Complex software tools make it possible to move storage from proprietary hardware devices to software-driven services that easily fit into the rest of the software-defined data center. Many practitioners hail the emergence of virtualization as a turning point in data center history, and they're quite right.
The Current State of Storage 7
In addition, the hardware-centric approach creates an inflexible system that cannot always be converted to meet new needs. It prevents easy integration between suppliers, drives up storage costs in the long run, and ultimately slows down innovation.
The Current State of Storage 9
The solution usually involves adding a shelf of disks, but this can be expensive due to the inherent nature of legacy storage and the often high cost per gigabyte of additional storage. Legacy storage systems apply compute-intensive and capacity-intensive services to all hosts in an environment, disregarding the requirements of each guest VM.
The Current State of Storage 11
Instead of depending on rigid hardware constructs to meet the needs of all your workloads, you can apply features and policies through the hypervisor. Because vendors build services on top of the virtualization layer, development must be done independently of the actual hypervisor running in the environment.
Software-Defined Storage Basics
Software-Defined Storage Basics 15
Today, the storage layer is gaining attention among vendors as they strive to build products that reduce or eliminate the challenges inherent in traditional storage. As is the case with SDN, SDS seeks to separate the physical storage hardware from the storage logic that determines where data is placed and which services are applied during read and write operations. The end result is a storage layer that is highly flexible, which can be adapted to meet the changing needs of the application.
Providing such features through software allows them to be expanded and improved over time. Dynamic tiering: Because emerging storage systems generally support a powerful combination of high-performance flash storage and slower, but larger, hard drives, software-based dynamic tiering allows the storage tier to automatically shift data between storage tiers to optimize performance. Snapshots: Snapshots provide a point-in-time copy of the storage system from which they can be recovered.
Software-Defined Storage Basics 17 Virtualization and per-VM
To illustrate, consider the consequences if storage logic depends on using a hardware device such as a RAID controller to provide availability. In this case, all future flexibility becomes tied to this hardware device and the system moves away from software-defined. For example, if a drive in the hardware layer fails, all distributed computing resources can participate in the rebuilding and data recovery process.
Software-Defined Storage Basics 19
In addition to being open from a hardware perspective, a key feature of SDDC and SDS is interoperability facilitated by open application programming interfaces (APIs). A specific application can e.g. need a particular type of storage and could use an SDS vendor's open APIs to provide that storage using the knowledge of the environment gathered by the SDS management layer. As tempting as it may be to label some well-known constructions as SDS, many products marketed as such are not SDS in the true sense of the term.
Although virtualization is a critical component of SDS, it is only a small part of the larger SDS picture. Although federated storage systems can provide some sets of SDS workloads and business functions, they may use proprietary hardware to do so, so they are not true SDS systems. In the real business world, ongoing support for infrastructure solutions is a necessary function.
Software-Defined Storage Basics 21
Given today's powerful processors and underlying hardware, such as flash-based solid-state drives, it should come as no surprise that an SDS solution can keep up with even the largest proprietary storage systems on the market in terms of both features and performance. Given that many legacy storage systems force customers to take an a la carte approach to enterprise storage capabilities such as deduplication, SDS may be able to provide customers with a more comprehensive range of out-of-the-box features and capabilities. The emerging software-driven infrastructure may be based on the advanced architectures of the world's largest cloud players, but you don't need to use a cloud provider to use cloud solutions.
As enterprise-focused companies have worked to transform cloud solutions — such as networking, computing and storage services from Google and Facebook — into components that businesses can use, they've done so with the understanding that customers will generally deploy the solutions to private data centers and not to cloud providers. At the same time, customers will demand on-premise programming and cost-effectiveness to rival public cloud storage services. This is where it becomes most apparent that the emerging SDS solutions were born in the data centers of cloud providers and are now being carried into enterprises.
Key SDS Principles and Enablers
Key SDS Principles and Enablers 25
Block storage: This type of storage provides lower-level protocols for fine-grained data transfers, but lacks metadata and requires a higher-level file system to make it usable. File storage: This type of storage is a higher-level structure that is easier to manage but often difficult to scale. A Windows file server's storage is an example of a storage system that uses file-level storage.
