Server Consolidation & Virtualization

Wednesday, December 06, 2006

Virtualization and server consolidation are the hottest buzzwords in the industry today. The interesting thing is that virtualization used to be one of the techniques for doing server consolidation. Today, virtualization can be used for more than just server consolidation. It can be used for disaster recovery and business continuity, testing and training, securing desktop environments, and much more. Similarly, virtualization isn't the only technique for doing server consolidation. Here, we'll look at the technologies in each area in more detail.

Let's look at server consolidation first. The most common definition of server consolidation is that it reduces the number of servers that you have in your data center and other offices by moving the Operating Systems and applications that were running on them onto a fewer, more powerful servers. This is done with the objective of reducing administrative and management costs of your existing fleet of servers. But that's not the only definition for it. According to Gartner, server consolidation can be divided into three types: logical consolidation, physical consolidation and rationalized consolidation. There are vendors who have also categorized server consolidation differently. IBM for instance, divides server consolidation into centralization, physical consolidation, data integration, and application integration. In all cases, the objective is to simplify your server infrastructure. Here, we'll go by Gartner's definition.

Logical consolidation
This is about what most IT managers dread the most: documentation. How well documented are the servers in your organization? Do you know which server is located where, its configuration, how well is it utilized, and what applications is it running? If you do, then you've already achieved one part of logical consolidation. The other part is to understand how to use this information effectively to plan your next server upgrade or purchase better, or improve troubleshooting.

Physical consolidation
This is nothing but reducing the number of physical server sites that you have by moving servers from those locations into a fewer locations. By doing this, you can save not only the cost of maintaining so many servers, but you also save on real estate costs.

True Utility Computing with Jack PC

The PC has been shrinking into various shapes and sizes, but one could never have imagined that it would actually fit into a wall jack. That's exactly what Israel based Chip PC Technologies has done. Termed as the Jack PC, it's a thin client that mounts into a LAN jack on the wall. The Jack PC has USB ports to connect your mouse, keyboard, or other devices; a VGA port for display, and jacks for a mic and a headphone. What's more, Jack PC supports PoE (Power over Ethernet), and therefore only consumes 5 Watts of power, so you can imagine the power savings. It doesn't expose any removable components, so it can't be pulled out of the wall very easily, except with special tools. This makes it more secure. The Jack PC supports all major thin client technologies, including Citrix's ICA, Microsoft's RDP, and has IE 6 web browser, and some other plug-ins for local applications. You could for instance, use it with VMware's Virtual Desktop Infrastructure, which is another new technology that let's you virtualize your desktop fleet (see “glimpses of VMWorld 2006” towards the end of this issue). The company also has other thin client solutions. Though these don't fit into a wall, they're compact nevertheless. Email: sales@chippc.com, website: www.chippc.com

Rational consolidation
This is the most commonly known part of server consolidation, and the most complex. It dramatically improves server utilization, but you must also have failover support, as doing this is like putting all your eggs in one basket. In this technique, you either put multiple applications onto a single Operating System on one server, or you install multiple Operating Systems on a single Operating System on the server, and then install applications on top of each OS instance. Gartner calls the former technique as Workload consolidation and the latter as partitioning. Workload consolidation needs a robust scheduler that can balance the hardware resources amongst different applications. There are two different ways of doing workload scheduling-processor binding and software based resource allocation. In the former, different applications are tied to different processors in a multi-processor machine. IBM's AIX Operating System for instance can force processor affinity through a 'bindprocessor' command. IBM defines processor affinity as the probability of dispatching a thread to a processor that was previously executing it. HP also allows this on its HP-UX processor using its HP-UXi 11i OS. In Software-based resource allocation, the OS itself tries to allocate the hardware resources to the applications.

Server Consolidation using Partitioning
Partitioning is another way of doing server consolidation. You can do it at the hardware level, which basically means allotting processor(s) and memory to specific applications. This technique is a fairly old one and has been around in mainframes and the UNIX world for a long time. It's only done on multi-CPU servers, and physical hardware is used to create partitions on the server. Each partition can thus have one or more CPUs, dedicated memory and I/O components. These are electronically isolated and therefore are completely safe from each other. Even if one partition goes down, the other partitions remain unaffected.

