Friday 19 April 2013

Intel CIO Kim Stevenson on big data, OpenStack, women in IT | ZDNet

By for Between the Lines |

Intel CIO Kim Stevenson, who has been at the helm for a little more than a year, said OpenStack is the most useful cloud architecture for avoiding lock-in, outlined how the chip giant is using big data techniques, and talked capacity planning for her company.
Stevenson, formerly the vice president of global operations and services at Intel, has a dual role at the company. First, she's in charge of the IT operations for Intel globally and has the usual CIO headaches. And second, Stevenson is the guinea-pig-in-chief for Intel's products.

Intel's IT group is typically the "first implementer" for new hardware and software, said Stevenson. In other words, Intel eats its own dog food and gives product people honest feedback on what needs to happen. Stevenson quipped that the conversations from an IT customer perspective don't always go well.
Here are a few key highlights from a breakfast conversation in New York:
Big data. One area where Intel is the first implementer is on its Hadoop distribution. Like other large enterprises, Stevenson's group is looking toward big data and analytics to boost revenue, improve time to market and cut costs. Here are Stevenson's biggest takeaways on using big data internally.

  • Talent: "Whether developing or implementing big data, talent is in short supply," said Stevenson. She added that Intel is working university partnerships and programs to get talent, but it takes time to turn interns into big data visionaries. Intel is also training employees, but noted that the transition to big data can be difficult. "It's not easy for a database person who thinks in rows and columns to learn Hadoop," said Stevenson. "It's also difficult for business people because they usually want clean data."
  • Time to market is the big data win: Intel's biggest goal with big data and analytics "is to reduce the product cycle time," said Stevenson. After all, Intel is entering the smartphone market and the product cycles are twice as fast as PCs. To speed up product development time, Intel looked at the post silicon validation process. This process revolves around making sure that all of the transistors---all 1 billion or so---are functioning properly. The data out of a silicon wafer equates to 750 terabytes. Intel deployed a machine learning model that isolates testing and will cut 25 percent of a five quarter product cycle period.
  • Manufacturing: Big data is also used to cut costs in manufacturing by analyzing silicon wafers and how they are cut. The goal is to have less waste and more processors for better yields. Hadoop and Intel's analytics stack -- which consists of IBM's Netezza, Hadoop and proprietary algorithms -- logs information, finds errors and best practices and passes the data along to the next tool. If each step improves, there are better good die yields and lower unit costs, explained Stevenson.
  • Marketing fraud: Big data is also used to detect fraud and errors in the chip maker's Intel Inside marketing. In a nutshell, Intel partners market Intel Inside and get reimbursed by the company on their marketing spend. Stevenson said that Intel used its machine learning engine to look at fraudulent claims, mistakes and incomplete reimbursement requests. Intel also looked to cut down on manual post audits. The return was $7 million in savings in a quarter and the expulsion of bad guys. Reimbursement time also improved.
The cloud approach. Stevenson's architecture is cloud heavy---for some things. Other areas such as product design and manufacturing data will never see the cloud or come anywhere near it.
Here's how the cloud fits in with Intel's key areas: Office applications, silicon design and manufacturing.
  • Typical enterprise applications: According to Stevenson, 75 percent of Intel's typical office systems such as email and ERP are in the cloud in a private-public hybrid model. "In 2012, we were bursting to other Intel regions and now we burst out to other providers," she said. One public cloud use case is Will.I.Am. The artist can offer a free download on Intel.com and get a ton of Web traffic. That use case is perfect for the public cloud.
  • Design: "Silicon design will never go out to the cloud. That's our core IP," said Stevenson. She added that no cloud service level agreement or chargeback would ever compensate for Intel's intellectual property being leaked. Instead, product design runs on a high performance computing grid that's internal.
  • Manufacturing: Manufacturing is another area that won't be put into the cloud. The information is housed in small data centers near the manufacturing site and later aggregated.

Beyond design and manufacturing, though, Stevenson requires that an application request has to prove why it needs physical resources. In other words, applications need to be virtualized or they won't get capacity.
Capacity planning. Stevenson said the goal of Intel is to have 15 days of capacity on hand. This 15 day rule means that if the typical usage occurs within Intel with the usual new requests, the company will be out of compute in 15 days. For Stevenson, compute capacity is her inventory. The 15-day capacity cushion is based on memory mostly since that's the first to go when resources are being taxed. Compute and storage are also watched. Stevenson acknowledges that the 15-day capacity best practice may not work for all companies and could even be shorter. "Some companies may have to convince the CFO that having 15 days of excess capacity sitting around is a good idea," said Stevenson. "We didn't have to convince our CFO of that."

Playing the cloud provider game. Stevenson said Intel has gone with OpenStack, a cloud architecture that's gaining momentum. Why? Intel uses multiple cloud providers working under a master services agreement and wants to hop between them to maximize performance and costs. "We use several cloud providers mostly in the U.S.," she said. "We've had no issues with SLAs, but we're only bursting. We're careful with what workloads we use in the public cloud." OpenStack enables Intel to use multiple providers and avoid lock-in, said Stevenson.

About those cloud brokers. I asked Stevenson what she thought of cloud brokers---companies that would manage providers to maximize savings. "Cloud brokers are transitory. That dog doesn't hunt as an independent business model," she said. Ultimately, the ability to hop between cloud providers will be built into an infrastructure like OpenStack and automated.

Women in IT. Finally, Stevenson talked a bit about women in IT and was asked about Sheryl Sandberg's Lean In: Women, Work, and the Will to Lead, which drew some fire over what the Facebook operating chief had to say. Stevenson said the book was largely on target and added that Sandberg had good advice for anyone in business not just women. Her main takeaway on women in IT is that they have to understand the communication differences in a male dominated industry. In other words, know how to navigate the subcultures such as sports chatter and international differences. Also use good context at the start of a business problem to navigate functional and relational communication styles.

Source: http://www.zdnet.com/intel-cio-kim-stevenson-on-big-data-openstack-women-in-it-7000014221/

No comments:

Post a Comment