The Cloud: An Alien Technology Redefines Data Centers
By Mike L. Chase
Much of the technology that powers the cloud is derived from the intellectual property of the cloud service providers themselves or an emerging market that caters to only them. It is an alien level of technology that much like master magicians is hoarded as much as it is sought after.
While the ignorant claim that the cloud is mere virtualization, there is a powerhouse of technology combined with sophisticated financial models to create overarching software and hardware combinations that pioneer cloud solutions at a dizzying rate at costs that even the Fortune 50 can't touch. Outsiders usually only get a mere glimpse of such technology on some TV show featuring a quick blurry flash down one aisle of a data center—just enough to tease and amaze.
Why You Aren't the Cloud
For starters, the world's top clouds don't use your favorite hypervisor vendor, your favorite server vendor, your favorite storage vendor, RAID, iSCSI, Fibre Channel, blade systems, tape backup or whatever else you might run in your own data center—they just don't. Such technologies are outdated, too expensive, don't integrate well and waste time. While most are "standards," they are wildly implemented by various vendors whose aim is never to interoperate.
The Cloud—Alien Technology Run Amuck
Servers: At the high end, a typical cloud server in a 2U of rack space has 8 CPUs—which, combined, yields 128 virtual cores, 2 terabytes of memory and a minimum of 80G bps of upstream ultra-low latency network bandwidth using special NICs/switches.
Storage: Distributed Object Storage has replaced RAID in every major cloud on the planet. RAID6+2 can take two to three days to rebuild 1TB of data when a disk or other component fails. For every 720 terabytes to 1 petabyte of data, you're talking about an average of 1,200-plus disks, most of which were purchased around the same time. So what do you think the odds of failure are at some point over a three-day rebuild period? Think you'll lose one drive sooner or later? Two? At three failed disks, you can say good-bye to your entire array, which is why it's very popular now to buy two arrays from the same vendor and mirror them. However, this raises the average cost of your data in a 36-month financial model from 12.3 cents per GB per month to 24.6 cents per GB per month. And that's unencrypted because adding encryption is complex and costly on most enterprise arrays. Now you know why clouds don't use traditional storage. It's easier to keep three copies encrypted within distributed object storage systems, which can rebuild 1TB in 20 minutes, not three days. Cost per copy? Less than 1.5 cents per GB per month.
Networks: Infiniband and ultra-low-latency Ethernet (Broadcom's Trident+ chipset used by vendors such as Cisco, Brocade and others) mean that everything from Hadoop to virtual desktops, as well as other high-impact transaction-oriented services, can thrive in the cloud often at 40G-bps or 10G-bps speeds minimum, all the way up to 100G bps+.
Power/Space: Typically, efficiency in all of the above gear when plugged into a cloud data center is 95 percent or higher. Most Energy Star ratings that enterprise customers like to pound their chest about only start at 80 percent by comparison. Better power conversion ratios mean less heat, requiring less cooling, requiring less power. An entire 42U rack of cloud gear can be powered by as little as two to three (at most) 30-amp 208-volt circuits. I've seen single blade systems and some network switches that take 3x 30-amp 208-volt circuits just for themselves. That's why you don't see them in the cloud.
Transport: Do you find yourself paying more than $4 per month for 1M bps of Tier 1 bandwidth to the Internet? Do you own a BGP ASN # or thousands of public IPs so you can deploy and re-route your network worldwide within seconds? Have you mastered the CCIE Jedi art of how to block an intermediate carrier from snarling your path from point A to point B via the Internet? Are you peered to 15-plus carriers for Internet, Multiprotocol Label Switching (MPLS), private links, private IP networks and have deep contractual understandings for these services, which involve very mature, defined QoS/DSCP (quality of service/differentiated services code point), service-level agreements and more? Then you're not the cloud.
Solutions: Whether engineered internally or when some components are used from the vast array of software or hardware "merchant silicon" on the market, cloud service providers are keen to use their army of programmers to inject their own innovation and signature touch to differentiate themselves immensely. All clouds offer similar services, but if you peel back the lid, many of them are wildly different. Nonetheless, the ability to monetize and offer a unified product to the world that can still be customized to a reasonable level is what makes the cloud ultimately untouchable.
All the factors I've mentioned are merely the tip of the iceberg in a vastly diverse digital landscape that is cloud. Every day, real people are doing unreal things in the cloud. Sure, you could huff and puff into a plastic bag and call it your own private cloud, but it's just not the same.
The cloud's offerings, cost models, technology, support, insane talent pool and pioneering spirit are something to be leveraged in your personal and corporate life, not competed against in a career-limiting or career-ending private cloud foray that largely results in doom. The question really comes down to how many public clouds you should you be in and which set of digital Legos are the best fit for your strategic goals.
Mike L. Chase, J.D., also known as "Dr. Cloud," is executive vice president and chief technology officer for dinCloud.