Cloud Myths

    Recently, I have quite a few clients migrating to the clouds, mostly to Amazon and they are all subject to about the same myths about the clouds. And I have to constantly dispel these errors and misunderstandings. In this article I will talk about IAAS (infrastructure as a service), these are cloud companies such as Amazon, Rackspace, Linode, digitalocean, for PAAS everything looks a little different.

    The clouds are reliable, there will be a good uptime and no backups are needed.

    Clouds relieve some of the problems with hardware, due to virtualization and live migration of virtual machines from problematic machines. But clouds are also prone to network layer and data center or cluster problems. Here you need a couple of examples, quite often the data centers of the cloud provider fall completely and at the same time all the virtual machines of the cloud fall. Sometimes there are network problems and the entire DC is not available or its traffic is sent to a black hole. There are problems with the disk subsystem and sometimes they lead to complete data loss (recently, such letters of happiness came to some Amazon customers, and who did not have backup lost all of their data).
    Western clouds are less prone to the above problems, and those who use Russian clouds probably knew all the pain described above (remember the closed scalaxi, the fall of the clodo, the fakapa selectel).
    The reliability of the cloud lies in the ability to create a distributed system with large cloud providers, choosing different data centers and continents, using convenient methods of load balancing, creating backups and more. That is, if you bought one cloud vpsku, then you still completely tasted all the charms of the cloud, and now all the vpsk hosts have probably become cloudy and are spread over a heap of servers on different continents.

    Clouds are about vertical scaling.

    Top providers in the world still do not support automatic updating of memory and processor on the fly, since these mechanisms were not very simple in older versions of virtualization, and in current ones everything is not smooth, and there are also some problems with memory in linux.
    Here it should be noted Russian clouds that are almost universally capable of vertical scaling. But in general, clouds are built so that they are focused on horizontal scaling, not vertical.

    I launched my application in the cloud and it will scale itself.

    Usually, after dispelling this myth, customers get a little sad, but you can't just take and scale some kind of wordpress by transferring it to the cloud. Any scaling requires certain actions to modify the application infrastructure and often the application code. The simplest is the removal of the code to the file server, the removal of the database and the scaling of the application server, and somehow try to scale the database.

    Are clouds faster than not clouds, or will any particular cloud stand my application?

    Here I try to explain to the client that the megahertz of the processor and the megabytes of memory are the same for everyone and it all depends on how much he is willing to buy them and the price of the cloud. And that there is no difference in the performance of 4 core Dediks at 3 Hz and with 8 GB memory and a cloud with the same parameter, be it Amazon or, but it is the first serious cloud provider with a lot of goodies and dozens of data centers around the world.

    Clouds are cheap.

    There is a myth from Russian users, and here goes the story of how I got into some sort of selectel and it cost me 50 rubles. In fact, clouds are very expensive, using scaling (vertical or horizontal) and having a sinusoidal load, you can save on unused resources two times, only two times, while the clouds are sometimes ten times more expensive than the same cheap Dedics.
    To save a Dedic with 32 gig of RAM, 8 cores, it costs hetzner 70 euros, a similar config in Amazon Linux on cc1.4xlarge will result in almost $ 1000 without traffic and disks.

    Clouds it is not clear

    Yes, there is a certain threshold for entry, mostly associated with differences in terminology and options for work.
    For example, all clouds are essentially ordinary vps in a cluster and I simply decrypt the terms
    ec2 instance - vps
    ebs - hdd
    elastics ip - dedicated static ip for vps
    and so on

    The clouds are safely moved and all is well

    In general, clouds provide a certain number of mechanisms for protection, but in general, all protection from everything falls on your shoulders and you do not become more secure in the cloud. Well, in the clouds sometimes prose pass pci compliance to accept credit cards.

    Of the non-myths, I note the slow disk subsystem of almost all clouds, this is well understood when you raise the instance to ssd clouds like digital ocean. In Russia, the clouds are worse than in the west due to the lesser infrastructure and software testing, that is, Russian clouds are younger and still cones are still filling.

    If you are going to migrate to the clouds and you have questions feel free to ask them in the comments
    or email

    Good luck in the clouds.

    Also popular now: