LAB REVIEW: A one-two combo to KO storage costs

Posted on March 20, 2007

RssImageAltText

SAN volume virtualization is only the first step in cutting storage management costs; really knocking down costs takes virtual SAN connections and the ability to reconnect virtual volumes at will.

By Jack Fegreus

—For IT to function successfully, it must master change as a normal occurrence in an environment where change is constantly accelerating. In such an environment, IT must frequently deploy new resources, or re-purpose existing resources, such as turning an older database server into a Web server. Any such re-purposing of servers typically requires significant changes in storage, which is a highly labor-intensive task for storage administrators. To make that process significantly more cost-effective, Scalent Systems offers a new twist on virtualization software: the Scalent Virtual Operating Environment (V/OE).

Since the early days of data processing, IT has dealt with anticipated resource changes by over-provisioning—just in case. When the capital costs of such over-provisioning are combined with the operational costs associated with manually reconfiguring a SAN fabric, real potential arises to save significant sums of money by provisioning new resources "just in time" and configuring a SAN logically at will. What's more, there are equally significant ancillary benefits for data-center resiliency and business continuity when IT has the capability to readily tear down and rebuild diverse infrastructure environments.

The benefits derived from the virtualization of storage resources and operating systems are well understood. As a result, SAN-based storage virtualization and OS virtualization environments have become very sophisticated, as sites have moved to create single-resource pools. For storage, the goal is a single pool of virtual storage blocks that spans all vendors' arrays and provides a means to create easy-to-manage logical disk volumes that can raise the level of disk utilization to 80% or more. Meanwhile, for systems, the goal is a single compute space that spans all physical servers in which virtual machines run.

However, the independent development of storage and server virtualization technologies has created an IT environment characterized by distinct "silos" of technology. Even in data centers where virtualization has been aggressively implemented, storage administrators need to be involved in even the most basic server provisioning and maintenance tasks, including routine server moves, additions, and changes. Storage administrators need to schedule time to update the SAN fabric to account for any changes in either the unique worldwide name (WWN) of a Fibre Channel host bus adapter (HBA) or a Fibre Channel port (WWPN).

Worldwide names of server HBAs and ports are the links between the virtual world and the physical world of storage resources. That makes WWNs and WWPNs the bridges between virtual machines and virtual storage. That scheme, however, runs counter to the notion of simplifying IT resource management, which requires resources to be unconstrained by physical limitations such as WWNs and WWPNs.

Scalent's V/OE provides administrators with the means to provision virtual connections between systems and storage resources. In particular, V/OE creates resource pools of virtual WWNs and WWPNs for SAN HBAs and their ports, as well as resource pools of virtual media access control (MAC) addresses for LAN network interface cards (NICs). Via these logical connections, administrators can rapidly change what a server is running and how that server connects to storage and network resources. Also, an administrator does not have to physically reconfigure SAN or LAN connections or alter SAN or LAN topology. In essence, V/OE provides IT with the means to physically wire SAN and LAN architectures once, and reconfigure storage and network resources at will.

Pools of stateless server images—the OS, application software, Scalent's agent software, along with the server's virtual I/O topology—are maintained either in online repositories for easy deployment or on a server's local disk, which can be a SAN boot volume. Storing those images locally binds them to the server. Dubbed a "persona," each image can be made network bootable via a PXE Ethernet LAN connection. What's more, each persona encapsulates a fully provisioned server environment.

Using an independent Ethernet connection that supports lights-out system management, such as IPMI, iLO, or others, Scalent V/OE control software can reboot and re-purpose a target server running the Scalent agent software with a particular persona. As a result, V/OE eliminates the SAN and LAN administration overhead associated with any provisioning for server re-purposing by accounting for that provisioning when creating a persona.

Within Scalent V/OE, a network-resident system persona can be booted on any server at any time. From a practical viewpoint, however, there will be physical constraints on servers, such as internal memory and I/O bus speeds that will make certain applications fit better on different servers. To deal with those issues, most sites will create pools of servers with common attributes and assign personas to boot only on servers within certain pools.

Scalent V/OE significantly lowers the capital and operational costs associated with provisioning business continuity solutions. By making it easy to re-purpose a server, V/OE makes it possible to institute an N+1 sparing topology. Resource pools of physical or virtual servers can be created to supply a set of servers available to function as a fail-over platform for any application. For even greater efficiency, servers running non-critical applications can be automatically powered off via a lights-out management scheme and re-purposed to fill a disaster-recovery task or new business role in real-time.

OpenBench Labs first examined the operational functionality of a test Scalent V/OE deployment from the perspective of simplified administration and operations management. Included in this evaluation was the addition of a new system to the environment along with the creation of two new personas. These personas were used to test the ability to re-purpose the server to run a different application requiring different SAN connectivity.

