio, research vice president at Gartner.
Data center operators don't have a lot of time to research new generations of hardware and software, weigh their capacity needs and budgets, and refresh advantageously, said Henrique Cecci, a Gartner research director based in Brazil. But the smart ones make time.
2. Accept failure.
"Anti-fragile IT" is a tenet of Web-scale production environments. Enterprises must transition from rigorously protecting hardware components and striving for the best array, best server or other component to accepting failures within a resilient system. The pieces might not be perfect or reliable, but the complete data center works together to be, said Cameron Haight, research vice president at Gartner.
3. Utilize server standardization.
The benefits of server standardization will outweigh the benefits of customization in a scalable data center. Facebook, for example, runs about 200 applications on five server types. There must be a very compelling business case made for any additional server types, because customization for applications will limit your ability to scale and innovate, according to Claunch.
To follow the hyperscale data center model of commodity hardware, turn to original design manufacturers or simply ask your server vendor to tailor a product to your needs. Server manufacturers are becoming more accommodating of data center requests as they see the rise of original design-manufactured servers, said Peter Grant, managing vice president at Gartner.
4. Build it.
If the right tools to provision and manage your IT environment don't exist, create your own. Make use of open source development and also contribute to it. Strip out any management options that are not strictly necessary for your infrastructure. Give admins more places to look, and you're wasting productivity, Claunch said.
5. Embrace DevOps.
DevOps is another Web-scale strategy that trickles down to enterprise IT. Enabling development and operations teams with the same goals and metrics pays off with more modular, easier rollouts and roll-backs, faster updates, and fewer disruptions to applications. Applications are designed leaner when operations teams are involved from the beginning, analysts said.
DevOps can also shake IT out of a rules-bound funk, re-energizing the team to move beyond operations as they've always been.