Introduction

Virtualization is transformational to the way that IT services are delivered—it will fundamentally change how businesses use, access, and pay for IT services.  Why? Once virtualization is implemented in a specific area of IT, it becomes viral.  Virtualizing a specific IT asset leads to virtualization of other IT assets … and so on.  It leads to more efficient delivery of IT services, which, in turn, creates opportunities to extend virtualization and transform the environment for business impact.

Situation Analysis

The number of challenges faced by IT organizations runs the gamut from the high cost of delivering services, to not being able to keep pace with data growth, to security vulnerabilities, to energy issues and more.  The economic climate adds pressure to IT budgets and resources.  However, IT organizations are still expected to deliver cost-effective and quality IT services in response to new business initiatives—with agility.

IT management is most concerned with the “IT triad:” mitigating risk to existing business activities, reducing cycle time to deliver innovation, and reducing costs to fund innovation.  It’s often difficult to achieve all three of these objectives, so some organizations may make sacrifices in one area to deliver in another.

Ideally, IT should be directed toward business initiatives that provide a competitive edge.  That could be delivering business applications, creating new business channels, improving processes for employees and business partners, and more.  Unfortunately, IT’s need to “keep the lights on” and simply run the business often take up 70% of time and resources, leaving only 30% for changing-the-business-to-be-more-competitive activities.

Oftentimes, the imbalance is a result of IT sprawl.  Silos of legacy, inflexible hardware and software stacks are inefficient and costly.  These resources are often under-utilized and create multiple points of failure.  The situation also results in operational inefficiency: managing more infrastructure, reactive “firefighting” activities, and delivering baseline service levels.

The reality of the situation creates conflict between the business and IT.  Line of business (LOB) owners desire rapid return on investment (ROI) and higher service levels.  To achieve high service levels, IT might over-provision or create dedicated stacks to guarantee performance.  However, this is in conflict with ROI requirements and LOB owners’ (and IT’s) needs to reduce costs and achieve business agility.

The result?  Over-stressed IT teams are struggling to meet IT triad goals while simultaneously maintaining high quality of service for their business constituents.  Likewise, business constituents could have low satisfaction if quality of service needs are being sacrificed by IT in order to address other concerns.

That’s where virtualization comes in.  Virtualizing resources breaks down the silos.  This, in turn, can reduce costs, improve operational efficiency and agility, and help mitigate risk.  A virtualized infrastructure can also contribute to IT’s ability to maintain high service levels. Finally, virtualization can facilitate a flip-flop on the 70/30 maintain/innovate ratio.

Virtualization

What is Virtualization?

In simplest terms, virtualization is an abstraction of reality—the process of abstracting the logical from the physical—to turn physical assets into a pool of independent logical assets.  As previously mentioned, this can be transformational for IT since it alters the way services are delivered, which enables transformation of the business.

Virtualization Maturity Model

So how is this transformation achieved via virtualization?  It’s an iterative approach that happens over time.  Small steps lead to benefits that enable next steps, and so forth.  As organizations progress through specific phases, greater cost optimization and process efficiency accompany the progression.

To illustrate the concept, ESG’s virtualization maturity model details the characteristics of the environment at distinct phases in the virtualization journey (see Table 1).

The first phase, basic, is to initiate resource consolidation and centralization through virtualization.  The resulting logical resource pool can help in eliminating single points of failure and reducing vulnerabilities, as well as improving utilization of resources. A shared infrastructure may facilitate automation, streamline management, and simplify IT’s ability to monitor and analyze trends.  This phase can drive the next one: standardization.

The aptly named standardization phase involves creating standard configurations, processes, and policies.  Policy development and adherence can deliver automation and more predictable budgeting as well as greater predictability for service levels.  The result for IT may be process improvement, enhanced management, and greater business resilience, quality of service, and responsiveness.

As IT organizations progress to the enterprise phase, virtualization and its benefits become more pervasive.  Service delivery is service catalog-driven and IT may have initiated a utility computing model.  Business policies are understood and aligned with IT.  Automation and self-service are more often in place.  Business constituents benefit from predictable service levels and costs—which is likely to result in higher satisfaction.  The IT triad objectives of greater agility resulting from process improvements, minimized risk, and cost reduction are within reach.

The culmination of a highly virtualized environment is the dynamic phase, where operations are highly automated and based on a utility model.  Resources are dynamically provisioned and used, and service delivery is tied to business-linked service level agreements (SLAs).  IT benefits from an enterprise-wide-view of resources, which streamlines planning and management.  All combined, dynamic-level virtualization enables the business agility companies desire to gain competitive advantage.

Extending Virtualization Principles

The most predominant form of virtualization today is server virtualization.  Some 76% of organizations with 2,500 or more servers are currently using virtualization in a production environment, compared to just 27% of organizations with less than 25 servers (see Figure 1).[1]

Server virtualization is expanding in two dimensions: consolidation of more physical servers onto virtualized platforms and expansion of the number of applications running on virtual machines.  As server virtualization expansion follows the aforementioned maturity curve, the dividends of the virtualization journey pay off.

Storage virtualization is often combined with server virtualization. Storage virtualization (functionality that takes multiple storage systems and treats those devices as a single, centrally-managed pool of storage) provides an additional opportunity to reduce IT costs and optimize efficiency—especially as server virtualization proliferates. ESG research indicates that among current server virtualization users, 86% report they have storage virtualization initiatives as a top spending priority over the next 12 to 18 months (see Figure 2).[2] The combination often amplifies the benefits of both forms of virtualization: lowering costs, improving resource utilization, increasing availability, simplifying upgrades, and enabling scalability.

As organizations become more comfortable with one form of virtualization, they don’t have to make great intellectual or philosophical leaps in order to grasp the concept of virtualizing other data center domains.  Oftentimes, IT organizations undertaking complete data center refresh initiatives have virtualization as top of mind and would look to extract all possible efficiencies in one fell swoop by deploying virtualization in multiple technology areas.

Creating Efficiency and Business Agility

As organizations mature in their use of virtualization, time and cost efficiency enable further innovation.  For example, virtualization could extend to leveraging the elastic and potentially infinite resources in cloud computing models.  In this case, compute, storage, communications, and applications are abstracted into services and delivered over the Internet, enabling cloud-based, on-demand computing.

Cloud computing can be IT-owned and -operated or leased on a subscription basis (usage-based) from a third-party.  In a cloud computing approach, IT’s capital expenses are converted into operational expenses.  Because IT organizations can more easily partition and provision resources in a multi-tenant environment, this aids in providing customer-specific SLAs.

As more organizations virtualize the infrastructure layer in corporate- or third-party-data centers, IT service delivery becomes a reality.  A shared resource model facilitates a shared services model.  Instead of the application silo approach, applications are de-coupled from the infrastructure and are delivered via a utility model.  An application services approach can offer increased availability considering that infrastructure provisioning is dynamic and application workloads with greater criticality could be prioritized over less critical ones.

Implications for Data Protection

Virtualization of servers and storage is a catalyst for changes in data protection, too. These abstractions allow server and storage resources to be used more efficiently to capture and store data copies for operational and disaster recovery.  This can result in changing both the way that data protection services are delivered and the cost structure.

