There's no shortage of buzz about networking server-attached storage to create new networks from existing DAS, SAN and NAS. See if it's right for your shop.
The recent proliferation of available storage technologies has led to an abundance of choices for storage architects to select from when building their ideal solutions. Product differentiation gives users distinct options, which has long been the case for storage managers. However, gone are the days of a clear distinction between DAS, NAS and SAN -- traditionally considered the three pillars of enterprise storage—as software-defined storage gains interest.
An analogy might be a color wheel where blue, red and yellow are unique. The addition of software-defined storage (SDS), data-defined storage (DDS) and object storage takes attributes from the base "colors" and blends them to create a spectrum of storage technologies. While this increases IT consumer choice, it also leads to confusion as to where one technology starts and another ends. In addition, vendors may eschew technological purity and adopt different labels for marketing benefit, thereby exacerbating the muddle.
SAN and NAS still prevalent
SAN and NAS remain the mainstays of most commercial applications, serving small and medium-sized businesses (SMBs) through enterprises. For more than a decade, DAS has been used only for local operating system (OS)-related needs and very small, local applications. Veteran storage managers can remember when general-purpose computers were used as storage controllers in the late 1990s and early 2000s. These products disappeared from the market largely because specialized ASIC-based controllers offered such substantial performance improvements as to overcome the simplicity of PC-based implementations. SDS may be built upon general-purpose components, but it's not "back to the future." Although one might conclude that SDS is targeted at SMB consumers, most vendors position themselves as enterprise suppliers where performance will be of paramount importance.
Quick Q&A: Server-attached storage
What does server-attached storage mean?
Server-attached storage refers to the continuum of storage-related software used to orchestrate DAS, NAS and SAN into a seamless storage topology.
What's the most important element in server-attached storage products?
Generally speaking, it's the clustering functionality of the storage nodes that bring DAS, NAS and SAN into one environment. Most products, but not all, use a proprietary file system or clustered file system. Regardless, the clustering capabilities provide common functionality and resilience across the systems and constitute the secret sauce for most products.
What's the key difference between server-attached storage and traditional storage systems?
Server-attached storage are all-software products that don't rely on specialized ASICs for functionality. Although some server-attached products may be delivered in a proprietary manner, they all operate on commodity componentry.
Definition of SDS is elusive
One effect of the "blending" of storage technologies is the mere notion of server-attached storage itself. In a way, all servers are attached to storage, whether directly or indirectly through a network. By the same token, storage internal to a physical server may be shared with other physical servers across a network. Perhaps the lack of a clear definition is for the best, because a definition is inherently limiting. To extend our color analogy, a point on a spectrum can define a color, but a color can't be used to describe a spectrum. Similarly, we can describe a range of technology configurations, but one product or technology doesn't define an industry. Here, we'll examine a variety of offerings from both established and emerging vendors as illustrations of the breadth of the server-defined storage market.
Hypervisor rising as storage host
The dominant emerging platform for server-defined storage is a set of storage services built on top of a hypervisor. Not surprisingly, emerging vendors are the fastest to adopt this architecture, as they have the fewest legacy considerations and the least intellectual property tied to proprietary hardware designs. Those offerings may variously address DAS, NAS and SAN. Thus, they virtualize the physical storage not just at a device level, but at an architectural level. How they do this is the secret sauce to almost every product. Therefore, IT decision makers should fully understand the clustered file system or other clustering methodology of each product, as that element differentiates it and determines its suitability for a specific purpose.
Sanbolic Inc. Sanbolic Platform
Sanbolic's Platform is one of the emerging solutions built as a "kernel-mode" platform on top of a hypervisor (VMware, Hyper-V or XenServer), though it will also run on a physical server. With all storage services controlled at the hypervisor level, it doesn't matter where the data physically resides in a specific storage architecture. Sanbolic can serve mixed block and file data from the same machine. The built-in file system and volume manager virtualize the storage hardware and create a distributed architecture. Storage services such as RAID, replication, high-availability (HA) failover and volume management are controlled from a single point. This includes Microsoft Volume Shadow Copy Service (VSS), by virtue of the integration with Windows-based servers. Sanbolic is a scale-out architecture, supporting up to 70,000 storage elements. Depending upon version and options, an active/passive HA arrangement can be configured. Sanbolic claims to be entirely hypervisor, server and storage device agnostic. Common workloads include file serving and virtual desktop infrastructure.
