Despite enhancements on both cloud and virtual computing products, major vendors aren't taking into account many of the ways even a technology designed to save IT resources can unintentionally waste them.
Some companies are adopting products from relatively obscure third-party companies to keep resource-wasting virtual machines from eating their storage budgets, or add cloud-based computing resources without administrative overhead they can't afford.
Cloud-computing providers like Amazon, for example, provide a lot of power and flexibility on demand, but it takes too much time for a resource-constrained IT shop to set up or modify, according to Steven Peltzman, CIO of the Museum of Modern Art (MoMA) in New York.
Peltzman is plotting a long-term virtualisation/cloud-computing plan for MoMA, which currently has more than 100 servers, most at less than 50 percent utilization, to support a heavily digitized makeover of the museum and its offerings.
Along with mainstream SaaS, cloud and virtualization providers, Peltzman is is testing Cloudshare Pro, a free service that creates a generic-computing environment IT managers can hand over to business partners, or build to give internal users additional capacity.
Unlike Google's Gmail, which Peltzman likes as a low-cost, highly efficient way to outsource email, Cloudshare isn't limited to a single function. Once created, it provides a portal through which specific users get access to up to six Linux or Windows virtual machines, 10 user accounts and free access to Microsoft Office 2010 or 2007, SQL Server, Oracle and other applications.
"If you're a really good sysadmin with experience setting up huge environments, you can crank those out on Amazon or another cloud to get what you need," Peltzman says.
"That still takes time and expertise. With this [Cloudshare] you call up the interface, yank down a menu to load Windows 2000 or XP, yank down this app, that app, and sign up your user. It makes the process less cumbersome."
Even companies that have the time and staff to do heavy engineering work tuning their own systems end up wasting money and hardware in places they don't always examine - like the input/output (I/O) of their virtual servers, according to Roger Johnson, technical lead for the Enterprise Systems Group at high-end audio/video reseller Crutchfield Media.
Johnson, a Microsoft Hyper-V user who blogs about his success tuning Hyper-V servers to increase their stability and efficiency, says the way virtual machines send data to and from storage devices degrades server performance and wastes storage.
Physical servers sequence data through the I/O bus so that writes and reads to storage are in sequence with their location on the disk, not spread randomly so the hardware wastes time hunting for the right spot.
Virtual machines try to do the same thing, but since they share one machine, they end up dumping all their data on the hypervisor, which is designed for CPU load-balancing, not sophisticated management of data streaming and I/O optimisation, Johnson says.
"The hypervisor is having to compensate for what's almost a random stream of I/O, so it's a lot slower than it could be," says Johnson, who maintains about 150 physical servers and 200 VMs using Microsoft's Hyper-V, and speaks at Microsoft conferences about how server tuning can increase the number of VMs each Hyper-V server can support.
Johnson is testing an I/O and storage optimisation tool called Virsto One from Virsto Software that intercepts streams of data from virtual machines, sorts out the data and streams it through the bus to allow the writes and reads to each individual disk to be done sequentially, rather than in random order.
Using Virsto's product improved I/O for Microsoft Hyper-V-based virtual machines by an average of 30 percent, Johnson says. The product's other major feature - a virtual storage manager that can use a single server image to launch and maintain a host of VMs rather than using a separate image for each - increased storage efficiency by about 20 percent, he said.
"If you can reclaim 30 percent of your I/O, that's a couple of [storage] spindles you don't have to buy because the main way to address I/O problems is to add more spindles," Johnson says. "If you're spending a couple of thousand dollars per spindle, that adds up pretty fast."
You can get around either storage I/O or resource limits by throwing hardware at them, but that's not either an elegant or a cost-effective solution, according to Tom Becchetti, senior storage engineer for a global medical-device manufacturing company based in Minnesota.
Companies can alleviate virtual machine I/O bottlenecks by adding fibre-channel or multiple Gigabit Ethernet interfaces on the server, but ultimately all you're doing is building a bigger pipe through which to push a furball, Becchetti says.
"Virtualisation has really thrown the storage industry on its ears," according to Becchetti, who is also testing Virsto's product. "We're putting a strain on disk subsystems that really weren't designed to support multiple hosts."
Cloud computing has also changed the landscape for end users, but not always in a way that allows them to make the most use of its potential, Peltzman says.
"We have an unbelievable project list in the next three years and only about 45 IT staff. We're not going to be able to manage that well if we don't do this big strategic slide and look for strategic resources - outsourcing in India, cloud computing, virtualisation," Peltzman says. "[Virtualisation providers] underestimate the need SMBs have for cloud or virtual services that give us resources that don't cost that much in staff time."