Encapsulation and portability of the operating system, applications, data, and configuration setting in virtual machines enables mobility within and across data centers.  Transitioning from on-premises, siloed infrastructure to an environment that includes IT-owned-and-operated or third-party shared components creates new possibilities for protection strategies.  Data protection solutions can capture and transfer active virtual machine disk files as a whole, enabling rapid local or remote recovery.  Furthermore, a virtualized, shared infrastructure eliminates IT’s dependence on one-to-one, physical-to-physical mirroring strategies and creates an opportunity to scale availability and recovery architectures cost effectively.  This outcome is clearly resonating with end-users.  As shown in Figure 3, ESG research respondents ranked “facilitated disaster recovery” as the number one benefit of deploying server virtualization.[3]

Applying virtualization to data protection resources can result in benefits similar to those seen in server and storage virtualization environments: consolidation, automation, enterprise-view of resource management, higher utilization, etc.  Importantly, it can also address IT triad objectives by minimizing risk, delivering more predictable service levels and costs, and maintaining agility. ESG research respondents ranked decreasing time and budget requirements and improving SLAs in the top five outcomes of implementing server virtualization.

The Bigger Truth

Virtualization is vital for IT optimization and there is a compelling business case for adoption. It enables scalability, ease-of-management, improved utilization, increased availability, and system resiliency.  This, in turn, contributes to efficiencies in IT responsiveness and costs.  It brings qualitative changes to the way IT is run and services are delivered.

Virtualization is the cornerstone of the service delivery movement. As IT resources “go virtual,” IT organizations advance on their transformational journeys from an environment where resources are over-provisioned and underutilized, deployment is slow, costs are high, and processes are inefficient to one where resources are pooled and virtualized for right-sized consumption, deployment is on-demand, costs are optimized, and processes are agile.  Reaching the “dynamic” end-state in the virtualization journey takes a considerable investment in time, capital investment, and skill enhancement for IT resources; however, the results could be well worth the price.

[2] Source: ESG Research Report, 2010 IT Spending Priorities, February 2010.

[3] Source: ESG Research Report, 2010 Data Protection Trends, April 2010.

Background

ESG research indicates that increasing the use of server virtualization will top the list of IT priorities over the next 12 to 18 months.  As shown in Figure 1, securing, protecting, and managing ever-growing stores of information assets are top priorities as well. [1]

The research also indicates that the vast majority of organizations that have deployed server virtualization have done so in conjunction with networked storage (87%);[2] compared to islands of direct attached hard drives, utilization is greatly increased when applications share a pool of networked storage. Applications deployed on virtual machines sharing a networked pool of storage are more mobile and available than those deployed on direct attached hard drives.  Put it all together and it’s clear that forward-thinking IT managers are looking to combine the benefits of virtual server and virtual storage to increase the efficiency, flexibility, and security of their IT infrastructure.

IT managers are looking to extend the benefits of a virtual infrastructure beyond the physical walls of the data center.  The lofty strategic goal is to create a centrally managed pool of virtual resources spanning all data centers in an organization. A practical first step is achieving the ability to move a running application to another data center without having to declare a disaster.

Introducing EMC VPLEX

VPLEX is a combination of hardware and software deployed between servers and Fibre Channel-attached disk arrays.  It turns a heterogeneous mix of disk arrays into a distributed, federated pool of virtual storage that can be spread over multiple data centers. One of the key goals of the VPLEX architecture is to overcome the scalability and distance challenges that until recently have made deploying a large pool of centrally managed virtual resources that spans data centers prohibitively complicated and expensive.

VPLEX hardware is deployed as a rack-mounted solution (see Figure 2) with VPLEX engines at the core. The field-tested server hardware within enterprise-class EMC disk arrays is used for VPLEX engines.  Each engine is comprised of a pair of highly available directors, each with a pair of quad core Intel CPUs. Directors have 32 FC ports, each with 8 Gbps of bandwidth, for connectivity to hosts and disk arrays.  The solution is equipped with a management server, Fibre Channel switches, Ethernet switches, and power supplies.

A cache coherent, active-active cluster spread over one to four VPLEX engines is supported in the first release.   In other words, up to four VPLEX engines can be used to create single view of storage.  Two VPLEX clusters can span two data centers at synchronous distances.  All of the engines can be actively used and if any one of them fails, resources in other engines are used to ride through the failure without interruption. The first release of VPLEX has been qualified to work with a heterogeneous mix of disk arrays and IT infrastructure components.   This list will grow as EMC continues to invest in VPLEX interoperability testing.

VPLEX has been qualified to work with a wide variety of physical and virtual servers, operating systems, and applications. In the family of VPLEX products, VPLEX Local is intended for use within the confines of the data center while VPLEX Metro can be deployed within, across, and between data centers up to 100 kilometers apart.  VPLEX Geo, which is planned for a future release, is designed to stretch the benefits of VPLEX between data centers separated by more than 100 km.

How VPLEX is Used

Some of the more powerful ways that the first release of VPLEX can be used include:

• Online mobility of applications between servers, storage arrays, and data centers.
• Online movement of VMware virtual machines between data centers up to 100 km apart.
• Online data mobility between EMC and non-EMC disk arrays located within and between data centers.
• Centralized online mobility of storage capacity between tiers of storage.
• Standardized processes for the configuration and deployment of storage capacity.

Now that we’ve seen what VPLEX is and how it can be used, let’s take a look at how it fits into an existing virtual server infrastructure.  As shown in Figure 3, VPLEX is configured between SAN attached hosts and storage. In the traditional configuration shown on top in Figure 3, a virtual server is accessing storage through a SAN using a LUN presented by a disk array. As shown towards the bottom, VPLEX engines have been inserted in the SAN as a cluster, between the servers and the disk arrays. In this example, two clusters have been deployed: one in data center A and another in data center B.  The WAN-connected (up to 100 km) VPLEX clusters work together to present a cache coherent view of application data (VPLEX LUNs).  Behind the scenes, network RAID is used to mirror LUN data to the second data center (VPLEX R1) using caching to make it seamless.  This is what makes VMotion over distance a reality as it removes the need for a replication step.  Instead of having to do a VMotion and a Storage VMotion, VPLEX is VMotion only. With VPLEX, Distance VMotion is as easy as VMotion within the data center.

ESG Lab Validation

The balance of this report was designed to explore how VPLEX can be used in an existing VMware environment. VMotion between two data centers is demonstrated using a heterogeneous mix of disk arrays.  The performance impact of VPLEX deployed within and between data centers is examined.

ESG Lab testing was performed in an EMC solutions lab in Hopkinton, Massachusetts.   The test bed was designed to evaluate real-world application workloads running in a VPLEX enabled environment.  Oracle, SAP, Microsoft SQL Server, and Microsoft SharePoint were tested using industry-standard workloads.   ESG Lab’s analysis of the VPLEX was based on a combination of hands-on testing, an audit of EMC performance test results, and conversations with VPLEX Beta customers.

The ESG Lab Test Bed

The configuration used for ESG Lab testing is shown in Figure 4.[3] SAP, Oracle, SharePoint, and SQL Server applications were deployed within VMware ESX-enabled virtual machines running on four physical servers in data center A.

The servers in data center A were SAN connected to a pair of VPLEX engines, which was SAN connected to an EMC Symmetrix VMAX and an EMC CLARiiON disk array.  The virtual infrastructure in data center A was WAN connected to a simulated data center B.  WAN emulators (one for the LAN and another for the SAN) were used to simulate 100 km of distance between the two data centers.  A Cisco UCS and an EMC CLARiiON within VCE vBlock 1 were used for the servers and storage in data center B.

A VMware vCenter view of the test bed is shown in Figure 5.  Note that the two sites are not configured as separate data centers; instead, they are defined as two clustered pairs within one logical data center.  It should also be noted that Converged Network Adapters (CNA) from Emulex and Converged Enhanced Ethernet (CEE) switches from Cisco were used to consolidate Ethernet and Fibre Channel networks onto a 10 Gigabit Ethernet network.  This reduced the number of cables that needed to be connected to each server.

