Amazon's 7 Cloud Promises: Hype Vs. Reality
Amazon says cloud computing is an indispensable enabler of seven important computing transformations. We hype-test the promises with a few reality checks.
![](https://eu-images.contentstack.com/v3/assets/blt69509c9116440be8/bltc0182b2356ae8eed/64b83949410a1b4c0bd7459b/IW_generic_image.png?width=700&auto=webp&quality=80&disable=upscale)
Seven crucial technology transformations have strong foundations in cloud computing. That's the bold claim recently put forward by Werner Vogels, CTO of Amazon, at the recent Amazon Web Services Summit 2012 in New York. In some cases the transformations have been either enhanced or accelerated by the cloud, as in the cases of big data analysis and mobile computing. In other cases, these transformations weren't possible before cloud approaches came along, as is the case with on-demand scaling and agile innovation, he asserted.
We felt the need to throw a little cold water on those claims, particularly after Amazon's Adam Selipsky, VP of product marketing, sales, and product management, upped the hype factor at the event with the even bolder claim that enterprise data centers will be largely replaced by cloud computing within 20 years. Our coverage of that prediction sparked more than a handful of reader comments, with one reader calling it "a load of dung."
Vogels' comments seem less grand by comparison, but we don't see public cloud approaches as the only way to go in all the scenarios he would suggest. In particular, our reporting on cloud service failures, cloud security breaches, and cloud cost and performance drawbacks might give you pause.
Don't get us wrong: Public cloud computing offers many incredible possibilities, like the prospect of doing supercomputer-level processing on demand and at an incredibly low cost. For example, Cycle Computing is helping pharmaceutical firms do hugely important research on cancer treatments and other diseases at lightning-fast speeds and very low costs. The company figures it put together the equivalent of a $20 million supercomputer on AWS at a cost of less than $5,000 per hour of use. And when the project was over, so were the fees.
Cloud is also great for startups and in other fast-growth scenarios where it's too hard to find people and you don't want to get slowed down by conventional IT. For example, three years ago PBS was serving up about 200 terabytes of streaming video per month from its site. But the iPad and video-capable smartphones have unleashed a wave of demand. Today, one year after the debut of a PBS iPad app, the content provider is streaming more than 40 petabytes of video per month. A PBS exec says the network could not have kept up without capacity from a public cloud.
Judge for yourself. Dig into these seven transformations and you'll likely see scenarios where your company will have to make the public cloud vs. private cloud vs. on-premises decision.
It's a given that computer systems fail, so computer scientists have known for years that the way to mitigate risk is to rely on highly distributed, fault-tolerant architectures. But it's easier said than done running all those servers and networks and synchronizing redundant, geographically isolated data centers. With cloud computing, running on reliable, distributed systems "becomes relatively easy," asserted Amazon CTO Dr. Werner Vogels.
It's obvious that tapping into a distributed cloud service is easier than building one from scratch. In Amazon's case, you can spread your deployment across eight global regions, each with multiple Availability Zones, and, within Availability Zones, multiple data centers, each located on separate seismic plates and running on separate power grids. Amazon runs distributed services--including Simple Storage Service (S3), DynamoDB NoSQL database service, and the Amazon Relational Database Service (RDS)--across multiple availability zones. It also offers distributed Elastic Compute Cloud (EC2) processing power and administrative services including Simple Workflow Service (SWF), Simple Queue Service (SQS), and Simple Notification Service (SNS).
Plenty of other cloud vendors have globally distributed architectures, but AWS has a 59% share of the infrastructure-as-a-service market, according to The 451 Group, so its advantages in scale should, in theory, translate to higher levels of overall capacity, scalability, system distribution, and redundancy.
Amazon does a lot to shield its customers from the complexity of creating their own highly distributed systems on AWS infrastructure, but Amazon itself can't internally avoid complexity or the certainty that systems fail. Despite its global scale and many redundancies, Amazon has a less-than-perfect record of keeping its systems up and running.
Amazon suffered two significant outages in 2011 alone, with one multi-day incident in April that took out services on the U.S. East Coast, and another multi-day incident in August that impacted multiple availability zones in Europe after power was lost and backup generators failed at Amazon's data center in Dublin, Ireland. The biggest concern in the August incident was that a problem in one Availability Zone ended up taking out another Zone. That's not supposed to happen if you have a highly distributed, fault-tolerant architecture.
Game site Zynga, which relies heavily on AWS, avoided disruption during these outages because it also has its own private-cloud capacity, further distributing and adding separation and redundancy to its available compute capacity. Zynga also determined it was better to own than rent, so it has flipped from 80% dependence on AWS to 20% dependence, with public-cloud usage focused on scaling up new games before taking them in house.
Likening conventional enterprise data centers to a castle with a moat for security, Vogels said, "as the castle starts growing and growing, as happens with most enterprises, it becomes much harder to protect what's inside the castle." With the cloud it's possible to offer an "end-to-end security guarantee," he said, calling security Amazon's number-one priority and area of investment. He pointed to application-level tools that make it possible for customers to protect themselves "in ways that were not possible" before the cloud.
Sure, Amazon and other cloud providers comply with plenty of rigorous security standards, but are they really more secure than far more anonymous on-premises systems? In January, Amazon-owned Zappos became one of the latest e-commerce sites to suffer a security breach, potentially exposing credit card information. Aside from damaging credibility and trust, that incident also precipitated a lawsuit.
There's some evidence that cloud computing tools help malware creators, as in an example last summer when cybercriminals were found to be using AWS to spread financial-stealing malware. The attackers apparently deployed registered accounts to wage the attacks on 11 international banks. Amazon was notified of the problem by Kaspersky Labs, but it then took Amazon 60 hours to shut down the malicious links, according to the security firm. Of course, malware creators will use any and all current technology tools at their disposal.
On premises, scaling is something you have to architect for and work hard to achieve. But distributed systems built on cloud principles let you scale on demand. No argument there. When using an on-premises database cluster, you have to continually configure, tune, shard, and re-partition to maintain performance. Vogels says the AWS approach removes database-as-an-application bottlenecks because you call for capacity with specific performance characteristics on demand. This makes application scaling a much simpler affair.
Paying a cloud infrastructure-as-a-service (IaaS) provider is obviously easier, but as InformationWeek's Art Wittmann has documented, taking the virtual, private-cloud route can save you money and deliver higher performance. That's one reason IaaS adoption is actually slowing. Public cloud makes tons of sense when there's spiky demand, but for baseline capacity that's a given, it's cheaper to do it yourself.
Amazon says its systems rank 42nd on the list of the world's most powerful supercomputers. Purpose-built, research-oriented machines top the list, because they tend to use super powerful graphical processing units, not the general-purpose CPUs in service on AWS. Still, public clouds can deliver incredible compute capacity by the hour.
Amazon customer Cycle Computing uses AWS capacity to handle supercomputing challenges. In the company's largest project to date, Cycle Computing was able to draw on 51,000 compute cores from across Amazon's global AWS compute capacity to test potential cancer drugs for the pharmaceutical research firm Schrodinger. The system ran completely on AWS and delivered answers in three hours that would have taken more than 12 years to process on a single core.
No argument here, our only quibble being that cloud makes the most sense for spiky demand; if you have a constant baseline workload, there are advantages in owning rather than renting. That said, research tends to be spiky, so it's no surprise that cloud is compelling even for large pharmaceutical giants that have ongoing supercomputing needs. Cycle Computing CEO Jason Stowe said his firm was able to harness more than 50,000 cores of AWS capacity within a matter of hours, and he figures that system packed the equivalent power of $20 million to $25 million worth of supercomputing infrastructure. At Amazon's cloud rates, the cost was just $4,828.85 per hour.
"This means any researcher with a National Science Foundation grant or any person at an academic institution or anyone at a large corporation can now do science that's impossible to do on an internal system in so short a timeframe," Stowe said. We say, bravo!
To drive innovation, Amazon CTO Werner Vogels makes the solid point that you want to experiment often and fail early. In other words, you want to do experiments quickly, so if the concept doesn't work out, you haven't spent too much time and money on a failure.
While on-demand compute capacity is very compelling for business experiments, the reality is that with virtualization, on-demand can be delivered publicly or privately. If, for example, a company has $7 million in virtualized infrastructure in place to serve its baseline needs, wouldn't it want to tap into reserve or excess capacity, as available, to support research? This is another case where we see all-cloud enterprises as the exception rather than the rule.
So the question is, do you have excess, quickly available capacity internally, or do you need to turn to the public cloud? Both options make sense. If you need big capacity that outstrips what you have available internally, public cloud wins.
There's a lot of work in handling big data, particularly if it's growing in variety and velocity, as it is at many fast-growing Internet startups. Big data is different from the old style of business intelligence, Vogels said, in that you don't always know what questions you want to ask. That means there's uncertainty about the resources you'll need, making public cloud a good fit. Scalable, distributed cloud services like Amazon DynamoDB (NoSQL database), RDS (relational database service), and EMR (Hadoop-based Elastic Map Reduce), or rival cloud offerings from the likes of Microsoft or Google), make it possible to quickly respond to big data analysis opportunities.
