At Chicago-based railcar operator TTX, Rob Zelinka, has taken the company's "Forward Thinking" motto to heart.
From the beginning of Zelinka's virtualization efforts, he wanted to be more aggressive than the average company. That meant an ambitious initial goal of virtualizing 75% to 80% of the company's servers.
Turns out, that goal wasn't ambitious enough.
TTX has already virtualized 75% of its server infrastructure, and is pushing to get as close to 100% as possible – and that includes desktops as well as servers, Zelinka says. Today, TTX has nearly 130 virtual machines (VM) running on production HP ProLiant DL servers and another 140 or so VMs for testing and development.
But this year in particular, virtual networking has been on Zelinka's mind – as it has been in the thoughts of many other enterprise IT executives, as virtualization vendors talk up their networking strategies.
For TTX, Zelinka has decided to do switching within a blade chassis using HP's Virtual Connect, in essence a software switch that will allow his team to move servers around without bugging the storage and network managers.
But the decision to go with HP, made prior to Cisco's announcement of its Unified Computing System (UCS), could have gone another way, Zelinka says. In UCS, Cisco packages blade servers along with storage, network and virtualization resources. "Today we'd have to look strongly at whether we want to do servers in what is traditionally a network chassis or network switches in what is in essence a server chassis."
Switch or broker?
And then there are so-called virtual switches integrated into hypervisors, such as the Cisco Nexus 1000V and VMware vSphere 4.0 suite, or the XenServer virtual switch Citrix Systems has in the works. Tom Nolle, CEO of CIMI, an IT consulting firm, cautions not to be confused by terminology. The Nexus 1000V is less switch and more resource broker, he says -- and enterprise IT managers ought to understand the difference as they explore where to go next with virtualization.
This switching/brokerage function comes into play when enterprises move beyond static virtualization -- loading an application onto a virtual server -- and begin creating a pool of virtual partitions on different servers for resource sharing among applications. Since the place an application winds up is now variable, IT has got to be able to bind a user to it in some way, he says.
"The guy who has the resources has to publish the resources and the guy who needs the resources has to pick something from a published available set and bind to it," Nolle says. "The network can do the latter of the two, connecting the user to a virtual application image somewhere. But the only guy who really knows how to get that virtual image published is the guy who owns the image, which in this case, is VMware [or other hypervisor provider]."
Watch for a variety of "fascinating" developments in the area of network virtualization to start popping this fall; vendors will be looking to grab mindshare as IT executives launch into 2010 budgeting, Nolle says.
"You're going to see a partnerships and management integration. Any virtualization package has a set of management tools that would allow you, for example to determine the status of a given server and the virtual partitions available on that server. That information is exposed through an API to provide integration for network vendors, who could use that process as the beginning of resource publishing."
Virtual starting point
In the meantime, Jeff Allison, network engineer with Health First, is eyeing the switching functionality found in vSphere 4.0 . That will come in handy for supporting virtual server growth at the Rockledge, Fla., health care system, he says.
Using VMware Virtual Infrastructure 3, the organization currently has 400 VMs in its production environment on 27 physical hosts. Its server infrastructure is nearly 60% virtualized -- and still growing. "We're waiting for eight more VMware servers to show up today," said Allison in a mid-July interview. "We’re mostly looking to the switching for ease of management."
In vSphere, VMware offers a collection of networking capabilities under the vNetwork label. One of the most interesting, experts say, is vNetwork Distributed Switch (vDS), which treats the network as an aggregate resource.
In other words, it abstracts individual, host-level virtual switches into a single vDS that spans multiple hosts at the data center level. Port groups span multiple hosts, which VMware says ensures configuration consistency for VMs and virtual ports necessary for such functions as live migration with VMotion. Previous VMware technology, vSwitch, handled networking and configuration on a per-host basis.
"As we grow, it will help to know that can just plant something into it," Allison says.
VDS also integrates with third-party virtual switches, starting with the Nexus 1000. Besides working through the vDS APIs to provide network services, the Nexus 1000V leverages Cisco's NX-OS operating system to provide deeper management and a feature set similar to its physical gear.
Vince Biddlecombe, CTO at Transplace, a logistics provider in Frisco, Texas, points to the optional Cisco virtual switch as one of several "pretty good" enhancements in vSphere 4.0. "This would give the network folks the ability to manage the network and have traceability right into a particular virtual machine," he says, "where right now they can really only see up to the host, the actual physical server."
Schultz is a freelance IT writer in Chicago. You can reach her at bschultz5824@gmail.com.