Getting Started

ESG Lab testing began with the configuration of VPLEX between existing virtual machines and disk arrays.  There are two ways that VPLEX can be configured with existing virtual machines: non-disruptive and disruptive.  The non-disruptive method uses VMware Storage VMotion to migrate existing storage to a newly configured pool of storage. This requires additional storage and takes time, but can be done online. The disruptive method does not require new storage capacity, but does require that the virtual machines are not running during the configuration process.  After VMware has detected the now VPLEX-managed LUN, the virtual machines can be powered on and resume normal operation.  During ESG Lab testing, the disruptive method was used to configure VPLEX for use between an existing virtual machine and an EMC CLARiiON disk array.[4] The process was later reversed to confirm that the VPLEX can be taken out of the configuration with no loss of data.

ESG Lab Testing

The home screen of the VPLEX Management console provides a good overview of the configuration process. As shown towards the bottom left in Figure 6, a storage volume within a disk array (shown in orange) is discovered and divided into extents.  Extents are used to build VPLEX devices (shown in green). VPLEX devices are optionally mirrored using extents from a second disk array.  The VPLEX device is used to create a virtual volume which is presented to hosts.  Hosts that have been registered as initiators (shown in blue) access VPLEX volumes through a storage view (shown in red).

ESG Lab configured an existing virtual machine deployed on one of the physical servers in data center A.   EMC refers to this process as encapsulation.  An overview of the configuration process is listed below. Note that  the VPLEX  LUN encapsulation steps listed in step 4 can be performed using wizards listed towards the right on the VPLEX management console shown  previously  in Figure 6.[5]

1. I.            The virtual machine was turned on and a text file was created and saved to the desktop.
2. II.            The virtual machine was shut down.
3. III.            The LUN that was servicing the virtual machines was removed from the vSphere inventory.
4. IV.            The LUN was encapsulated:
   1. The LUN was discovered.
   2. The LUN was claimed.
   3. Extents were created.
   4. A VPLEX device was created.
   5. The VPLEX virtual volume was created.
   6. The virtual server’s initiator was registered.
   7. A storage view was created and mapped to the initiator.
5. V.            The vCenter console was used to rescan and discover the VPLEX LUN.
6. VI.            The virtual machine was configured to use the VPLEX LUN.
7. VII.            The virtual machine was booted and the text file on the desktop was verified.

Thirty-three minutes after getting started on a pre-wired and -zoned test bed, an existing virtual machine was up and running in a VPLEX-enabled environment.  The process was repeated in reverse to confirm that an existing virtual machine can not only be configured, but can optionally be de-configured without losing data.

VMotion over Distance

VMware VMotion is used to move a running application from one server to another within a data center.  VPLEX Metro extends this capability as it enables the online migration of applications between data centers up to 100 km apart.  VMotion over distance has not been practical until now due to the challenges associated with creating a seamless, latency free connectivity to storage at distance.   VPLEX creates a decoupled layer between the application and storage as it maintains a virtual, mirrored, cache coherent copy of data in both data centers.   As a result, VMotion works the same as it does within a single data center. There is no need to fail over the storage to a mirrored copy at a second data center (i.e., Site Recovery Manager is not needed).  There is no need to pre-allocate storage and wait for the data to be mirrored by VMware (i.e., Storage VMotion is not needed).   The combination of VPLEX Local and VMware VMotion provides similar value within a local data center. A storage engineer at a global financial institution explained the value of this when he said:

“We present large LUNs to ESX and carve them into a bunch of smaller VMware file systems for guests.  The problem we have now is that if we want to move a guest to another server, then we have to disrupt all of those other guests that share the same LUN. Since VPLEX makes the LUN active/active in two locations, that means we can fail over without disrupting the other guests. That’s what got us interested in VPLEX.  I know that you could argue that we could have used smaller LUNs so each guest has its own storage, but that would increase admin overhead and complexity.”

ESG Lab Tested

ESG Lab used VPLEX Metro and VMware VMotion to move a virtual machine between two data centers located 100 km apart.  As shown in Figure 7, a SQL server virtual machine in data center A was moved online to data center B.   VPLEX maintained a mirrored copy of the data on an EMC CLARiiON disk array at site A and a vBlock 1 at site B.

The VMotion operation began with a drag and drop of the VM from cluster A to cluster B.   After dragging and dropping the VM from a server in cluster A to a server in cluster B, the “Change host” option in the wizard panel shown in Figure 8 was used to kick off the VMotion operation.

Note that the “Change datastore” option, which is used to start a VMware Storage VMotion, was not selected because VPLEX was maintaining a mirrored copy of the VM in data center B. In other words, the storage didn’t have to be moved to data center B because it was already there.

A write intensive IOmeter workload was running during and after the migration. The VMotion completed in one minute and twelve seconds. The workload continued throughout the migration and completed without error.   The virtual machine remained up and operational throughout the move.  The process was repeated in reverse as the VM was moved from cluster B to cluster A.

Performance Analysis

VPLEX uses advanced caching algorithms running on a scalable cluster of powerful hardware engines to overcome the performance challenges associated with deploying federated storage at scale and over distance.  When a federated storage abstraction layer is inserted between hosts and storage, each I/O operation will take a bit longer.  In other words, a federated storage abstraction layer adds latency.  Scaling this abstraction layer to run over multiple nodes in a cache coherent cluster adds a bit more latency.  And last but not least, stretching the abstraction layer between data centers adds even more latency.  The efficiency and advanced caching algorithms at the core of the VPLEX architecture were designed to mitigate the effect that these latencies can have on application performance.

ESG Lab Testing

A one terabyte online transaction processing system (OLTP) database application was tested using the Microsoft TPCE Benchmark kit. The application was designed to emulate multiple users within a brokerage firm as they managed customer accounts, executed customer trade orders and tracked customer activity.[6] Results were captured before and after VPLEX devices were configured with a goal of determining whether users would notice a difference in performance.  The performance of the five busiest LUNs is depicted in Figure 9 and Table 2.

What the Numbers Mean

• IOs per second (IOPS) is an indicator of the amount of work being done by the storage solution.
• As more IOPS are supported, more users can be supported. In other words, more IOPS is better.
• VPLEX Metro over 100 km reduced the number of IOPSs that the first Broker (B0) could perform by 3.4% (from 570 to 550 IOPS).
• Response times increased slightly, but not to an extent that users would notice.  All response times were within Microsoft guidelines.

The next test measured the time it took to VMotion a virtualized Oracle application as it was being load tested. The test was designed to ensure that the user experience remained the same while VPLEX-enabled VMotion was running.  Using the Oracle Load Testing for Web application, each of two test scripts was allowed to ramp up to ten virtual users each.  As expected, the average response time of transactions increased slightly (from 1.56 seconds to 1.62 seconds), but remained within acceptable limits. The time it took to complete the VMotion of the Oracle VMs from data center A to data center B is depicted in Figure 11.

What the Numbers Mean

• The graph depicts the elapsed time required to VMotion four virtual machines actively servicing 20 users accessing an industry-standard web server load testing tool from Oracle.
• As expected, the VMotion which moved the running virtual machines to a data center 100 km away (VPLEX R1 Metro) took a bit longer than the VMotion within a single data center (VPLEX R1 Local).
• In both cases, the VMotion completed in less than five minutes with no noticeable impact to users.