Gridstore Inc. Hyper-V
Gridstore's Hyper-V comes with Virtual Controller (vController) technology, but it's optimized for Hyper-V. Gridstore differs in several other important ways as well. First, Gridstore Hyper-V is delivered pre-packaged with Dell 1U servers for simple implementation and scaling, but it supports common-off-the-shelf (COTS) storage nodes on the back end. The key advantage of being tightly integrated with Hyper-V is that Gridstore knows which virtual machine (VM) is generating the I/O requests and automatically tunes itself to fit customer-defined Quality of Service (QoS) parameters on a per-VM basis. Gridstore leverages Windows' storage services to provide snapshots, deduplication, thin provisioning and so on. The company also has a relationship with Vision Solutions Inc. to provide Double-Take technology for data replication. The key factor in its ability to blend different storage architectures is that it makes LUNs look like physical drives. This makes navigating storage more intuitive to Windows users. As a best practice, the company recommends using three or more nodes to provide greater efficiency and resilience. Gridstore's top three use cases are as primary storage for Hyper-V implementations, as a backup target and for SMB operations.
Hewlett-Packard Co. StoreVirtual Virtual Storage Appliance
Hewlett-Packard (HP) is an interesting case where an established company has assembled an SDS palette from its SAN/NAS lines, servers, acquisitions and industry standards. HP has used its acquisition of LeftHand Networks to combine elements into its StoreVirtual Virtual Storage Appliance (VSA) SDS product. VSA can be purchased as standalone software installed as a VM on either VMware or Hyper-V, or as appliance. It can federate storage from DAS, NAS or SAN. It interfaces with other systems using industry standards from OpenStack and the RESTful API. Though VSA doesn't use a clustered file system, it does have a quorum HA methodology for resilience. All the storage services that you would expect from an enterprise-class storage system are available. In addition, VSA leverages HP's Adaptive Optimization capabilities to provide automated tiering when solid-state drives (SSDs) are included in the storage mix. SMBs may consider using VSA as a primary storage system, while enterprises may use it for remote office storage with centralized management using VMware vCenter and Microsoft System Center plug-ins. HP also targets service providers looking for an all-VM service offering. Recently, HP began shipping all ProLiant systems with a free license for the 1 TB version of VSA, which allows up to three servers to be clustered at no additional charge.
Tarmin Inc. GridBank
Tarmin doesn't use the SDS label, rather it describes its products as DDS. The key difference between DDS and SDS, at least with respect to Tarmin's GridBank product, occurs at data ingestion. GridBank's ingestion process includes full text indexing, object ID and referential metadata. GridBank is similar to SDS in that it includes many storage management functions and, like SDS, it's media-independent, supporting a variety of SSD and hard disk drive systems plus tape. Its functionality is enabled by a proprietary distributed file system for peer-to-peer networking, failover, deduplication and encryption. Tarmin recommends a reference architecture of at least two nodes in a scale-out configuration, although four is more common. GridBank is storage agnostic, but the company partners with Dell Inc. and IBM, which offers an IBM XIV grid-based storage system model. Because of the data ingestion process, GridBank is targeted at big data analytics and Hadoop clusters.
NetApp Inc. FlexArray
NetApp is another established vendor reworking its product lineup to address the server-attached storage market. The company's recently announced FlexArray is a software-only product, but it's a replacement for the current V-Series arrays in combination with the new FAS8000 Universal Data Platform. Although a FAS8000 is necessary in the configuration, Data Ontap Edge is now available on Clustered Data Ontap, which can be installed as a guest in a hypervisor. FlexArray can address NetApp, EMC and Hitachi Data Systems arrays, bringing the full functionality of Data Ontap regardless of the hardware back end. Flash Pools and Flash Cache are supported, but not on non-NetApp arrays. In any event, it makes the most sense to put SDS into the FAS8000, where full data orchestration can occur. FlexArray has elements of SDS in that it's a software product supporting a variety of storage systems; yet it retains many elements of more traditional storage virtualization. It's designed for storage consolidation opportunities and applications where heterogeneous data replication is required.
EMC Corp. Isilon
EMC's Isilon is touted by the company as "commodity hardware with intelligent software" and is targeted at hyper-scale clients. Although it operates on an x86 platform, the software isn't installed on a hypervisor. Rather, it's a Berkeley Software Distribution (BSD)-based kernel and is presented as a file system. By not utilizing a general-purpose OS, the kernel strips all functionality not related to storage and therefore has no wasted CPU cycles. The software also provides a single point of storage orchestration, though that doesn't presently extend beyond Isilon systems.
DAS, NAS and SAN previously represented storage siloes, albeit well defined. The new generation of SDS, DDS, VSA -- and undoubtedly more acronyms to come -- offers a central orchestration point for the diversity of storage architectures found in most organizations. Whereas SAN and NAS abstracted low-level storage issues from application developers, server-attached storage products abstract the architectures as well.
About the author:
Phil Goodwin is a storage consultant and freelance writer.
- CW ANZ: Taming the data beast –ComputerWeekly.com
- IT Definitions: Storage Special –ComputerWeekly.com
- Taking Steps to Prevent Your BC/DR Plans from Being Derailed –DataCore Software Corporation
- The Multi-cloud Storage Management Essentials –TechTarget