In-depth analysis of the eight public cloud services

There are significant differences in performance, complexity, and speed between Amazon and Windows Azure,iaas cloud.

The information on the cloud is always quite simple: hand over your worries, IT managers and we will help you solve everything. Forget to install the server and the engine backup that needs to be double-clicked, and don't worry about thousands of things going wrong. Just give us your credit card number and your data and we can do everything for you.

For the past few months I have been living in a dream where I have built a vast computer kingdom across the globe. Ubiquitous devices turn my data into tiny bits and then process them into more numbers. The personal network transmits my scattered secret information between different devices so that others can process the data and transform it into beautiful pictures. Sure, my desktop is a bit old and should be able to use more memory, but I created a global robotic army with my browser, and I can manipulate them as easily as the wizard apprentice in Fantasia.

The good news is that unlike apprentices, when I let these machines go, they can disappear more or less. This is the beauty of the cloud! When you need it, you buy what you need. Oh！ Using Microsoft Windows Azure Cloud to store that little bit of data string, we need to make an indefinite recurring fee, but fortunately technical support is already investigating how to eliminate the cost and I hope it will be faster. Every time I see a penny bill appearing on my credit card bill, it always reminds me of these data strings.

Then the other equipment came along with a small penny-denominated fee. Most of the dollar stores are cleverly renamed to accommodate items that sell for less than 5 dollars, but on the cloud, it is possible to buy equipment just like a cheap penny candy. It's time for some of these old-fashioned equipment makers and manufacturers to get back to life.

Key differences

When the survey of all the clouds was completed, the most striking finding was that the cloud world was so dramatically changeable. Those who think cloud facilities are just commodities are wrong. The market team insists that cloud allows us to switch between computers and storage devices as easily as we can replace Lego blocks, but that is not entirely true. All vendors are trying to make their products and services unique by offering slightly different, slightly superior things. Sometimes it takes time to find out the difference between different products correctly, but for anyone who has a lot of work to do, change is always important.

Variability is always accompanied by an operating system. It's easy to think that Linux is everywhere because Linux is ubiquitous, but that ignores the contrast in rationing. Although standard distributions such as Ubuntu are ubiquitous, companies are still creating their own enterprise versions with subtle or less subtle technical enhancements. For example, Amazon Web Services (Aws,amazon Web service) and Google Compute Engine (GCE, Google Computing engine), both have their Linux systems dedicated to the cloud. Rackspace users can choose from a batch of free versions, or you can pay a monthly fee to purchase the enterprise-class Linux for Red Hat software.

Linux is not the only option. Many cloud systems can use a Microsoft Windows system, but require an extra charge. But if you're using Microsoft Windows Azure and Dairyun apps, you don't have to pay a premium, and they want to attract Microsoft stores through the service to help them move more and more computing to the cloud more easily. Any company that is steadily spending money on Microsoft technology can use Microsoft Windows Azure seamlessly. There is also a joyent Cloud, which features a more efficient open source code, called the Smartos (Intelligent operating system).

The deeper differences are manifested visually. While all of these devices are almost indistinguishable from buying Intel processors to putting them on your rack, the fact is that they are a large number of paranoid blades that are cut into virtual machines for ease of use. You don't have to hire a separate device on your own site-you just need to register a set of devices (one of them) or share them over time.

Benchmark Equipment

When you start the timer, the difference between the blades is quickly revealed. Companies have created units of measurement that measure CPU performance to help us, but those are just very rough metrics. With the Java path collected by DaCapo, a superb Java application test, I put my computer in my vast computer kingdom. DaCapo is equipped with a variety of test content for these tests, such as drawing in Java, starting a Tomcat server, and so on. Because each standard can exert different pressure on the device, people who do not use Java can have a comprehensive and broad understanding of the performance of different devices.

The difference in results is so dramatic that it is hard to believe that the same device is doing the same test. This is because different device virtualization layers are different and device drivers vary greatly. The words "different" every time in a particularly funny way, but created a huge difference in the results.

Think of Lucene, the common tool for indexing large collections of text files. In the index creation test, the base SoftLayer device can easily reach more than twice times the basic speed of Amazon. But after the creation of the index, SoftLayer searches the index only a little over 30% faster than Amazon.

The performance of different devices in the same cloud can also be shocking. Google, for example, offers a large number of devices that are not as you would like. The speed of a high CPU device is roughly the same as that of most standard devices, and the quickest and fastest. In the tomcat simulation, Google's devices ran twice times faster. Inexplicably, in the Avrora standard test, its speed is 3 times times slower than standard equipment.

In testing the Google device, the old rules of thumb are always true, but they may not be the truth you expect. Of course they often make mistakes, adding CPUs can help with some tests of multithreading, but occasionally slows down, and adding memory strips can often work but not every time. All of these obvious fixes are more confusing for changes in the speed of testing, and most tests tend to repeat over the same period of time, but some (such as the Xalan parser) vary enormously.