A long running Oracle operation was used to demonstrate how VPLEX caching can actually improve performance for read-intensive, cache-friendly applications.   The time that it took to gather schema statistics for the Oracle web server databases was measured before and after VPLEX was configured.

What the Numbers Mean

• An Oracle gather schematics operation is a long running, read-intensive batch job that is a good candidate for VPLEX caching.
• The Oracle batch job took five hours and forty-four minutes to complete before VPLEX was configured.
• VPLEX caching improved performance (6% with VPLEX Metro and 26% with VPLEX Local).

ESG Lab Validation Highlights

• The VPLEX management console and command line interface were used to configure an existing VMware-enabled virtual machine.  Thirty-three minutes after getting started, the virtual machine was running with VPLEX configured between the virtual server and an EMC CLARiiON disk array.
• The VPLEX configuration process was executed in reverse to ensure that customers can optionally de-configure VPLEX without losing data.
• A running virtual machine was moved between two data centers 100 km apart in less than two minutes. The simple drag and drop VMotion interface that VMware administrators are familiar with was exactly the same with VPLEX.
• ESG Lab audited the results of exhaustive EMC application-level solution testing with SAP, Oracle, and Microsoft SharePoint and SQL Server.
• ESG Lab confirmed that the performance overhead introduced by VPLEX is manageably low. With four VPLEX engines working over two data centers located 100 km apart, the cached virtual storage abstraction layer has a measureable, but nominal, performance impact (0.2% to 4.0% less IOPS for a Microsoft SQL Server OLTP application).
• ESG Lab quantified the difference in the amount of time that it takes to execute a VPLEX-enabled VMotion within and between data centers.   VMotion of four Oracle virtual machines took between 7.3% and 13.65% longer with VPLEX Metro. In both cases, VMotion completed in less than five minutes with no interruption to host applications.
• ESG Lab confirmed that for some application workloads, VPLEX caching can actually improve application performance.   An Oracle batch job that completed in 346 seconds before VPLEX took 256 seconds to complete with VPLEX Local (26% faster) and 324 seconds with VPLEX Metro (6% faster).

Issues to Consider

• WAN bandwidth. Planning is needed to make sure that you can have a properly sized WAN connection between VPLEX Metro-enabled data centers. Like most organizations that have deployed synchronous remote mirroring for disaster recovery between two data centers up to 100 km away, a direct fiber connection (a.k.a., dark fiber) is recommended.   If you have already installed Fibre Channel and Ethernet connections between buildings within a campus or metropolitan network, you should be all set.  If not, you’ll need to work with EMC and your WAN provider to make sure you can afford the WAN bandwidth needed to ensure that application performance is ensured during peak periods of activity.
• GUI vs. CLI. ESG Lab was impressed by the home screen of the VPLEX management GUI—especially the one-click access to wizards for each of the steps in the VPLEX configuration process.  That said, the GUI could benefit from a number of enhancements with a focus on helping administrators manage an existing VPLEX deployment (e.g., inventory reporting, audit logging, topology viewers, and wizards for common tasks). As the GUI evolves into a more fully functional graphical user interface, administrators can use the well documented VPLEX command line interface (CLI). 
• More to Come. There are a number of features and capabilities that will be needed before customers can realize EMC’s ambitious goal of creating widely distributed storage federation unfettered by the limitations of scale and distance. Besides EMC’s stated intention of supporting more scalability and distance in the future (e.g., VPLEX Geo), there are a number of incremental enhancements that ESG would like to see in future releases including Microsoft Active Directory integration, GUI enhancements, tighter integration with VMware vSphere, thin provisioning, and integration with EMC FAST.
• Your Mileage may vary. The performance results presented in this report are based on industry-standard benchmarks deployed in a controlled environment. Due to the many variables in each production data center environment, capacity planning and application level testing is recommended to assess the performance impact that VPLEX will have within your organization.

The Bigger Truth

VPLEX extends the benefits of virtual server technology to the storage infrastructure located within and between data centers.  Like virtual server technology, VPLEX increases the flexibility and agility of consolidated infrastructure as it creates a single point of control for a consolidated pool of federated resources.

For those with experience in the storage industry, this should sound familiar.  Storage virtualization solutions from EMC and others have been around for more than a decade.  While it starts with the same goals of providing a flexible abstraction layer between servers and storage, VPLEX uses a radically new approach to tackle two fundamental challenges with traditional storage virtualization solutions: scale and distance.  It uses advanced clustering and caching to extend the benefits of a virtual infrastructure beyond the walls of the data center.  A cache consistent pool of storage capacity located within—and between—data centers turns a chaotic mix of disk arrays into a distributed federation of centrally managed storage capacity.

What are the practical implications of this new technology? The easiest to understand is the ability to non-disruptively move virtual server and storage resources to another data center.  Instead of failing over to another site (e.g., with VMware Site Recovery Manager), ESG Lab moved a running virtual machine between two data centers in less than two minutes. The simple drag and drop VMotion interface familiar to VMware administrators was exactly the same with VPLEX.  Compared to a VMware Storage VMotion, which could take hours to run in the background, the drag and drop approach was completed in minutes as VPLEX maintained a single cache consistent image of federated storage at both sites. Oracle SAP and Microsoft SQL Server and SharePoint virtual machines were moved between data centers with no interruption.  An existing virtual machine was configured for use with VPLEX and the process was reversed.

ESG Lab confirmed that the performance overhead introduced by VPLEX is manageably low. VPLEX Local deployed within a single data center introduced measureable, but nominal, performance overhead (between 0.3% and 4.1% percent) during SQL Server testing.  During Oracle load testing, VPLEX Metro configured between two data centers 100 km apart was only 3% slower than VPLEX Local in a single data center.  As expected, VMotion between data centers 100 km apart was slightly slower than VMotion within a data center (7% to 14% longer)—but both were totally non-disruptive and completed in less than five minutes. Last, but not least, VPLEX caching actually improved performance for a long running Oracle batch job (7% faster).

VPLEX leverages the hardware engine at the core of EMC’s enterprise-class disk arrays and field tested clustering software obtained through acquisition.  Both have been deployed in extremely large mission critical environments.  It was no surprise that the Beta customers ESG Lab spoke with were impressed with the stability of the solution during their evaluation of VPLEX with physical and virtual servers.

VPLEX is an innovation with the potential to change the way organizations deploy and manage storage infrastructure.  Using clustering technology that’s been field tested at scale in production environments, VPLEX has the potential to create a single view of storage that spans the globe.  Existing EMC and VMware customers intrigued by the possibility of dynamically moving virtual servers between two locations within 100 km should view taking VPLEX for a test drive as a no-brainer.  Customers shopping for a storage infrastructure upgrade that can maximize the value of the storage that they already own should consider adding VPLEX to their shopping list. Try it for a VMotion between data centers or an online migration between disk arrays and you can look forward to turning an out-of-control heterogeneous storage infrastructure into a centrally managed storage utility.

Appendix

[2] Source: ESG Research Report, The Impact of Server Virtualization on Storage, December 2007.

[3] Configuration details are documented in the Appendix and an EMC white paper entitled Distance VMotion for Microsoft, Oracle, and SAP Enablde by VCE vBlock ,  EMC Symmetrix VMAX, EMC CLARiiON, and EMC VPLEX Metro, April 2010.

[4] ESG Lab confirmed that the non-disruptive mode is supported and routinely tested by EMC.

[5] ESG Lab used a combination of VPLEX GUI wizards and VPLEX command line operations to complete the configuration.

