The American IaaS "8 Wolves"

We "will help you forget the complex system, get rid of server installation trouble, no longer worry about backup, and no longer because of all kinds of mistakes caused by loss, concentrate on the perfect enterprise core competitiveness." As long as you take the data and credit cards out and pay a small fee for yourself, the rest of the work can be done by the cloud server.

This propaganda has become the slogan of cloud computing. More tempting is its convenience (to buy when it is needed, to pay when it is needed), and to be cheap (most services and sweets are about the same price, less than 5 dollars).

However, the industry "outside" people may imagine the "cloud computing Services": are identical, or as a Lego toy can be as random collocation. But that is often not the case, and these services are generally "colourful". On the one hand, is because cloud computing service provider's propaganda demand, hoped that their service compares with the competitor to have the difference; On the other hand, in order to subdivide the market demand, the different product and the service difference aggregation together, certainly also has the very big difference.

To this end, the 8 most important IaaS service providers in the United States: AWS, GCE, Azure, Rackspace, SoftLayer, Dell Cloud, HP Cloud, Joyent,infoworld have issued a horizontal evaluation report.

Benchmark Test

The first is the operating system. Although Linux is already ubiquitous, various standard editions, distributions are available, such as Ubuntu. But some companies still create their own versions, even if only marginally or in general.

AWS (Amazon Web Services) and GCE (Google Compute Engine) have their own versions of Linux. Rackspace can choose a number of free versions or a monthly fee like Red Hat Enterprise Linux. Linux is not the only option, and many cloud service providers can offer services such as Microsoft Windows, but at an extra charge. Of course, Microsoft's Windows Azure and Dell Cloud can be excepted (they already provide related services). In addition to Microsoft and Dell, these companies want to move into their own cloud through services such as the Microsoft Store, which have already invested a lot in Microsoft technology. But it's not easy. The other, Joyent Cloud, formed its own smartos through the transformation of the Open Solaris.

Secondly, it is not easy to see the difference. It may be argued that the bottom tier of these services is that Intel hosts are placed in their respective racks, but the fact is that a small portion of the high-performance hardware that has been split by different virtualization software is the service base for consumers. In other words, you rent the land, just one of the apartments or timeshare (timeshare).

Finally, the specific data should be based on the benchmark test.

When the timer began, the differences between the machines began to become apparent. These companies are trying to help us create units to measure CPU power, but clearly these guidelines are sketchy. Testing with DaCapo collection of Java routines is clearly appropriate for building Java applications tests.

The DaCapo suite tests different tasks as creating images and starting Tomcat servers in Java. And for those who do not use Java can also be seen by the functional comparison of the general view to get some reference, after all, each benchmark will be on the machine product a different stress test.

Detailed comparison results are shown in the table below.

Be aware of virtualization

The results of such dramatic comparisons are hard to believe in the same test that the same machine runs. Chasing the root cause, or because different machines use different types of virtualization layers to handle different device drivers. When these "differences" are applied in an interesting way, the resulting results are naturally quite different.

In considering Lucene, a common large number of text document indexing tools. When you create an index test, the SoftLayer provides at least twice times faster than the Amazon basic instance. But after the index was built, SoftLayer was only a little over 30% faster than Amazon when searching for indexes.

(Responsible editor: Fumingli)