While server hypervisors have become a well-defined and established server virtualization solution, the term “storage hypervisor” may raise more questions than it answers. Whether a particular solution can be classified as a storage hypervisor is a subject of much debate, but to truly earn the “hypervisor” moniker, a storage hypervisor must virtualize storage hardware – that is, it must consolidate the resources from multiple storage devices into a common pool. However, it is not clear what capabilities are needed to supplement the virtualization function. IBM and DataCore believe that a strong management platform is key. Virsto, with its recent announcement of support for VMware, highlights the need for significant I/O optimization. These attributes are both needed to create a complete solution, and the ideal storage hypervisor would merge them into a single product.
During 2011, several vendors, including IBM, DataCore, and Virsto, highlighted technologies and product packages that they describe as storage hypervisors (HP describes its solution as storage-hypervisor-like). Whether or not a true storage hypervisor exists, among the products that are marketed as storage hypervisors, there are essentially two schools of thought. The first is management consolidation; the second is performance through workflow management.
IBM’s Storage hypervisor and DataCore’s SANsymphony-V, both storage hypervisors according to their respective vendors, apply the first school of thought.
IBM: TPC and SVC
IBM’s solution is comprised of the Tivoli Storage Productivity Center (TPC) management suite and SAN Volume Controller (SVC) virtualization appliance. Several other Tivoli software products are also included as part of the IBM Storage hypervisor, but the majority of the key features and functionality stem from these two products.
The SVC appliance offers a number of storage-enhancing features covering a wide range of customer needs, from disaster recovery to improving utilization. Easy Tier, for example, is an automated tiering solution for performance optimization in mixed HDD and SSD storage pools. SVC also includes both local and remote copy services and will thin-provision managed storage. While the additional features are valuable options, the virtualization capability remains SVC’s core value proposition in the context of the IBM Storage hypervisor.
Like SVC, TPC includes a host of functions for managed storage. The Standard Edition, boasting the full suite of capabilities, offers end-to-end storage resource management options, from enterprise-level reporting to drill-down performance management. Many of these options complement SVC features, such as enterprise-wide replication management of SVC copy services and Storage Tiering reports utilizing data collected by SVC. However, TPC’s chief value is in its far-reaching and consolidated management, which offers a single management console for all storage operations.
DataCore: The Software-Based Solution
Like the IBM Storage hypervisor, the primary features of DataCore’s SANsymphony-V are storage virtualization and full-featured management, providing a single point of management for provisioning, reporting, and performance management within a single platform. Unlike the IBM Storage hypervisor, DataCore’s solution is entirely software-based, utilizing Windows-based servers to power the product. This allows administrators to easily configure the amount of cache and computing power available. SANsymphony-V also offers a full suite of storage capabilities, including copy services, thin provisioning, and automated tiering.
Although both the DataCore and IBM storage hypervisors offer a wealth of features, the primary message is virtualization, followed by comprehensive management. The performance enhancements, high-availability options, and other capabilities are highly useful, but their value would quickly diminish without capable and centralized management.
Virsto: Focus on Performance
Virsto's storage hypervisor applies the second school of thought – performance through workflow management – to its solution. While Virsto also virtualizes storage, it focuses primarily on getting the best possible performance out of storage. Virsto’s storage virtualization software initially supported only Microsoft Hyper-V, and it now supports VMware ESX as well. The software installs as a virtual storage appliance (VSA) and improves the way a server hypervisor interfaces with storage by acting as a kind of regulator for block-based I/O. Administrators can continue to use the management functions of the server hypervisor while Virsto operates in the background, transparently optimizing the placement of, and access to, data blocks on storage devices. Thus, instead of creating a new management console, Virsto leverages the hypervisor’s console, VMware vCenter or Microsoft Management Console (MMC), through a plug-in.
When Virsto is installed, the server hypervisor is presented with Virsto vDisks, which, to the server hypervisor, appear as standard “thick” virtual machine disks (VMDKs) to an ESX server, or “fixed” virtual hard disks (VHDs) to Hyper-V. While both of these objects appear to the server hypervisor as fully-allocated NFS data stores, in fact, Virsto vDisks are thin provisioned and maintain the performance of thick volumes through regulation of the I/O path.
Normally, deploying a number of virtual machines (VMs) causes significant storage performance degradation due to what Virsto calls the “I/O Blender.” In this phenomenon, the I/O patterns from VMs are consolidated into a single I/O stream, creating a highly random, and typically, write-intensive workload for the underlying storage.
This is where Virsto comes in. When the server hypervisor sends a write request to storage, the request is actually sent to the Virsto vLog. By utilizing a log in front of primary storage, Virsto can immediately acknowledge write requests to speed response time and reorder data into sequential blocks before finally destaging data to storage. Read requests are also improved, since data for each VMDK or VHD is arranged in sequential blocks by Virsto mapping algorithms rather than being strewn about the storage pool in smaller pieces in a write-anywhere approach. Overall, it is typical for customers to see a tenfold improvement in performance utilizing the same hardware, in addition to improved capacity utilization through thin provisioning.
Note that in addition to approaching the storage hypervisor problem from different angles, the two approaches also view server hypervisors differently. The DataCore and IBM products support and integrate with server hypervisors from leading vendors through plug-ins. Virsto has chosen to embed itself entirely within the server hypervisor, effectively extending the server hypervisor’s storage capabilities. Though Virsto’s approach may ignore the portion of the market that doesn’t use server hypervisors, it also represents a level of integration and management consolidation that is difficult to match. This unique implementation also allows Virsto to manage storage at the VM level, rather than at the LUN or volume level. Administrators can snapshot, migrate, and replicate individual VMs rather than having to drag along the rest of the volume with them.
The basic question that any storage hypervisor needs to address is how it can help users get more out of their storage. Virsto has shown that it may not be enough to continually improve and tweak the status quo. There have been major changes in the way data is handled; changes that create bottlenecks in existing storage architecture. The products from IBM, DataCore, and Virsto cannot yet match all of the virtualization benefits that are possible with server hypervisors, that is, superior utilization, flexibility, performance, and manageability. Current storage hypervisors take important steps towards reaching some of these goals, but it will be the next generation that delivers on all of them in a single solution.
Comments