[6]For more details  on the performance benchmark methodologies and parameters, see the EMC white paper entitled Distance VMotion for Microsoft, Oracle, and SAP Enabled by VCE vBlock ,  EMC Symmetrix VMAX, EMC CLARiiON, and EMC VPLEX Metro,  April 2010

[7] Source: ESG Research Report, The Impact of Server Virtualization on Storage, December 2007.

What VPLEX is and Does

Let’s start with what VPLEX is not. First, it is not simply storage virtualization in terms of abstracting physical elements into logical views to provide non-disruptive mobility and management across arrays; to such functions, it adds the dimensions of scale and distance in a federated manner. Secondly, it’s not about replicating between systems. It is one integrated system, with one namespace. The traditional idea of a “master” and “slave” that dominates storage (where one storage entity owns a particular volume) needs to be forgotten.  Instead, “cache-coherency” works with a shared cache directory that amounts to a map of where all the data is; data is not replicated, map coordinates are transferred.

This is the essence and secret sauce of VPLEX since the role of VPLEX Metro in a SAN environment is both as a target and an initiator. From the host perspective, VPLEX Metro is a target and from the back-end storage array perspective, it is also an initiator. What this boils down to is seamless and predictable local or remote data access and mobility between two VPLEX Metro clusters within synchronous distances. Data center walls are no longer the barriers they have been. This new concept is what EMC calls “AccessAnywhere” storage (and what might more prosaically be called a stretched LUN for storage-focused readers) and it  allows data to be moved, accessed, and mirrored transparently between data centers, effectively allowing storage and applications to work as though those physical boundaries were not there.

Functional Capabilities: EMC refers to this as a “next generation private cloud infrastructure.” Whatever label it ends up with, these new capabilities offer both valuable and intriguing potential.  Here are some, in no particular order:

• Non-disruptive data mobility of virtualized storage arrays between EMC and non-EMC platforms, and sharing LUNs between VPLEX clusters—perhaps for migration or technology refreshes.
• Pooling and aggregating capacity, which can improve storage utilization and change the way “virtual-super-data-centers” are managed.
• Distributed mirroring across mixed platforms without requiring host resources, offering increased resiliency and availability including enabling applications to run continuously without restarts or interruptions.
• Transparent sharing and balancing of resources within and across physical data centers, where virtual machines can now be moved in toto, storage included, while applications are kept running. This can be done for workload balancing or to move work to low energy cost locations (as the geographic ability of VPLEX grows, this follow the moon approach for operations could be augmented by a follow the sun approach for knowledge workers, since physical separation of the data center and the work will no longer matter). Support for VMware to enable VMotion over distance between ESX clusters is also supported at general availability.
• Instant, real-time data access for remote users, where data is distributed and access is shared across sites, allowing concurrent read and write access to data by multiple hosts across two locations, providing real-time data access to remote physical data centers without local storage.

While the actual available products will take center stage, EMC’s VPLEX announcement also covered its general architectural approach and future strategy. It is about bringing federation to bear in order to relieve the geographic limitations that have prevented storage from playing a full, flexible role in the growing wave of resource-pool-based computing. As EMC itself states, “Federation enables IT to quickly and efficiently support the business through pools of resources that can be dynamically allocated. This flexibility elevates the value [that] IT offers within the business, as application and data movement is possible for better support of services. Together, cooperating pools of server applications and storage enable a new model of computing—dynamic IT as a service.”[1]

The Actual VPLEX Announcement: This is not the place for an exhaustive description of VPLEX. However, a quick summary of the highlights is important to put the later analysis in context:

Architecture: VPLEX combines scale out clustering, advanced data caching, and distributed cache coherence to allow data on distributed federated storage resources to be accessed and shared between locations.

Platform: An appliance based on a field-proven, redundant-everything EMC storage engine (for the technically focused, this has Intel multi-core processors using 8-Gbps Fibre Channel ports to provide 16 host and 16 back-end connections, and 64 GB of read cache memory).

Products: The two initially released products are VPLEX Local (provides local federation across heterogeneous arrays) and VPLEX Metro (provides distributed federation where two clusters can be connected within 100km synchronous distance). Both products support a range of hosts (VMware, Vblocks, and Cisco’s UCS) and storage arrays (EMC Symmetrix /CLARiiON, plus high-end Hitachi and IBM on day one with HP and NetApp to follow). The asynchronous VPLEX Geo is slated for 2011, and VPLEX Global is also planned.

Amidst all the market chatter and debate that will be generated around VPLEX, it is also worth looking at the market that the products enter.  Of course, the areas where users will be investing (shown in Figure 1) were investigated before VPLEX was commonly known; however, it is telling that the product can play a role in addressing the top planned storage investment areas cited by users from SANs and replication to virtualization and power-efficiency.

The User and Market Value of VPLEX

ESG Research Shows Practical Value: Until now, data access over distance has usually required replication (regular mirroring or copies of some kind) often accompanied by pain—whether operational, financial, or both. The elegance of VPLEX in addressing this issue should not be allowed to get in the way of appreciating the pragmatic appeal of its capabilities. After all, as the data in Figure 2 reminds us, the most important considerations for justifying IT investments remain very much about financial and business value. In this regard, VPLEX looks well placed for success as it can address many of the top considerations affecting IT investment, including the top two considerations cited by respondents: reducing operational expenditure and business process improvement. 

The potential sales news is even better for EMC when the data in Figure 2 is ranked by “spending mode”—those users that describe themselves as being in “growth mode” have slightly different considerations leading their investment choices. For these organizations in spending mode, their number one consideration is business process improvement. This is an area where VPLEX clearly delivers value, not just by doing “known tasks” better, but further by opening up the opportunity for hitherto impossible tasks. Furthermore, the same research also shows that “leading edge” users (those that stay on top of technical trends and purchase as soon as new tools are available—in other words, perfect VPLEX target purchasers) are far less focused on cost reduction than most (17% less than the average) and far more focused on improved business intelligence and delivery of real time business information (50% more than average). This is an area where VPLEX can deliver enormous value.

Even when the research is viewed from the perspectives of broader market segments, the results augur well for EMC’s new baby; when asked about their most important IT priorities, enterprise users—obviously a stalwart target segment for VPLEX—responded highly above average in favor of certain priorities that will help drive interest in VPLEX, most notably the increased use of server virtualization (the number one priority), regulatory compliance initiatives, the implementation of IT governance frameworks, and data center consolidation.

ESG Lab Testing Shows Operational Value: ESG Lab has also tested VPLEX and found it both excellent and very easy to use. Not surprisingly the headline story is VMotion over distance, where ESG’s Lab experience was very positive. As already covered, instead of failing over to another site using SRM or a stretched host cluster, users can move VMs at will between data centers (up to 100km apart with VPLEX Metro). What cannot be fully appreciated from EMC’s materials is how easy and transparent this is to do. From a VMware management perspective, the simple drag and drop process used to move VMs between physical servers in a data center is exactly the same with VPLEX. It’s fast (minutes) since the data is on both sides of the mirror. Another way to do this is with VMware’s Storage VMotion, but it can take hours to move the data and finish the migration compared to minutes with VPLEX.  During testing, ESG Lab moved a VM running a heavy IOmeter write workload 100Km in just one minute and twelve seconds with no interruption at the application level.

Unexpected surprises are good if they are also pleasant and EMC has delivered a couple with VPLEX. For instance, there’s what could be called the “easy test drive” ability. It’s something that isn’t in the EMC materials: the claim process used to virtualize an existing VM with VPLEX can be reversed. That’s a good thing—users could perhaps try it with their existing applications and VMs over a weekend and go back to their traditional tested infrastructure on Monday. This was a notable and missing capability in many early storage virtualization solutions. The Lab team was also impressed by the heavy investment in exhaustive real-world solutions-based testing with Oracle, SAP, SharePoint, and SQL Server, resulting in well documented solutions-based proof-points available at G.A.