How we tested

For our tests, we used a collection of HP ProLiant DL 380-G3 servers. These servers were powered by Intel Xeon EM64T processors and included an integrated intelligent management processor for HP's integrated Lights-Out (iLO) management package along with a dedicated Ethernet port for iLO communications. We configured one of these ProLiant servers as a Scalent V/OE Controller, which runs the software that manages the physical and virtual hardware, software, and network configurations within the Scalent managed environment. In our test configuration, we also used the Scalent V/OE Controller as a repository for Linux-based personas, which all of the servers accessed via NFS.


The physical view of the Scalent-managed environment shows the physical infrastructure based on the notion of a virtual rack (vRack). In our test environment, we had five active physical servers, including the Scalent Controller, which is at the top of the vRack. Each active server has icons on the left for the active persona showing the application and OS. Icons on the right are for SAN and LAN connections. Right clicking the SAN icon gets virtual connection details. We ran the Scalent VMRack Manager in its own virtual machine on our VMware ESX server. That allowed the ESX server to appear like a blade server chassis (see inset) and the virtual machines to be displayed like Individual blades.
Click here to enlarge image

The other critical piece of hardware required to create a Scalent-managed environment and deploy V/OE is a layer 2 managed Ethernet switch. In the openBench Labs test configuration, we employed two Cisco Catalyst 2950 Gigabit Ethernet switches. The Scalent Controller manages each of the switches. Through these switches, all of the management communications passes between the controller and Scalent agents running on target servers. This kind of hardware collection—Scalent controllers, managed Ethernet switches, and target servers—is referred to as a virtual Rack (vRack).


To complement the physical view, the Scalent Console provides a logical view of the environment that is connection-centric. Instead of seeing the vRack with five physical servers, an administrator sees seven virtual servers in a virtual network with LAN and SAN connections based on virtual switches and virtual WWPNs. That simplifies dealing with composite IT applications, such as Web services and complex SOA environments. Once the physical network is wired, logical, application-centric and virtual networks can be created at will.
Click here to enlarge image

Scalent V/OE supports Windows Server 2003, Red Hat Enterprise Linux, SuSE Linux Enterprise Server, and Sun Solaris on X86- or SPARC-based systems. Thin-client services are supported for Microsoft Terminal Services and Citrix Presentation Server (formerly Citrix MetaFrame). V/OE also provides support for operating systems virtualization through support for the VMware ESX 2.5.x or 3.0.x Server and the Xen 3.x hypervisor. In particular, for the VMware ESX Server, with Scalent's ESX management software running on the ESX server, a VMware ESX Server will appear in the Scalent console similar to a blade server chassis with full real-time details of each virtual machine running on the ESX Server, just as if it were an independent blade server.

Our assessment tests were designed to examine the ability of Scalent V/OE to simplify the administration of a Web services or SOA environment. To that end, we ran personas built on Red Hat Enterprise Linux and SuSE Linux Enterprise Server. Virtualization was tested running VMware ESX Server. For application environments, openBench Labs ran Apache Web servers, BEA WebLogic application servers, and an Oracle database.

There are many tools that provide virtualization services for specific resources, such as the virtualization of a vendor's storage array or a particular system virtualization scheme, along with management software for those resources. There are very few tools, however, designed to help system and storage administrators bridge the gaps between silos of virtualization technology by providing centralized management for all virtualized devices.

By virtue of the fact that Scalent V/OE software virtualizes connections between systems (both physical and virtual systems) with storage and network resources (again, both physical and virtual resources) system and storage administrators get a clear understanding of the all of the relationships among the physical and logical devices from the Scalent Console, which is a Web-based GUI to the Scalent Controller. In addition, deep support for system virtualization environments, such as VMware, provides centralized support for those environments. As a result, an administrator can take control of the full V/OE from a single-pane-of-glass interface.

Using the Scalent Console, a system administrator has a number of different views of the environment that cover all aspects of IT operations. Within all of these views, administrators can access a wealth of context-sensitive help and utilize drag-and-drop techniques to create, extend, and manage the virtual environment. Dragging and dropping personas, virtual switches, and virtual WWPNs changes the reality of the operating environment. By dragging out a new icon, an administrator starts a physical server and connects it into an operating network. More importantly, the essence of the Scalent V/OE value proposition is the ability to create and manage multiple virtual environments without disturbing the physical environment. This is particularly important for sites running compound applications over multiple servers, such as Web services.

To simplify both the creation and provisioning of virtual environments, the Scalent Console provides Catalog Views of all V/OE resources. In the various Catalog Views, administrators use the left-hand panel to search for elements in the Scalent-managed environment and use the right-hand panel to configure those elements. For example, to create a new persona from a gold master template in the persona catalog, an administrator first clones the template and then assigns appropriate virtual SAN and LAN connections to the new persona to add the proper SAN and LAN connectivity.