Yes, scalable, cloud-based services will particularly appeal to companies that are getting their data from the Internet to begin with, but privacy- and security-sensitive firms won't be rushing into the cloud. Hospitals, banks, intelligence agencies, and other firms have good reason to be paranoid, and perceptions of superior on-premises security still prevail.
Even in the sports business, teams are using big-data sources such as Stats' SportVU system for the NBA and the Pitch FX system used by Major League Baseball since 2006 (and soon to be joined by a Field FX system that collects data on fielding). These camera-based systems track positions of players, the ball, vectors, and velocities on every pitch and play, generating huge amounts of data in the process. That's all publicly available data, yet according to Kevin Goodfellow, an executive at Sportsdatahub.com, teams are usually loath to put their own data on a public cloud. Sportsdatahub does dedicated hosting as well as on-premises deployments of Hadoop-based systems designed to help teams spot promising athletes, team strategies, and marketing approaches.
Mobile is a hot trend, and when agility matters, cloud-based approaches have real appeal. Jon Brendsel, PBS's VP of products, says his network could not have scaled to meet fast-growing mobile video viewing demand without AWS. PBS serves up content and streaming video to more than 30 million unique visitors per month and an average of 115,000 unique mobile visitors per day. The network's AWS-based mobile architecture includes nearly 70 databases running on Amazon EC2 and more than 170 storage "buckets" on the Amazon S3 storage service.
Three years ago, PBS was serving up about 200 terabytes of streaming video per month. Today, one year after the debut of a PBS iPad app, the content provider is streaming more than 40 petabytes of video per month. "We've grown a lot, but with Amazon's infrastructure, we're set to scale significantly," Brendsel says.
The reality is that markets mature and eventually level off, but given that it's early days for mobility, this is one area where CIOs facing big mobile growth might be wise to avoid continuous capacity buildout until the potential is better known. Once things do start leveling off, big enterprises with virtualized capacity will be able to shoulder predictable volumes more economically.
The reality is that markets mature and eventually level off, but given that it's early days for mobility, this is one area where CIOs facing big mobile growth might be wise to avoid continuous capacity buildout until the potential is better known. Once things do start leveling off, big enterprises with virtualized capacity will be able to shoulder predictable volumes more economically.
Seven crucial technology transformations have strong foundations in cloud computing. That's the bold claim recently put forward by Werner Vogels, CTO of Amazon, at the recent Amazon Web Services Summit 2012 in New York. In some cases the transformations have been either enhanced or accelerated by the cloud, as in the cases of big data analysis and mobile computing. In other cases, these transformations weren't possible before cloud approaches came along, as is the case with on-demand scaling and agile innovation, he asserted.
We felt the need to throw a little cold water on those claims, particularly after Amazon's Adam Selipsky, VP of product marketing, sales, and product management, upped the hype factor at the event with the even bolder claim that enterprise data centers will be largely replaced by cloud computing within 20 years. Our coverage of that prediction sparked more than a handful of reader comments, with one reader calling it "a load of dung."
Vogels' comments seem less grand by comparison, but we don't see public cloud approaches as the only way to go in all the scenarios he would suggest. In particular, our reporting on cloud service failures, cloud security breaches, and cloud cost and performance drawbacks might give you pause.
Don't get us wrong: Public cloud computing offers many incredible possibilities, like the prospect of doing supercomputer-level processing on demand and at an incredibly low cost. For example, Cycle Computing is helping pharmaceutical firms do hugely important research on cancer treatments and other diseases at lightning-fast speeds and very low costs. The company figures it put together the equivalent of a $20 million supercomputer on AWS at a cost of less than $5,000 per hour of use. And when the project was over, so were the fees.
Cloud is also great for startups and in other fast-growth scenarios where it's too hard to find people and you don't want to get slowed down by conventional IT. For example, three years ago PBS was serving up about 200 terabytes of streaming video per month from its site. But the iPad and video-capable smartphones have unleashed a wave of demand. Today, one year after the debut of a PBS iPad app, the content provider is streaming more than 40 petabytes of video per month. A PBS exec says the network could not have kept up without capacity from a public cloud.
Judge for yourself. Dig into these seven transformations and you'll likely see scenarios where your company will have to make the public cloud vs. private cloud vs. on-premises decision.
About the Author(s)
You May Also Like