Issue link: https://cp.revolio.com/i/661564
18 | THE DATA CENTER JOURNAL www.datacenterjournal.com they Live (in An UndeAd Kind of WAy) A ccording to research unveiled in mid-2015 by the Anthesis Group and Stanford University researcher Jonathan Koomey, some 30% of physical servers are zombies—or what an Anthesis press release calls "comatose," defined as serv- ers "that have not delivered information or computing services in six months or more." Regardless of the metaphor, this surprising number points to a major problem for an industry already under rising scrutiny for its energy appetite. e percentage works out to about 10 million servers globally, or approximately $30 bil- lion just for the physical boxes, assuming about $3,000 per server. Naturally, different data centers will suffer from this problem to a different extent depending on their goals and in- frastructure model, whether it be a purely in-house facility or the cloud or some- thing in between. But given that the time frame for defining a server as comatose is at least six months, we can't dismiss these idle servers as just extra capacity waiting for periodic demand spikes. Every data center will have some capacity that's inactive for a portion of each day or each week, as demand will vary. Equipment that does nothing for months, however, screams wastefulness. On the other hand, many data center operators have recognized this problem and taken some steps to amelio- rate it—for instance, the virtualization/ consolidation wave that has largely swept the industry. Unfortunately, however, it hasn't completely solved the problem, if the Anthesis study is any indication. Jeremiah Ervin, senior solutions architect at Information Innovators Inc. (Triple-i), said, "e underutilization rate really depends on whether the environ- ment is largely made up of physical standalone servers or highly virtualized ones. Organizations with mostly physical standalone servers have higher underuti- lization rates, upwards of 50%, because server resources are typically deployed to meet peak performance thresholds." Such a number seems to be in the ballpark, since the overall 30% estimate for zombie servers would mean some data centers are (much) higher while others are lower. But Ervin believes the main factor determin- ing underutilization (the presence of zombie servers, at least in the extreme case) is not data center size, but configu- ration-management practices as well as a lack of virtualization and continuous infrastructure monitoring. Why do ServerS become ZombieS? Conceivably, poor capacity planning or an unwillingness to take a modular/ just-in-time approach to data center design could be to blame for a large outbreak of zombie servers, but in many cases, the problem builds over time. For example, "When application workloads are idle or applications are decommis- sioned without also decommissioning the server hardware, you end up with higher underutilization rates," noted Ervin. Particularly in situations where personnel face heavy workloads or tight schedules, the temptation might be to decommission the application on the soware side now and to address the hardware side later— only to find that the hardware decommis- sioning never really happens. Such factors may drive up under- utilization rates in any data center, but large facilities face particular challenges. "In larger environments—including some agencies in the federal government— where multiple contractors are deploying and managing IT resources in the same data center, you can end up with pockets of underutilized assets because a contract may end or be canceled for whatever reason, but the underlying hardware remains," said Ervin. "is situation is really a breakdown in communication combined with a lack of strategies [for combatting underutilization]." Moreover, as beneficial as it may be, server virtualization can hinder the task of reducing underutilization. Because it creates a resource pool beyond the limits of individual physical servers, it enables data center operators to avoid wasting capacity by allowing them to consume ar- bitrary amounts of resources rather than preset increments. In this case, however, some or all of the servers in a facility may fall outside the Anthesis definition yet still be vastly underutilized. Instead of leaving obvious zombies to be consoli- dated, virtualization can create a larger number of semi-zombies, which may be difficult to identify owing to the absence of a one-to-one link between the applica- tion and the physical hardware (server). KiLLing the Zombie (Server) for good e question is then how to address this problem. Virtualization provides some benefits, but simply spreading ap- plications across physical resources can just obscure the presence and location of underutilized servers. Also, differing business goals for a data center may make the problem more or less urgent as well as more or less difficult to address. But the problem can indeed be mitigated. Ervin cites multiple strategies for re- ducing underutilization and putting zom- bies down for good. Of course, the initial line of defense is the aforementioned server virtualization. It should be the first step for any data center looking to run applications on fewer hardware resources. Beyond this well-known (if insufficiently employed) tactic is configuration and change management, which address a problem cited earlier: idle capacity owing to decommissioned applications or con- tractor changes that fail to take account of newly unused hardware. In addition, solid data center monitoring helps keep the facility manager informed regarding how resources are being used (or not). Ervin notes that continuous monitoring offers