Another form of hardware partitioning can be achieved with blade servers. They offer electronic isolation, and each server blade contains its own CPUs, memory, and hard drive. The good news about blade servers is that their costs have come down significantly.

The second type of partitioning is logical partitioning. This adds a firmware to the hardware for doing the partitioning. IBM for instance, adds a firmware hypervisor function in its pSeries servers that does virtual memory management, debugs registry and memory access, and provides virtual tty support. The benefit of logical partitioning is that you can alter resources to virtual partitions without stopping the OS. The last form of portioning technology is called software partitioning. This allows one Operating System to run several guest Operating Systems on top of it. In other words, there's a hypervisor on top of the host OS on the server, which allows multiple instances of an OS to run on top of it. In case you haven't figured it out, then this technique is also known as virtualization. Today, virtualization is one of the most happening technologies and has moved far beyond being a mere technique for doing server consolidation. Let's get dig deeper into it.

Understanding Virtualization
Many of the techniques we've discussed in server consolidation are a form of virtualization, and they've existed in the UNIX world for ages. What's relatively new is the introduction of virtualization technology on the x86 platform.

This is good because the real challenge of server proliferation has happened in this segment only. Most organizations have tons of x86 based servers, the result of which is high administrative overheads and operating costs, challenges in managing the server fleet, etc. In fact, there are many studies which indicate that most servers in the enterprises are not utilized beyond 15-20% of their actual capacity. This wastes a lot of compute power. It wastes a lot of precious energy that powers up servers and keeps them cool. Virtualization can improve server utilization by deploying more than one Operating System and applications on single server. It achieves not only server consolidation, but also improves resource utilization, reduces real estate requirement in a data center, and lowers energy requirements. Moreover, virtualization technology on the x86 platform is also available for desktops.

Types of virtualization
The oldest type of virtualization is emulation, which is running an OS that's meant to be run on one platform on another platform. The one that's gained popularity today is native virtualization, which allows you to run multiple Operating Systems on same platform. The difference here is that it doesn't let you run Operating Systems for some other CPU. Another form of virtualization is called paravirtualization, which also let's you run multiple Operating Systems on the same system, with the difference that it's tuned to provide better interaction with the CPU, memory, and other I/O devices. It also interacts directly with Intel's VT and AMD V technologies to give better performance.

Hardware Assisted Virtualization
The dark side of software virtualization on the x86 platform is that performance takes a hit as the number of virtual machines increases. To tackle this, Intel and AMD are coming out with virtualization technology in their processors. Intel calls it Intel VT, while AMD calls it AMD V (code name
Pacifica).

Intel VT
This was officially launched in 2005 and was available in most Pentium 4 6x2, Pentium D 9x0, Xeon 3xxx/5xxx/7xxx, Core Duo and Core 2 Duo processors. The technology facilitates a CPU to act as if it were numerous CPUs working in parallel. It has some extra instruction sets that are used in virtual machine emulation. These instruction sets reduce the communication process between virtual machines and actual hardware, which was earlier performed by VMM. This eliminates overhead and improves performance.

AMD-V
AMD introduced its virtualization extension in its 64-bit architecture in mid of August 2006. This technology uses Direct Connect Architecture, which provides a balanced mechanism to support virtualization software in managing, partitioning, and securing I/O devices. This is done for improved performance and less implementation complexity in providing I/O in virtual environments. It helps eliminate bottlenecks inherent in FSB architectures and gives high-throughput responsiveness and scalability for your applications to improve overall system efficiency. Right now this technology is available in AMD Opteron Processors.

Virtual Appliances
This is the latest feather in virtualization technology's cap. Simply speaking, it is a combination of OS and application. Instead of distributing apps alone, ISVs can bundle an optimized OS with them. The virtual appliance would then simply be dropped on top of a virtual infrastructure from some vendor, and would be ready to run.

This would save IT managers the trouble of first installing an OS on a hardware, and then installing and customizing the application. They can instead focus straightaway on configuring and customizing the application as per their organization's need. It's a fairly powerful concept, and is worth watching out for.

Page(s)   1