The Bigger Truth

There are two “mega trends” driving the storage market: value and simplicity. These jointly deal with the issues of continually increasing data volumes and operational complexity.

If a vendor or product does not have a useful contribution in either one of these areas—and ideally, they should contribute to both—then that vendor or product will probably die. Products that are overly specialized, are techno-geek ‘crack,’ or that require an army of specialists to tweak all the knobs will be increasingly irrelevant as storage continues to become more of a flexible and automated servant of IT and less of a rigid and frustrating shackle restraining it. We must remember that storage is no more and no less than a tool—no one makes money by having storage in grand isolation. It is a means to an end, and not an end in itself.[2]

The bottom line for VPLEX, and hence the good news for EMC, is that measured against such criteria, it appears to have market applicability and relevance in two main ways:

1. The practical and immediate business impacts of delivering financial and operation value, which are a prerequisite for any market success. This includes management ease; better resource utilization; and power, cooling, and space advantages.
2. A more conceptual and longer term “play” (strongly tying back to value and simplicity) is that VPLEX (with its flexibility and heterogeneity) puts EMC squarely in the current and growing rush to collapsed-stacks and resource-pools. Ignoring this and continuing to simply produce standalone systems would soon have placed EMC in the role of King Canute. Instead, EMC is now part of the advance.

Like many good things, VPLEX isn’t conceptually complex. Indeed, its underlying sophistication and engineering wizardry belies its straightforward value proposition. Removing distance as a storage barrier brings us one step nearer to fully flexible IT, which is more about applications and delivering business value than it is about implementing and running a bunch of technologies. Server virtualization was one piece of the solution, VPLEX represents another. Yes, it still has to be field-proven, and yes, there are bound to be a few minor initial implementation limitations, and certainly the impact of truly massive distances on performance is some time from even being tested, but the supposed rule of distance as a virtual brick wall is broken as VPLEX Metro breaks through the real walls of data centers. Of course, other vendors are likely to cry humbug about such “unnecessary” functions until they, too, offer similar “vital” functions! After all, who needs radios in cars or computers in homes!? Since VPLEX is not a full storage virtualization solution (such as Hitachi’s USP-V, IBM’s SVC, NetApp’s V-Series, or even EMC’s own Invista), EMC would do well to work to make it at least aware of traditional virtualization tools (such as thin provisioning for instance); for now, many users will choose to run VPLEX in association with their existing traditional storage virtualization and we would not be surprised to see EMC integrate these functions over time. VPLEX federation is something new and different. Moreover, we absolutely know the business need exists because of the limited and specialist capabilities that have sprung up in this space before—but never with the scale, distance, ease, and affordability of VPLEX. Detractors might also say that the world already has remote mirroring using tools such as EMC’s own SRDF, but that is to overlook the massive rise of server virtualization and the new use case of VMotion over distance. The new capability will also open up new operational opportunities—and other ideas that have yet to be imagined.

Indeed, whatever else happens, we will all have to start thinking differently. After all, according to conventional wisdom in 1491, the world was flat. In 1492, it was not. In 2005, Pluto was a planet. In 2006, it was not. Similarly, in 2009, data access and management across distance was a major anchor. In 2010, that rule changed. Perhaps we need a new name for data centers—or at least a new conceptual understanding of what constitutes a wall?

[2] Source: ESG Brief, What’s on the Block for Block Storage in 2010?, January 2010.

Overview

Currently, enterprises adopt horizontal standardization, deploying common technologies at the storage and network layers. Storage offers block, file, and object access over a standard physical driver layer. Networks offer standard Ethernet- and IP-based physical drivers to the physical platforms.

The cloud changes this model by introducing a new virtual to physical layer into the IO layer; this adds complexity and obscures the underlying shared services, but delivers application mobility. In much the same way that (to date) enterprises have adopted a multi-vendor approach to operating system and hardware selection in order to provide support for their mix of legacy and best of breed application portfolio, today’s cloud-enabled enterprise is just as likely to need to support multiple hypervisors. This has significant implications for the selection and configuration of shared service layers.

The Changing Landscape

Legacy Windows platforms have been consolidated onto VMware farms for many years now as enterprises try to regain control over the server sprawl they built up over the last decade. VMware has been successfully deployed as an after-market fix that has enabled enterprises to execute on a physical to virtual consolidation without making code or environmental changes to their applications.

The next wave of virtualization is now starting, with new applications developed from the outset to support virtualization. This new world offers significant additional benefits to the consolidated/virtualized farms of today. IT will be able to migrate certain applications, live, between servers and within a data center without user impact. This offers the prospect of compute on demand with dynamic provisioning, improved availability (migrate on failure), and greater operational flexibility.

Key Technologies

The key virtualization technology here is the hypervisor and there are a number of competing platforms.

• VMware ESX
• Citrix Xen
• Red Hat KVM
• Microsoft Hyper-v
• Virtual Desktop Infrastructure (VDI) – [Xen, Hyper-v, or VMware]

Technology Diversity

From an end-user perspective, there are four conflicting drivers for the choice of hypervisor technology.

• Oracle VM (which uses Xen) is the recommended and fully supported virtualization platform for the 4,000 Oracle applications on the market and is likely to receive significant support from CIOs.
• Hyper-v is the recommended and fully supported virtualization platform for Windows platforms and services from Microsoft.
• VDI often leverages Citrix Xen, but can also use Hyper-v or VMware at the discretion of the integrator.
• VMware has a large install base of consolidated platforms.

Adoption of a single hypervisor is becoming an increasingly unlikely prospect for the enterprise, so developing a strategy and technology set that will fully support disparate hypervisors across the shared services layers of network and storage is essential.

What Are We Trying to Achieve?

Given that we must support multiple physical platforms and hypervisors simultaneously, we need to determine at the outset what is desirable and what is essential in our cloudscape.

Desirable. It might be desirable to transparently migrate applications between hypervisors (inter-hypervisor), but this is unlikely to be an immediate and essential requirement even if it were possible to achieve at present.

Essential. It is certainly essential to be able to migrate applications within a hypervisor environment (intra-hypervisor) as this enables load sharing and operational flexibility as well as an opportunity to improve availability and business continuity. Today, this is possible (with varying degrees of sophistication depending on the hypervisor), but it is dependent on fully functional virtual IO capability at the storage and network layers.

What Do We Need at the Network Layer?

Modern network ports are configured with more than MAC and IP addresses. It is common to configure QoS (quality of service), VLAN, and ACL parameters; these switch configuration parameters must be maintained as applications are migrated from one physical server to another server connected to a different switch port.

Uniquely, Blade Network Technologies provides support for all of the important hypervisor technologies, including VMware VMotion, Xen XenMotion, and Microsoft Hyper-v Live Migration. Its VMready is switch-resident software that provides the ability to transparently migrate a virtualized machine from one physical server to another whilst maintaining all of the essential quality, connectivity, and security settings.

In fact, VMready delivers against the desirable and essential network layer requirements for live virtual machine migration, both intra-hypervisor and inter-hypervisor. All that’s needed is to have the hypervisor capability in place to make live inter-hypervisor migrations a reality.

What Do We Need at the Storage Layer?