All this means that you need enough knowledge and a lot of experimentation to determine the basic question: how much is the device worth? When the bill expires, we should consider using these basic standards to measure (equipment). If you're creating a Lucene index instead of just searching the index, it's obviously softlayer more valuable. This discrepancy suggests that if we want to maximize our benefits, it is necessary to extend the test time as much as possible. You can't take it for granted that the 3 cents per hour cloud device must be more than the 4 cents per hour device.

Data storage

The place where virtualization affects a lot is data storage. The database relies heavily on the input and output signal speed of the disk drive, which can be reduced by a little extra increase in virtualization. Some clouds don't have much to say about it because they probably think everyone wants to use their own database devices.

Some cloud platforms offer special data storage services that are billed in bytes. SoftLayer, for example, provides specially optimized mongodb for reading and writing data to those individual devices, and performance is certainly better than the MongoDB installed on the enterprise's own devices. The HP Cloud and the Rackspace Cloud provide MySQL services in unison.

Many companies try to do similar work with different database technologies. They reduce the virtualization layer and create APIs to help customers buy storage in bits instead of devices. They believe that a highly optimized operating system is certainly superior to the client's regular machine performance.

Other services emphasize different performance attributes. Amazon launches a package of data storage solutions that they pack bit data to return, but the most interesting cloud services may be glacier, a data storage service designed to retrieve time that might be "hours". It's not milliseconds, not seconds, it's not minutes-it's an hour.

Customers can of course choose another device and install their favorite storage scheme, but this managed solution is enough to tempt you into making a decision. If a cloud platform has your preferred tier of data storage, you can occasionally use other tools.

Network selection

Another troubling theme is the Web. Some clouds-such as Dairyun and SoftLayer cloud-provide a private network connected to devices. It is easy for a customer to create a database device that listens only to this private network command, which also makes the database more secure to evade attacks from the public network. Technology is not perfect because the security of the cloud is still a "cloudy" thing, but a good start.

Some other vendors provide more detailed geographic distinctions about their cloud. Customers know where to put their own devices and make it easy to decide where to store the data. In particular, there are suspicious business people who manage particularly important data. They can create a vast empire of equipment that stores data backups in different locations to better withstand storms, fires and other floods. Google, for example, is well aware of the bandwidth costs of different data centers, so the cost of transmitting data between different bandwidths is higher than the price transferred in the same data center.

Bandwidth metering can cause confusion. Dairyun, for example, does not charge any cost of incoming traffic, simplifies measurement and accounting, and creates a roach motel that specializes in customer information, but does not want output. A similar data plan is attractive if you create a huge data-processing machine, like a visitor from the "Hitchhiker's Guide to the Galaxy" that absorbs large amounts of data and provides only one answer.

Beyond the Basics

The most interesting place in the cloud is the equipment for specialized purposes. Even if they are not very well suited to the PHP code commonly used by current customers, they can deal with future challenges more easily. Amazon, for example, has a set of graphics cards (GPU) for any of the algorithms you dream about and use to perform simple and easy operations. Physicists, biologists and computer scientists have all changed their algorithms to run on this batch of graphics cards. This is just one example of the cloud that makes it easier for us to try out a new architecture.

These special computer devices also do not require special hardware devices at all times. It's cloudy--including Amazon, Joyent, Windows Azure, and so on--to provide special Hadoop devices to meet fanatical needs. They coordinate the underlying operational systems, optimize the JVM, and perform more remarkably. Joyent claims their equipment is 3 times times faster than before. Are you sure? It depends on what you are ordering the device to do. Amazon, in this section, introduces a cloud device that can work directly with Hadoop to help get standby time.

There are other features that ultimately don't matter to me. Some clouds perform better and are more secure. At first they attracted me, but then I stopped paying attention to it. Understanding the overall assembly benefits of the device, but most developers need to hack into their own statistics to gain a clearer picture of the throughput of the device mix. While customer demand may be volatile, other additional performance traits may be required by the customer.

Other similar performance may end up being more important. Some of the latest features on the cloud make it easier for automation equipment armies to change the configuration of each device a little bit. Amazon Cloud allows customers to create hundreds of new machines in the same image, and then configure the information to allow each device to modify itself without having to log in individually to each device and configure the relevant information.

The value of this kind of performance is largely dependent on the type of work being run. If your device is used only for statistical data, there is little difference between the features. But if you want to install and disassemble large units, the ability to automate configuration is important. The expectation of more support for such performance dominates people's choices, and their job is to deal with the occasional bursts of large data.

The most suitable cloud for you

If there is one that can show all of the above courses, it must be the answer will never be cut, no water. Your cheapest device may not be the cheapest for me; your best bandwidth-cost plan may be expensive for me; the underlying standards change as the price of data stores. We are forced by the system to process data and run tests before making a decision.

This part is interesting. The cloud may look like it can solve all the complex problems running on a string of servers, but the provider has to do is solve all the difficult challenges, while opening up the free attitude, allowing customers to choose different infrastructure facilities. Because we don't have to worry too much about backup generators and rack capacity, business choices are becoming more transparent and more convenient. After spending months in my huge machine kingdom, I just realized that I hadn't really done it.