The Age Of Virtualization.

Virtualization brings the true benefits of SAN.

A portion of this article appeared in Storage Inc. (Q4 2000, p. 18). It appears here in its entirety.

These past two years have seen glowing analyst predictions of explosive SAN deployment. Is the Storage Area Network (SAN) living up to its promise? The honest answer is: "It depends". In its narrowest definition, any fibre channel-attached storage device can constitute a SAN. This protocol-based definition is used by a number of RAID subsystem and interconnect vendors for publicity reasons. But this is not what users have in mind, nor what the promise is.

As the name implies, a SAN is a network: a fast, scalable, reliable, manageable way to interconnect a vast number of heterogeneous hosts and storage resources. These characteristics are what make SANs a very attractive concept for a majority of businesses:

* Speed of delivery and low overhead, combined with a reduction in LAN traffic, improve application response time and enable execution of very demanding Service-Level Agreements (SLAs).

* Scalability is critical to deploying new applications and sustaining organic growth when demand for capacity doubles every year.

* Manageability allows resource-strapped customers to deliver the above benefits on time within an acceptable Total Cost of Ownership (TCO).

So why is the SAN stuck? Simply because it has not delivered the management and bottom line benefits it promised. A recent Computerworld survey confirms that the two main inhibitors to SAN deployment are high implementation costs (38% of respondents) and lack of staff resources (36%). In other words, SANs have been a bleeding-edge technology, not a business proposition.

In 2000 and early 2001, experts agree: virtualization provides the missing ingredient bringing you the true benefits of the SAN. More and more vendors are jumping on the bandwagon. Should you?

What Is Virtualization?

Virtualization is a simple, time-honored idea: a client requesting a service does not care, or want to know, about what the server needs to do to fulfill the request.

Virtualization has been an integral part of computing for several decades. Virtual memory, the ability for a system to serve more memory than it physically has, has spelled success for IBM's 370 series and MVS operating system. Closer to us, Web serving performs two levels of virtualization: first by mapping a URL to a physical server's map address, then by building the resulting page on the fly from a collection of image files, pointers, and other embedded functions.

Virtualization also applies to telephony. When you dial my cell phone number, the network searches, then maps it to my particular location at the time. If I do not answer, it remaps it to my voice mail.

Storage networking has been remiss in not implementing these techniques. An administrator still needs to manually assign each RAID's address to a server or Host Bus Adapter (HBA). SAN addressing allows for connecting 16 million devices, but what administrator can manage and maintain the combinations?

This is where storage virtualization comes in. Virtualization brings the SAN to functional parity with the Internet or switched telephone network by performing a series of tasks:

* Discovering SAN objects.

* Enabling the administrator to sort all devices into "pools" according to service-level, location, or any other criteria, to carve out virtual "volumes" and assign them to application servers.

* Maintaining the server-level map of virtual-to-physical relationships.

Nice. Now what's in it for you, the user?

What Virtualization Can Do For You

The first benefit is a better utilization of your storage assets. Large data centers are typically built around racks housing tens or hundreds of Windows/NT or Unix servers. Each server has its own storage, with capacity dictated by vendor models available, predicted future usage, and safety buffers to absorb any peaks. The result? Average capacity utilization is about 30 to 35% for NT servers, 38% for Solaris. In other words, two thirds of your largest data center investment is idle. With a SAN physically regrouping these islands of capacity, and virtualization pooling all your unused capacity in a single "volume", you can start using that untapped potential. In concrete terms, virtualization will usually save you over a year's worth of storage purchases.

Better still: virtualization can pool any of your SCSI or fibre-attached storage, from the JBOD up to the most sophisticated enterprise RAID subsystems. You can integrate your existing assets, mix heterogeneous technologies, and evolve your storage and server assets independent of each other, transparent to the application. Virtualization keeps your existing assets working longer, and enables you to introduce new technologies with minimal regard for backward compatibility.

Virtualization also increases server uptime and productivity by eliminating much of the reassignment/reorganization downtime, enabling server-less backup without interrupting operation, and also matching logical pool features (capacity, bandwidth, non-stop attributes) to each server or application's target service level.

Does your company have aggressive plans of capturing new markets, increasing quality of service, rolling out new applications? Virtualization brings the promise of "just in time" responsiveness.

