Home > Articles > Web Services > Cloud Computing

  • Print
  • + Share This
This chapter is from the book


Virtualization of storage can be done in different ways to make physical storage transparent to consumers of I/O services. Block storage is storage handled as a sequence of bytes. In file-based storage systems, the block storage is formatted with a file system so that programs can make use of file-based I/O services to create and manage files. Virtualization can be done at both levels.

Block Storage

Usually, the person installing an operating system partitions physical hard disks in a physical computer. A disk partition is a logical segment of a hard disk. Partitioning a disk can have several advantages, including separating the operating system from user files and providing a storage area for swapping. The disadvantages of partitioning include the need to reorganize or resize if you run out of space on one partition. The classical example is running out of space on your operating system partition (C:) when you still have plenty of space on the other partitions. One advantage of partitions in virtual systems is that you can plan for a large amount of storage space but do not have to actually allocate that space until you need to use it.

Clouds can make use of partitions as well. In the IBM SmartCloud Enterprise, when you provision a virtual machine, you have an option to create only the root file partition. This optimizes startup time. If you have a large amount of storage associated with the image, the time savings can be considerable. Later, when you use the storage, it is then allocated.

A Linux logical volume manager (LVM) provides a level of abstraction above block devices, such as hard disks, to allow for flexibility in managing storage devices. This can make it easier to resize physical partitions, among other tasks. The LVM manages physical volumes, which can be combined to form a volume group. Logical volumes can then be created from the volume groups. The logical volumes can span multiple physical volumes, allowing them to be any size up to the total size of the volume group.

Copy on write is a technique for efficiently sharing large objects between two or more clients. Each client appears to have its own writable copy of the object, but each client actually has only a read-only copy of the shared object. When a client tries to write to the object, a copy of the block is made and the client is given its own copy. This is efficient when the object is only rarely changed by client programs, such as an operating system when a virtual machine loads and runs it. This technique can make starting the virtual machine much faster than first copying the operating system to a separate storage area before booting the virtual machine. In this context, copy on write is often used with a network-based file system.

The term direct attached storage is usually used to contrast local block-based storage with network attached storage. Direct attached storage is simple, cheap, and high performance. Its high-performance characteristics are due to the fact that it is directly attached. Its disadvantages include that its lifetime is usually tied to the lifetime of the virtual machine. In addition, it might not be scalable if you do not have physical access to the machine. In a cloud environment, you often have no way of increasing direct attached storage, so be sure to start with enough.

In an Infrastructure as a Service cloud, you do not need to be concerned with the different storage implementations the cloud provider uses. Instead, you should be concerned with the amount of storage and the level of performance the storage service provides. Cloud consumers need a basic understanding of the concepts to do informed planning. Generally, local storage comes and goes with virtual machines, and remote storage can be managed as an independent entity that can be attached to or detached from a virtual machine. In general, local and remote storage have a large difference in performance. Remote storage is not suitable for some applications, such as relational databases.

File-Based Storage

File systems provide a level of abstraction over block storage, to allow software to more easily use and manage files. As with block-based storage, a fundamental difference exists between local and remote file systems. Common local file systems in clouds are ext3 and ext4 on Linux and NTFS on Windows. Common remote file systems are NFS on Linux and CIFS on Windows. One huge difference between remote files systems and network attached storage, such as AoE and iSCSI, is that remote file systems are designed for multiple clients with simultaneous write access. This is not possible with remote block devices provided by network attached storage.

Some distributed file systems can span many servers. Apache Hadoop is an example of such a distribute file system used by many large web sites with huge storage requirements. Hadoop is discussed in the upcoming “Hadoop” section in Chapter 5, “Open Source Projects.”

Table 1.4 compares different basic storage options.

Table 1.4. Comparison of Different Storage Options

Storage Option



Local block based

High performance

Lifetime tied to a virtual machine

Remote block based

Can be managed independently, with a lifetime not tied to a virtual machine

Cannot be shared among multiple virtual machines

Local file based

High performance

Lifetime tied to a virtual machine

Remote file based

Can be shared among different clients

Relatively lower performance

The persistence of virtual machines and their local storage can vary with different virtualization methods. Some virtual machines’ local storage disappears if the virtual machine is deleted. In other implementations, the local storage is kept until the owner deletes the virtual machine. The IBM SmartCloud Enterprise uses this model. Some virtualization implementations support the concept of a persistent virtual machine. In a third model, some implementations boot the operating system from network attached storage and do not have any local storage. Be sure to understand the storage model your cloud provider uses so that you do not lose data.

  • + Share This
  • 🔖 Save To Your Account