Scalent to Support Solaris with Virtual Operating Environment

Published: August 31, 2006

by Timothy Prickett Morgan

As everyone now knows, the IT industry is in the midst of a transition from physical to virtual computing. But what many people do not know is that this transition is just getting under way, and that logical or virtual hypervisors, which virtualize processors and therefore the operating systems and applications that run on top of them, is only part of what needs to be virtualized. Network and storage interfaces have to also be virtualized, and that is why Scalent Systems has created what it calls the Virtual Operating Environment.

It is hard to talk about virtualization without getting technical, but to understand what Scalent's V/OE does--why Scalent added a slash to the acronym for Virtual Operating Environment, heaven only knows--you have to understand what current hypervisors do not do.

With a virtual machine hypervisor like the popular VMware GSX Server or ESX Server, the instruction set of a chip and the way it talks to memory (part of the functions of a chipset) are emulated in such a way that an operating system talks to a virtualized processor instead of a real one. That allows a single physical processor to be carved up into many virtual processors. This instruction set virtualization is all done in software, and that software exacts a relatively heavy performance penalty on the processor. No vendor has been particularly honest about this overhead, but the fact that Intel created the VT electronics for its latest Itanium and Core processors and Advanced Micro Devices created AMD-V tells you it was pretty high. VT and AMD-V implement instruction set virtualization in the chip's hardware, so that part of the hypervisor does not have to be done in software any more.

But that doesn't mean that I/O--connections to storage and to networks--has been virtualized. In fact, Intel and AMD are both working with hypervisor makers--mainly VMware and XenSource, which has created the open source Xen hypervisor--on creating I/O extensions to their respective VT and AMD-V electronics that will allow them to virtualize connections to storage and networks. Such virtualization will radically simplify the process of creating and migrating virtual machines within a network of computers. Right now, as Scalent correctly points out, network and storage connections are still shackled to the physical world, and the V/OE software addresses that problem.

But virtualized I/O is not the only problem that data centers employing virtual or logical machines need to solve. Companies need some sort of master software, almost acting like microcode in the mainframes of old, that can link server provisioning tools such as those from OpsWare, Hewlett-Packard, IBM, BladeLogic, and others to the so-called orchestration software--think of IBM's Tivoli Orchestrator or Cassatt's Collage. Such software, when used together, can provision a server and roll it in and out of production. Then, you have to have something that can control both physical and virtual environments, dealing with VMware, Xen, and Microsoft Virtual Server virtual machines as easily and fluidly as it does physical boxes. Scalent believes that V/OE is the glue that can hold all of these pieces--provisioning, orchestration, and virtualization--together and make it a coherent whole that truly automates the way servers are brought into being and put out of commission as workloads change in the course of a day, month, or year.

While everyone is keen on virtualization and the server consolidation and higher utilization it enables, the single biggest complaint that IT shops have with these technologies is that they don't work together. IT shops have to create their own policies and practices for making OpsWare play with Tivoli Orchestrator and various hypervisors. This problem, says Kevin Epstein, vice president of marketing at Scalent, is what V/OE is aimed at solving.

"The first wave of virtualization was chopping up boxes," he explains. "The next wave will be about coping with a mix of physical and virtual boxes, and coping with what needs to happen when you want to change things. What happens to network and storage connections when physical and virtual machines change?"

This may sound like an easy problem to crack, but it isn't. Server operating systems have been programmed to act like they own their network and storage interfaces, every bit as much as they were initially programmed to think they exclusively owned their processors and memory.

In an effort to not get hung up on the V word, Scalent also calls its software the means of doing rapid server infrastructure repurposing, putting an emphasis on the fact that V/OE, in conjunction with other tools, allows data centers to stop building servers that have capacity that is, on average, 120 percent higher than the peak load for the main applications that run on them. Epstein says that in many companies, the difference between average and peak load can be a factor of 20 to 1. Moreover, for every four servers in production, there is usually one server back in development.

All of this means there is a lot of spare physical server capacity that could be better utilized, even if companies do not go virtual just yet. Just being able to move applications around a network of flexible servers--mixing application development, production, and high availability clustering capacity on a single set of machines--can save companies a lot of iron, and therefore, a lot of money. For instance, if programmers are not doing QA runs on code, why shouldn't their servers be available for production workloads? Or, if there isn't a disaster, why shouldn't HA servers be used to support either production or programming workloads? In the event of a disaster, such as a bad software crash, backup production workloads could be quickly provisioned and brought online, kicking programmers offline for a bit.

Initial deployments of Scalent's V/OE software have chopped hardware costs by 54 percent, cut network infrastructure costs by 62 percent, reduced software bills by 66 percent; this results in an average one-time cost savings of 60 percent, plus a 41 percent reduction in recurring costs. The net effect is to reduce the total cost of ownership of a server farm by 56 percent. These are big numbers--and that doesn't even take into account the effects of virtualization.

The V/OE software runs on a server of its own, and its puts agents on the servers that hook into the operating systems, provisioning, orchestration, and hypervisor software running on that iron. The key thing is that V/OE remains out of the dataflow. If you have machines set up using V/OE and you walk over and shoot the server running V/OE, it does not affect any of the production workloads. Shooting that server will turn the configuration of your physical and virtual servers from dynamic to static, though. So don't shoot it.

V/OE 1.0 was launched last November, and it supported Linux and Windows platforms. With the V/OE 2.0, due before the end of the year, Scalent will be able to manage Solaris 10 servers, either the Sparc or X64 variety. The software will also include support for high-end Ethernet switches from Cisco Systems, Nortel, and others. V/OE 2.0 also has a set of Java-based application programming interfaces that will allow providers of operating systems and other tools to hook into V/OE.

The V/OE software can manage Xen and ESX Server hypervisors today, and Virtual Server hypervisors will be added at a later date to a future V/OE release. Epstein was not at liberty to say if support for IBM's AIX platform and its Virtualization Engine hypervisor or HP's HP-UX platforms and its Virtual Server Environment were coming in a future release, but that stands to reason.

Scalent does not provide list pricing for its software, but Epstein says that in a typical configuration, the cost of the V/OE controller software and agents for servers averages out to about $2,000 per physical server. Once the V/OE agent is on a box, it can be used to interface with that one physical machine or an unlimited number of virtual machines on that box.

Scalent also announced this week that it has inked a partnership with IBM, which will be a reseller of V/OE on BladeCenter blade servers and System x rack-mounted servers. Scalent was already a Sun partner.