At the storage layer, SAN addresses (also known as WWN – World Wide Number) are virtualized across all major hypervisor platforms using N_Port ID Virtualization (NPIV). NPIV enables a single physical host bus adapter (HBA) to log in to a switch multiple times. The hypervisor then manages the connections between virtual HBAs in the guest operating system and these virtual NPIVs.

The benefit of this standards-based approach is that zoning, LUN masking, and QoS  are all based on the WWN so as long as the guest operating system maintains the same WWN migration between physical servers. The key is that all HBAs and all SAN switches must support NPIV; SAN controllers however, treat NPIV-created WWNs transparently. The hypervisor must also be NPIV-aware so that it can manage the virtual fabric ports. In the case of VMware, it is important to recognize that RDMs should be used in preference to the VMFS file system so that LUN masking, zoning, and QoS functionality can be related to specific VMs rather than to all VMs in a cluster. In addition to mitigating the security complications and risk that can arise from a configuration error if all VMs in the cluster share the same storage, VM-related storage frees storage/VM pairs from being restricted to a pool of clustered storage, all having to share the same values.

Analysis

It is likely that enterprises will choose to support multiple hypervisors within their private clouds and perhaps others in the public cloud, posing support and integration problems. How are we to deliver compatibility between multiple hypervisors and an existing legacy physical estate whilst maintaining our horizontal standardization in the storage and network layers?

Network layer virtual machine migrations are not just about moving IP and MAC addresses; they are much more significant and complex. The network switch needs to be included in the migration—unless you are interested in running in lowest common denominator mode with no QoS, VLAN, or ACL parameters.

Choosing the right open technologies and the best fit configuration options will enable enterprises to leverage shared services layers across multiple hypervisors.

The Bigger Truth

The second wave of virtualization is starting. It is essential that organizations select technology directions at the shared services layer that are open enough to support live VM migration. Blade Network Technologies, through VMready, has recognized that the point of network connectivity is now the application and not a MAC or IP address. By taking this approach, Blade Network Technologies has been able to deliver a flexible and universally applicable solution for highly virtualized multi-vendor platforms.

In comes Virsto. Virsto One is initially compatible with Microsoft Hyper-V environments to solve the exact dilemma VMware solved with VMFS. Mixed I/O workloads are becoming more and more common as administrators look to drive up relatively low consolidation ratios–5:1 on average, according to recent ESG research. As companies target the virtualization platform for business- and mission-critical workloads, performance, maximum utilization, and ease of management quickly become top priorities. Virsto extends the value of server virtualization by virtualizing the storage as well–the value of virtualization is extended into the underlying storage infrastructure and IT can plan to scale with confidence.

Virsto has some company: companies like Sanbolic and Microsoft Cluster Shared Volumes are looking to improve the reliability, agility, and manageability of virtualized environments. As storage management, capacity utilization, and efficiency remain top priorities, it will be interesting to watch each of these solutions mature in a market that offers ample opportunity for growth.

Read more of Mark’s blog entries at Liquefying IT.

Click here for the PowerPoint slide.

Click here for the full ESG Research Report: 2010 IT Spending Intentions Survey. ]]> http://www.enterprisestrategygroup.com/2010/01/key-areas-of-investment-for-data-storage-over-the-next-12-18-months/feed/ 0 http://www.enterprisestrategygroup.com/2010/01/percent-of-organizations-increasing-spending-by-technology-2010-vs-2009/ http://www.enterprisestrategygroup.com/2010/01/percent-of-organizations-increasing-spending-by-technology-2010-vs-2009/#comments Sun, 31 Jan 2010 20:14:21 +0000 kevin http://www.enterprisestrategygroup.com/?p=13817

Click here for the PowerPoint slide.

Click here for the full ESG Research Report: 2010 IT Spending Intentions Survey. ]]> http://www.enterprisestrategygroup.com/2010/01/percent-of-organizations-increasing-spending-by-technology-2010-vs-2009/feed/ 0 http://www.enterprisestrategygroup.com/2010/01/2009-to-2010-spending-change-in-specific-technology-areas/ http://www.enterprisestrategygroup.com/2010/01/2009-to-2010-spending-change-in-specific-technology-areas/#comments Sun, 31 Jan 2010 19:59:42 +0000 kevin http://www.enterprisestrategygroup.com/?p=13813

Click here for the PowerPoint slide.

Click here for the full ESG Research Report: 2010 IT Spending Intentions Survey. ]]> http://www.enterprisestrategygroup.com/2010/01/2009-to-2010-spending-change-in-specific-technology-areas/feed/ 0 http://www.enterprisestrategygroup.com/2010/01/areas-of-technology-where-organizations-in-major-cost-reduction-mode-are-most-likely-to-decrease-spending-in-2010/ http://www.enterprisestrategygroup.com/2010/01/areas-of-technology-where-organizations-in-major-cost-reduction-mode-are-most-likely-to-decrease-spending-in-2010/#comments Sun, 31 Jan 2010 19:40:16 +0000 kevin http://www.enterprisestrategygroup.com/?p=13910

Click here for the PowerPoint slide.

Click here for the ESG Research Report: 2010 IT Spending Intentions Survey. ]]> http://www.enterprisestrategygroup.com/2010/01/areas-of-technology-where-organizations-in-major-cost-reduction-mode-are-most-likely-to-decrease-spending-in-2010/feed/ 0 http://www.enterprisestrategygroup.com/2010/01/storage-virtualization-isn%e2%80%99t-about-storage/ http://www.enterprisestrategygroup.com/2010/01/storage-virtualization-isn%e2%80%99t-about-storage/#comments Wed, 27 Jan 2010 21:07:13 +0000 Garrett Doherty http://www.enterprisestrategygroup.com/?p=12382 Virtualization is at the center of all 21st Century IT systems, yet many CIOs fail to fully understand all of the benefits it can deliver to the data center operation. When we think of virtualization, we think compute, network, and storage—and we mostly think about driving up utilization on each. Storage controllers have always offered the ability to carve out pieces of real storage from a large pool and deliver them efficiently to a number of hosts, but it is storage virtualization itself that offers improvements that drive operational efficiency. IBM has been quietly addressing storage virtualization with SAN Volume Controller (SVC) for the last six years, building up a significant technical lead in this space.

Overview – Storage in the Data Center

Strange but true: most infrastructure architectures are deliberately designed from the outset to need little or no change over their lifetimes. There are two main reasons for this:

1. Change often means outages and customer impact and must be avoided
2. Budgets are set at the beginning of a project and getting more cash later is tough

Typically, then, applications are configured with all of the storage capacity they need to support the wildest dreams of their business sponsors (and then some extra is added for contingency by IT). Equally, storage is always configured with the performance level (storage tier) set to cope with the wildest transactional dreams of the business sponsor (and guess what? IT generally adds a bit more for good measure.).

No wonder storage is now one of the largest cost components involved in delivering and running a business application.

Endemic Over-provisioning

The storage salesperson loves this approach. Always over-provisioning gets her more volume at a higher price than the much more sensible approach of buying what is really needed, just in time, would. In fact, many vendors’ storage sales strategies are aimed at just that issue, making it easy for the customer to over-provision.

Actually, the work behind capacity and performance management is boring, tedious, and needs active participation from both the business unit and application development groups.  Many organizations cut those IT service management jobs a few business cycles back as no one would notice immediately. The infrastructure architect knows this and builds in so much additional performance and capacity that it covers for the lack of attention to the capacity and performance management process.