Bringing in additional storage for a new application is a complex task: checking revision levels and compatibilities, configuring and masking LUNs, reorganizing volumes, reconfiguring and rebooting a number of servers, updating your backup plan. Two weeks' work is the norm; a month is not uncommon.

Contrast it with this: you roll in a new storage subsystem, plug its fibre cable into your fabric switch, map it to a set of logical drives on a GUI (you can choose to slice it into any number of logical units, or conversely to pool it with any other devices), and press "enter". Done. Your servers now see a larger volume and your backup plan already has this volume listed. Fifteen minutes. With testing and optimization, plan for one day. Now your data center can back the most aggressive growth plans--and turn on a dime if any adjustment is called for. Virtualization is the delivery vehicle for storage on demand.

Large data centers are faced with an exponential increase in storage management workload due to massive capacity growth, fragmentation of servers and storage, and increased overall complexity. Virtualization brings down the complexity of configuring, assigning, optimizing, and managing storage several notches. According to early studies, an administrator can manage between four and forty times more capacity by using virtualization.

Return on investment, business responsiveness, administrative productivity: virtualization packs a punch.

Planning Your Virtualization Project

Congratulations, you have decided to implement virtual storage. Now what?

As usual, your first task will be to cut through vendor hype. Every vendor claims or prepares to announce some form of virtualization. Are they all identical?: Will they give you the same benefits? The answer is a resounding no.

Server-based storage virtualization has been in existence for many years. Products such as VERITAS' Volume Manager have brought tremendous benefits to their users, for instance in optimizing Solaris clusters for serving large Oracle databases. But they do not claim or deliver any benefit in heterogeneous servers or storage integration, data center consolidation, or storage on demand.

Likewise, enterprise RAID subsystems also offer some virtualization capabilities. But these are designed to enable easier allocation within one homogeneous subsystem. Here again, no consolidation, server transparency, or multi-vendor reach.

To be effective, storage virtualization must be located where it can survey and manage the whole storage network (servers and storage), and not be based or dependent on any of these components. Two solutions for this: either introduce a redirection device within the storage data path (this is called symmetrical virtualization) or connect an appliance that sits outside the data path and tells each server what it should "see" (asymmetrical design). All storage virtualization products are either symmetrical (e.g., DataCore's SANsymphony) or asymmetrical (SanOne's Dynamic Allocation Manager). Your next step is choosing one architecture which fits your needs.

An attractive feature of symmetrical virtualization is that it involves no server-based code and can be implemented across any heterogeneous operating systems. These benefits are offset by limited scalability and the increased fragility introduced by flowing all storage access through an additional device (usually NT based): a device failure brings down the whole data center.

Asymmetrical virtualization, by contrast, can scale to any level, as the appliance does not handle any data flow or physical connections. Also, this design allows several appliances to easily cooperate in active fail-over. Should all appliances fail simultaneously, the data path would still remain open and unaffected. This makes asymmetrical design very robust. Asymmetrical virtualization involves a handshake with servers, which is now available for the most popular operating systems, such as Windows/NT, Windows 2000, and Solaris. Some exotic or older operating systems may not be supported. In the long term, most analysts predict that asymmetrical designs will prevail because of their innate scalability and robustness.

Your virtualization software must work with your other storage applications, such as replication, backup, and storage resource management. Your next consideration will be for the breadth of functionality enabled by each product. Good news: some virtualization products such as SanOne's DAM include a snapshot and mirroring capability, enabling you to take any number of "frozen" images of your data for site replication or test purposes. Similarly, virtualization vendors have been working with backup software providers to cross-certify their products. The situation regarding event monitoring and storage management is more complex, as these disciplines are still in their formative stages themselves, and management integration of your SAN is still best served by a virtualization-savvy multi-vendor integrator.

The last step is implementation including data migration. Here again, a number of vendors and service providers including SanOne have tools and techniques to deploy virtual storage and to reload your data into your target environment.

Should virtualization be your 2001 priority? If your storage is growing only modestly and you are not faced with major consolidation, heterogeneity, or scalability issues, you cant probably afford to wait another six months and be content with surveying vendor offerings and the results of early adopters. If capacity growth, data center consolidation, asset utilization, or heterogeneity keep you awake at light, do yourself a favor: bring virtualization in--now.

Mike Flannery is the general manager of San One (Phoenix, AZ).