Object storage: This type of storage is well-suited for scaling, but has difficulty supporting data that changes frequently or is transactional. Using Intel x86 processors means vendors don't have to do R&D on processors or other hardware, so they can focus on the software side - the real strength of SDDC. SDS aims to take advantage of storage hardware for which there is nothing factory-defined.
The role of hardware in SDS
Key SDS Principles and Enablers 27
In fact, if one were to simply build a storage system with commodity hardware and without a software management layer, the scale deficiencies of the underlying hardware alone would quickly become apparent. This is achieved by trying to keep data local to the virtual machine node whenever possible. In these cases, Nutanix doesn't just blow up the network fabric by performing a full-on transfer of a virtual machine's storage blocks from one host to another.
When a virtual machine calls storage, the fetched blocks are moved to the new virtual machine location, while the rest of the virtual machine remains on the original node. Over time, the entire contents of the virtual machine will eventually make their way to the new location, but the migration will happen in a natural way with no impact on the network. All of this scaling is achieved by implementing a software layer that manages all aspects of the storage environment.
Key SDS Principles and Enablers 29
Over time, the I/O needs of emerging services have far outstripped the I/O capabilities inherent in the hard disk market. Bold efforts have been made to squeeze every last ounce of performance out of hard drives. But as its cost has fallen, flash storage has become an important part of new classes of storage, such as hybrid storage arrays, all-flash arrays, and SDS systems.
Flash storage: By utilizing solid state drives for reading and writing memory, flash storage can speed up normally slow hard drives. Here's how it works: Keep in mind that hard drives prefer a sequential I/O pattern to maximize performance. Eventually, the software management layer will take steps to reorder this data—what would otherwise be random writes to the hard drives—into sequential I/O, which is then written to the hard drives.
Key SDS Principles and Enablers 31
Short stroking
Key SDS Principles and Enablers 33
If an entire hardware node fails, other storage virtual machines can simply pick up where the failed node left off by using replicated copies of the data. For administrators, the process is also transparent, which is one of the primary design goals of software-designed infrastructure. Decoupled control and data planes: The decoupling of the control plane from the data plane (see Chapter 2) enables distributed decision-making about what resources are provided and how they operate.
Late binding: A programming term known as late binding refers to the ability of the environment to remain flexible. In late binding, these items are software that is provided and managed as part of the runtime operation so that they can be tailored to a specific environment. This active systems approach to infrastructure assumes that all elements of the data center are dynamic.
How SDS Benefits Business
How SDS Benefits Business 37
It also adds additional data center costs in the form of space, power consumption and cooling load. Under the SDDC umbrella, a hardware appliance should be able to provide virtually any data center service (or services). CIOs are looking for every opportunity to simplify their data centers, because simplifying the data center ultimately leads to better service, lower costs and a better focus on the business.
A growing number of CIOs want to take advantage of the cloud, but keep data center resources under firm control of the organization. SDS is equipped to support private cloud initiatives – particularly automated and orchestrated data center infrastructures. Orchestration is a set of services that provides automation, coordination and management of complex data center services.
How SDS Benefits Business 39
Ten Key SDS Facts
Ten Key SDS Facts 43
Regular hard drives are really good for one thing: storing a lot of data at low cost. In SDS, for example, converging flash and hard disk storage together allows IT administrators to use the right level of storage for each application. Decoupling the control plane from the data plane (see Chapter 2) enables centralized decision-making about how resources operate, as well as the efficient delivery of new enterprise-wide services built in software.
It also allows IT departments to use cheaper hardware that is not readily available to lower overall infrastructure costs. The brains of the storage system are centralized, while the data plane, which is responsible for handling lower-level data placement, is distributed but remains under the umbrella of a central intelligence. It not only provides high performance without the cost of proprietary hardware, but provides a historically proven model that significantly increases computing power across generations of compressed products.
Ten Key SDS Facts 45
