Data Center Journal

Volume 30 | February 2014

Issue link:

Contents of this Issue


Page 23 of 32

THE DATA CENTER JOURNAL | 21 a drag on operation. Part of the appeal of this approach, however, is isolation: by running separate OS instances for each virtual machine, the system can better ensure isolation among the different virtual machines. is capability is particularly crucial in the context of cloud providers, who must prevent crossover of data and resources between one customer's virtual machines and another's. One possible solution is "contain- ers," which are essentially virtual machines that run in a shared operating-system environment. Relative to a standard virtual machine, "the overhead of a container is [very] low. ey start so fast that many configurations can launch on-demand as requests come in, resulting in zero idle memory and CPU overhead." One of the current impediments to containers, however, is their lack of standardization, which contrasts with the standardization of typical virtual machines. But the appeal of unloading the OS from virtual machine instance will no doubt lead to further work in this area to garner more efficiency. BalanCing viRtual MaChines Virtual machines running on a server or several servers must still contend with the problem of limited resources. One such resource is network bandwidth. For instance, InformationWeek ("IBM Solves Virtual Machine 'Noisy Neighbor' Problem") cites the case of Netflix using Amazon: "In the first two years of use, [Netflix] concluded noisy neighbors or bandwidth-demanding virtual machines on a shared host stole a small percentage of compute cycles from its virtual machines as their CPUs waited on idle for access to the network." is problem limits the capability of virtual machines to act like their dedicated-hardware counterparts. Given that virtual machines are not limited to a specific hardware instance, however, creates a means to address the "noisy neighbor" problem. In the case of network bandwidth, IBM developed a patented approach to moving virtual machines on the fly and thereby preventing individual instances from being starved of their required bandwidth. Essentially, the method involves dispersing bandwidth-hog virtual machines so that they are not grouped in one location on the network, forcing them to contend for insufficient bandwidth. In particular, the dynamic nature of the method is helpful for applications with widely varying bandwidth needs; examples include "online retailers and auction sites that endure daily spikes of activity or seasonal spikes of traffic; search engines, which must respond to surges of activ- ity on different topics; government and news media, where sudden spikes of traffic might be driven by natural disasters, emer- gencies, elections, or outbreaks of conflict; and online sites for sporting events." speCializeD viRtual MaChines Beyond just the general concepts of virtualization are particular virtual machines designed to tackle jobs better than native operation. One example is Facebook's PHP virtual machine, which the company claims "can execute the language as much as nine times as quickly as running PHP natively on large systems," according to InfoWorld ("Facebook invents a PHP virtual machine"). e company had deployed the so-called HipHop Virtual Machine (HHVM) on all of its servers early in 2013. ConClusions Virtualization technology has enabled increased utilization of exist- ing resources in data centers, allowing companies to cut down on physical IT infrastructure (server consolidation) and improve energy efficiency by reducing idle time on servers. e centerpiece of this approach is the hypervisor, which generalizes the underlying hardware to create a uniform operating environment. But it's the virtual machines that do the work. Virtual machines function much like operating systems on dedicated hardware, but the virtualization layer enables multiple instances of one OS or several different OSs to run simultaneously on a given set of hardware. Virtual machines are critical to cloud service providers, allowing com- panies to sell chunks of IT resources in units that are much more granular than entire servers. Customers can tailor their compute needs according to their projects and pay accordingly. Another space where virtual machines are making a mark is the PC, allowing users to run an operat- ing system from within another operating system—such as a Linux window running in Microso Windows. Virtual machines may help OSs like Android that are at- tempting to gain traction in the PC market against their much more well-entrenched competitors. New technologies for virtual machines aim to improve efficiency (for instance, containers running on a common operating system) and prevent one in- stance from starving another of resources (IBM's dynamic method for preventing network bandwidth bottlenecks). Virtualization is no longer the hot trend it was for data centers in previous years—at least in the headlines. But it remains an important means of increasing efficiency, particularly in the face of rising energy costs, potential regulations and a tough economy. Future headlines will likely focus less on virtualization per se and more on new products and technologies that expand the capabilities of virtual machines and address their limitations. n

Articles in this issue

Links on this page

Archives of this issue

view archives of Data Center Journal - Volume 30 | February 2014