Here is a real life (and personal) example of what actually happens on the ground:

An application has just been rolled out that, at launch, actually needs 20 GB of storage and enough performance to support 3 transactions per minute, peak. It is provisioned with 700 GB of storage on Tier 1 (320 GB, 15k Fiber Channel disks)—enough performance and capacity to deal with the best possible business case in 10 years.

When the application got ported to a new hardware platform five years later, it was actually using 40 GB of storage at 60 transactions per minute, peak. To put some perspective around the commercial impact, over the five year hardware life, this level of over-provisioning consumed more than 6 megawatt hours of excess electricity and thousands of dollars in capital depreciation and maintenance costs.

This example is repeated again and again across IT systems, making the cost of failing to do the job right, billions of dollars of waste, and many gigawatt hours of wasted energy and unnecessary data center plant.

Even if some capacity and performance management had been performed that was able to identify in-life just how over-provisioned applications are, the difficulty of migrating data and applications to a lower tier of storage, plus the resultant service outage, would make the exercise impossible to justify.

Operational Paralysis

In many high risk, high impact environments, such as wholesale banking and manufacturing, (where application outage costs can be greater that the cost of the hardware), provisioning totally separate infrastructure for each business unit is an entirely accepted practice. The thinking is that by separating the infrastructure, change risk can be mitigated and obtaining change approval or permission from the business to schedule an outage is easier.

The problem here is not the change itself, but the impact of the change. Changes that reduce service availability are unpopular, are difficult to manage, and sometimes significantly delay implementation. Software and firmware patches often can’t be installed; other times, unreliable hardware can’t be replaced—all because doing so would cause a service impacting outage.

Well run IT infrastructure organizations set themselves up to avoid the need for changes that impact service availability. Unfortunately, changes to storage have been notoriously difficult to execute without impacting servers and applications. Typically, applications are very unforgiving of data unavailability, even if the outage is only for a short time.

The Continuous Migration Architecture

In addition to improved utilization, virtualization offers the prospect of isolating the application from the underlying hardware to such an extent that application migration is viable in real time as part of normal business operations. At ESG, we refer to this as the “continuous migration architecture (CMA).”  It delivers simultaneously on operational flexibility, business agility, and reduced costs.  In effect, a CMA is the embodiment of the private cloud, including compute, storage, and network virtualization.

What Value Does Virtualization Deliver?

Storage virtualization is all about operational efficiency, service levels, and flexibility. Consolidation is nice, but it is only a small piece of the overall value. Storage virtualization, on the other hand, is NOT about disconnecting the logical from the physical—it is actually about breaking the link between the physical storage and the application. It is about being able to perform normal maintenance activities with no service impact—and being able to make very significant changes to the underlying physical environment without anyone noticing. Compute, network, and storage virtualization technologies all deliver two key benefits:

• Operational flexibility – the ability to separate the application from the underlying physical hardware
• High utilization of assets – sharing the resources between multiple consumers

It is the first benefit that makes storage virtualization important for the CIO. By separating the physical storage from the application, significant changes can be made without a service outage. The quantity of allocated storage can be changed transparently, storage volumes can be moved from one tier of storage to another, and whole storage controllers can be swapped in and out—all without any impact whatsoever.

The second benefit is also a benefit of the standard storage controller: enabling the sharing of a large pool of disks between many consumers in a resource efficient way. Uniquely, IBM SVC adds the ability to concatenate stranded storage between controllers, reducing wastage and improving storage utilization.

IBM SAN Volume Controller (SVC)

Like many Tier 1 storage controllers, SVC offers enhanced features that further improve the utilization of storage resources (such as thin provisioning) while improving their resilience (via synchronous and asynchronous replication).

Additionally, SVC offers this functionality across all supported storage controllers. In many cases, functionality such as replication support between different vendors’ equipment or, in some cases, between different models of the same vendor’s storage controllers is not generally available elsewhere. SVC acts as an integration layer, offering a single administrative toolset to the multi-vendor storage administrator.

IBM has delivered 19,000 SVC nodes that are running in more than 6,000 SVC systems and, as a result, has built up a huge amount of experience with most configurations and environments. IBM’s approach reflects the reality of today’s data center: the need to support multiple different hardware and software platforms is a consolidated and simplified way.

The ability to work across multiple server virtualization hypervisors (VMware, Hyper-v, XEN, PowerVM, zLPAR) means that data center managers can fully support multiple virtualization environments (spanning Oracle Linux [XEN], Microsoft [Hyper-v], and VMware) all at the same time and from the same virtualized storage platform.

The ability to work across multiple storage controllers—including those from EMC, Hitachi, HP, Pillar, IBM, Sun, Fujitsu, and others—enables incredible operational flexibility. Universal connectivity enables the introduction and retirement of storage systems in a flexible and non-disruptive way. It also allows for flexible storage tiering so when the workload increases on an application and the relevant disk component is unable to sustain the performance required, migration to a higher performance platform or tier can be managed transparently.

A less important, but nevertheless relevant, issue is the ability to introduce competition into the storage buying process because the issue of compatibility with what you already have is gone. SVC customers have the advantage of a single storage driver for a single device type across a heterogeneous storage estate.

Remote and local site mirroring between storage controllers of different types (e.g., CLARiiON to DMX, IBM to HP) means that cost effective disaster recovery platforms can be constructed from the best fit platforms.

Solid State Support in the Virtualization Layer

Solid state disks (SSD) offer the potential of extreme IO performance with low latency and low seek times, providing service for the most demanding applications. Conventionally, SSDs are either located inside the host computer (which undoubtedly gives the lowest possible latency and IOPS potential) or in a storage controller. The host computer location option makes high availability clustering and cloud operating systems difficult to support. The storage controller option is ideal for high availability designs, but maximum IOPS are generally limited by the performance of the central controller and shared cache designs.

SAN Volume Controller offers an innovative, alternative solution with SSD support at the virtualization layer. A fully configured SVC system can support up to 32 SSD modules spread across eight SVC engines configured to provide 800,000 read IOPS within a 2.4 TB volume. By sharing the load between high performance computers (Quad Core Intel 2.4GHz with 24 GB of RAM) connected by four 8Gbps FC ports, SVC leverages an excellent scale-out design to deliver great performance.

Although these extreme levels of performance can only be achieved in a fully configured SVC cluster, much smaller high performance designs can be assembled, starting at two SSD modules spread across a pair of SVC engines in a cluster.

Why is This Important?

Organizations that fail to adopt virtualization and the continuous migration architecture will find the costs of running their IT organizations growing faster than IT budgets will allow. The opportunity to make changes and react to business demands will diminish as the cost of maintaining the legacy becomes all consuming and the cost of change becomes unaffordable.

Businesses facing an unresponsive IT organization have typically responded in predictable ways, such as:

• Do nothing and let more responsive businesses win
• Outsource IT
• Make a once every five year major IT investment to clean up the mess
• Get a new CIO/CTO

Enterprises should not “pay and pray” that things will get better. Those organizations that do so will find costs increasing rapidly whilst business agility disappears. In order to attain any CMA, virtualization must play a part. An investment made in virtualization will pay back in four key areas:

1. Reduction of operational cost – reduced headcount
2. Reduction of change risk
3. Reduction of over-provisioning and over-spending risks – reduced capital and operational costs
4. Improved change cycle time, improved business agility

The Bottom Line

By adding storage virtualization to a toolset that already contains server and network virtualization, IT organizations have the opportunity to ready themselves for a continuous migration architecture where all applications are isolated from their underlying physical systems. The continuous migration architecture is the embodiment of the promise offered by the private cloud: flexible, autonomic computing that simultaneously improves business flexibility while reducing the cost burden of IT.