Cloud Computing: When It Rains, It Pours

It's fast. It's cheap. It's easy. It works.

What could possibly go wrong?

The CRN Test Center has reviewed hosted applications and services in the past, including the likes of Google Apps, Data Deposit Box hosted storage and hosted security. This month, we focus on several more: Amazon.com's Elastic Cloud Computing (EC2), Salesforce.com and Trend Micro's hosted security. Overall, we believe hosted applications, when deployed where it fits with a customer's needs, can be a powerful solution to vault an enterprise into high levels of operation in short order at a limited cost.

But let's be clear: Don't expect "the cloud" to provide five-nines of availability and don't expect it to be the default solution for those wanting cost-competitiveness. Expect the unexpected.

Before stopping to count all the profit you or a customer could make by deploying a cloud-based solution, consider that just one outage for the span of a few hours could eat up a measurable piece of a company's profit for a quarter. Could never happen, right? It did, though. On July 20, 2008, Amazon.com suffered a catastrophic outage in its S3-hosted storage business. While Amazon managed to get the service up and running in several hours, and the event is now just a blur to many, it impacted thousands of businesses and individuals who had gone to S3 as a convenient, cost-effective way to store lots of data easily.

And, over the past several months, search and online advertising giant Google has suffered notable outages in its Gmail and Google News offerings -- outages that the company has left largely unexplained.

The health of the broader, global Internet should also be on your regular checklist.

Since the days of dial-up connectivity passed into the history books, the relative health of the Internet has become little more than a passing, occasional thought in the U.S. and developed world. Fiber cables that telecoms dropped in places around the world from the late '80s to mid-'90s have provided bandwidth that has given us the Internet economy. Never mind that much of that fiber buildout has slowed, that higher levels of bandwidth are eaten every day for multimedia and voice applications over the Internet, and that the transformation of the Web into a mobile lifestyle is taxing infrastructure to the nth degree.

For now, experts and data demonstrate that the Internet -- with its many, global backbones -- is running smoothly. Even when taxed by historic events, like this year's inauguration of President Obama that caused massive, historic levels of bandwidth to be eaten by those watching the event's streaming video over the Web, global network performance only registers a passing blip.

Based on our review, which includes examining Internet performance, cloud product performance and discussions with experts including executives of Keynote Systems (see below), when considering whether a cloud application or service is the right approach, these practices should be deployed:

A complete audit and forecast of the business involved to develop a cost-benefit-risk analysis, on an application by application basis for a move to a hosted computing model vs. traditional client/server-data center models. An audit of the cloud service with a focus on issues including geographic redundancy, latency, packet transport performance and uptime guarantees. An audit of the business' own ISPs, including performance at connecting points between carriers including AT&T, Verizon, Savvis, NTT and others to determine potential future performance issues. A determination of whether the cloud-based application or service shares bandwidth with other companies that are "resource hogs." While most cloud services imply shared resource (that's what makes many of them economical), having a small bicycle shop share resources with a Wall Street brokerage might have negative consequences for the bicycle shop. Constant monitoring of those performance issues, once hosted solutions are up and running.

Using monitoring software and services (which are outlined below), we found areas of the Internet that ran continuously with little to no latency or packet loss and other points that routinely saw packet loss of 5 percent to 10 percent at least once or twice a day, as well as latency that could have significant—and negative—business implications if a hosted application depended on it for performance. These are the types of issues that are critical to keep in focus.

With that in mind, we think there's a lot of potential in taking a hybrid approach in many cases: keeping some or many computing and applications at a customer's, solution provider's or MSP's site while taking advantage of hosted applications where the technology and risk make sense.

Here are several solutions we've looked at and believe could fit in well in a nascent, cloud buildout.

Keynote's Keynote Internet Testing Environment (KITE) and Internet Health Report

Keynote Systems Inc., San Mateo, Calif., was among the earliest adopters of Internet benchmarking technology and remains a force in this space today. While researching this story, we turned to Keynote's free, over-the-Web Internet Health Report at internetpulse. net. This report is a real-time look at the health and performance of several Internet backbones in the U.S. and, importantly, backbone-to-backbone connecting points.

Keynote maintains more than 3,000 servers and PCs at 200 sites in 59 countries to monitor real-world Internet performance based on actual Web traffic and performance. The Internet Health Report monitors and reports on uptime, latency and packet loss throughout the U.S.

Another free Keynote product, the Keynote Internet Testing Environment (KITE), is a separate, rich desktop application that can also be used to monitor the performance of individual Web sites. The CRN Test Center likes, and recommends it, for use by solution providers working with customers in the earliest stages of evaluating a move to cloud-based services or solutions.

We monitored the Internet Health Report continuously for several weeks. What was impressive and valuable about this service, even as it's free and Web-based, is that it let us begin to see some performance patterns among carriers such as Savvis, NTT, Qwest, Level3 and Verizon. We were able to spot patterns of latency, for example, in areas where one carrier's network met another's at similar hours of the day and during different days of the week. If a particular business uses an ISP that deploys those two carriers, it could be a yellow or red flag to search for an alternative.

KITE, as a free application, is more of a taste of Keynote's paid applications and services. It's downloadable and works on Windows XP SP1 or later, or Windows Vista. KITE allows for the measuring of a Web site's performance from various locations around the world. In testing CNN.com's performance using KITE, the service performed an analysis in less than a minute that showed the site had about a 1-second response time at all of Keynote's measurement sites around the world—except for Hong Kong, which turned in a painfully slow 23-second response time.

Smart decisions require good information and good data. Smart decisions about cloud computing require good data about Internet-based performance. Keynote provides that in free applications and offers custom-tailored, pay services for even deeper, richer detail.

Keynote's Internet Health Report is a free, Web-based service that examines performance of the global network.

Amazon.com's EC2

Amazon.com has been among the most aggressive companies in building out a cloud computing offering, beginning with its S3 storage service and building out several more offerings including its Elastic Cloud Computing—a pay-as-you-go, Web-based hosted server solution.

Like most Amazon Web Services (AWS), EC2 is turnkey-simple. Within minutes, we were able to take a prewritten image (an Amazon Machine Image, or AMI), upload it to Amazon hardware, and turn on a server running Windows 2003 Server with Eclipse 3.4, Tomcat Server 6 and Java 6. However, Amazon makes scores of other images available in the same turnkey fashion, ranging from servers running IBM's DB2 to servers running Linux or OpenSolaris and a variety of different databases, including Sybase and Oracle.

Amazon also connects EC2 to other services, including its S3 storage service. For example, you can customize one of its system images, take a snapshot and store the snapshot and other data on S3. Amazon charges for storage space and for data transfer, in rates ranging from 12 and a half cents per hour for some AMIs based on Windows, to about 10 cents an hour for some systems using open source.

When we tested our server, we noticed no latency or performance issues; however, we didn't run anything that could be considered an intensive workload for many companies. Generally speaking, Amazon.com is given credit by many for the "geographic redundancy" of its infrastructure in that it maintains high-performance data centers in geographies around the world. That not only provides for redundancy in the event one site suffers an outage or performance issues, but it ensures that if a catastrophic event hits one geography (such as an earthquake or natural disaster), it won't take a week to get fuel to a location for even backup generators to run so basic service can be restored.

NEXT : Salesforce.com and Trend Micro