The Catalog View of personas provides a list of all of the possible server configurations that have been defined, such as the BEA WebLogic application servers used in our openBench Labs test scenario. In addition, the persona catalog includes templates as gold masters to define new personas. Typically, these templates will include configuration information for the OS and the application software, such as our template for an Apache Web server. The only configuration data needed to activate the template as a persona can be selected from catalogs of virtual SAN and LAN connections.
Click here to enlarge image

This is precisely what openBench Labs did to assess the ease of adding a new server and bootable persona into our Scalent V/OE framework. To the existing managed network, we added a new vRack into which we placed a ProLiant DL580 server running SuSE Enterprise Linux 9 and an Apache Web server.

The process of creating that persona for our server was very straightforward. After a standard local installation of the OS, the Scalent agent software and VLAN driver were installed to enhance the functionality of the server within the Scalent-managed environment. A reboot of the server made it a member of the managed network, which also made it visible in the Scalent Console. At this point, any additional application software should be loaded on the server and any vNICs and vWWPNs can be selected from a pool assigned for LAN and SAN connectivity that meet any fabric zoning constraints on resources.

However, to get the full value in terms of re-purposing persona on this server in the managed network, the server needs to be set to boot from the network, and the persona must be copied to the proper catalog on the NFS distribution server, which in our scenario was handled by the Scalent V/OE controller. Once we did that, the server could be loaded with any persona, and its persona could be loaded on any server appropriate for the persona.

For our final operational test, we installed an Oracle database for Linux on the ProLiant DL 580. Next, we created a second persona with the appropriate SAN connectivity that would be loaded from the Scalent V/OE catalog. We were able to select this new persona and reboot our test server from the Scalent Console. After the reboot, the server came up with its new application and the appropriate SAN connectivity for its new role. More importantly, all of this was done from the Scalent Console with no changes needed at either the SAN switch or the storage virtualization application for the storage array that the new persona accessed.

Working just from the Scalent Console, we re-purposed systems using an appropriate set of persona for our test V/OE environment at will. In each case, the system came up with all the appropriate SAN and LAN connectivity. As a result, we could now introduce N+1 redundancy for our test environment with one hot-spare server assigned to all of the servers within a particular resource pool. More importantly, the ease with which systems can be re-purposed makes it possible to provide system backup by shutting down non-critical systems, such as a system serving as a development test server.


After we cabled a ProLiant DL580 G3 server into the Scalent network and added Scalent's agent software and VLAN driver to SuSE Enterprise Linux 9 OS on that server, we were easily able to add that server to Scalent V/OE. In the Dashboard view of the Scalent Console, the new vRack that we created expressly for the ProLiant DL580 was visible along with the VMRack, which provides the virtual machine details for the ESX server running in our initial vRack of test systems. From Scalent Console, we could then reboot the server and change its persona (see inset).
Click here to enlarge image

Using Scalent V/OE, we had virtualized our SAN connections to create a wire-once fabric to simplify data-center system management. Through its Web-based GUI, Scalent Console provided us with a single-pane-of-glass interface for numerous system administration tasks. More importantly, the console provided a unified context for both the physical and logical topologies of our IT infrastructure, as well as a better understanding of the real-time relationships between the physical and virtual IT environments.

IT will increasingly apply virtualization to increase the utilization of new and existing resources and simplify systems/storage management. By extending virtualization to the way infrastructure resources—both physical and virtual—connect, Scalent V/OE ensures virtualization schemes work as expected, are easily managed, and provide all of their cost-savings potential.

Jack Fegreus is CTO of openBench Labs (www.openbench.com). He can be reached at jack.fegreus@openbench.com.

----

openBench Labs Scenario

UNDER EXAMINATION

Virtualization of SAN and LAN connections for a wire-once environment

WHAT WE TESTED

Scalent Virtual Operating Environment (V/OE) software

  • Architecture support for X86 (Intel and AMD) and SPARC
  • OS support for Linux, Windows, Solaris
  • Virtual machine support for VMware ESX 2.5.x, 3.0.x, and Xen
  • Fibre Channel HBA support for Emulex and QLogic
  • Network persona booting via NFS (Linux, Solaris) and iSCSI (Windows)
Two Cisco Catalyst 2950 managed switches

HOW WE TESTED

  • HP ProLiant DL580 G3 Server
  • SuSE Linux Enterprise Server 9

KEY FINDINGS

  • Scalent V/OE maintains pools of stateless server images
  • V/OE cleanly separates the management of servers from SANs and LANs through the creation of virtual HBAs and NICs.
  • V/OE is not in the data path so there is no performance impact or a single point of failure.
  • If Scalent components fail, there is no impact on the data-center systems.


Comment and Contribute
(Maximum characters: 1200). You have
characters left.