Deep within the bowels of your datacenter, unspeakable horrors -- the living dead! -- may dwell. Like zombies, these mindless minions feed greedily on, well, not brains, but electricity, a costly resource for a datacenter operator. They also take up precious space, something that many a datacenter operator can't afford to spare.
I'm talking about "zombie" servers, machines that, for whatever reason, have been left in the rack, powered on but doing very little or no work whatsoever. Now, if these were certain banking execs, they might receive huge annual bonuses for their non-efforts. But they're not, and it's in IT's best interests to weed out these near-dead machines as quickly as possible.
Some perspective as to how significant a problem zombie servers are: A study by McKinsey and Company titled "Revolutionizing Data Center Efficiency" found that among a total of 458 servers at four production datacenters, 32 percent (146 in all) were running at or below 3 percent utilization. Moreover, according to the same study, it can cost between $1,320 and $2,020 per year -- factoring in electricity, cooling, and other operational costs -- to support a single mid-tier ($2,500) server.
So then: Imagine you're running a 1,000-server datacenter and you discover that nearly one-third of your machines are superfluous. Well, you'd have a wealth of options. You could retire those machines and look at annual savings of, on the low end, $396,000 (that's 300 servers times $1,320). Or perhaps you could joyfully announce to the CXO that the expensive datacenter expansion project can now be put on hold because you've figured out a way to free up about a third of your datacenter space. (In this case, you might even get the bonus -- or it, too, might somehow end up in a banking exec's pocket.)
Unfortunately, though, hunting down and slaying zombie servers isn't particularly easy, especially in a large datacenter. Hardware vendors haven't designed machines yet to tap you on the shoulder and complain that they have nothing to do. But there are ways.
Among them, a company called Tideway offers a product called Foundation, an automated application mapping solution. It discovers all the hardware in your datacenter, from switches to servers, as well as what apps, OSes, and databases are running on them. On top of that, it maps all of the dependencies among them, including which servers are using which network connections and conversing with which databases.
This level of visibility has many benefits, but from a green-tech perspective, it means you can spot which servers are doing little to no useful work based on what apps they're running and how they're connected to the rest of your infrastructure. According to Tideway CTO Adam Kerrison, one customer (an investment bank) "identified 150 servers in its London datacenter that were effectively decommissioned. They found out they were still in there and plugged in. That's a breakdown in process that actually costs money."
Moreover, Foundation can provide a hardware profile of the machines in your datacenter, so you might discover a few dinosaurs that draw far more power than their shinier new brethren while delivering lower performance.
One feat Tideway's Foundation can't yet perform is telling you the utilization level of each server it finds, though according to Kerrison, the company is working on adding that feature.
Additionally, although Foundation can reveal which servers might be candidates for reassignment or dismissal, it can't take you by the hand and guide you to where the machine resides on the datacenter floor. If yours is a well-organized datacenter, that might not be a problem. Otherwise, you might need to follow the lead of Wachovia. The company combined Tideway's technology with that of a company called Interpoint to create a 3-D map of its datacenter operations.
Tideway's offering isn't the only product out there that can help pinpoint lazy servers. Nimsoft, for example, offers its NimBUS for Server Monitoring solution, which supports iSeries AS400, NetWare, Linux, Windows, and Unix from a single console. The solution monitors core server resources (CPU, memory, disk, event logs, counters, etc.) and enables centralized management of remote processes and services (i.e. automated and manual start/restart/stop).
Although my focus here has been on zombie servers, it's well worth pointing out one more statistic, lest the active servers out there get too cocky about how useful they think they are: The average utilization of the distributed systems in the datacenter today is between 5 and 30 percent, according to McKinsey.
The bottom line here is that underutilized servers represent a huge amount of costly waste in the datacenter. Rather than allowing that waste to continue unchecked, datacenter operators would be well served to determine which machines can be unplugged, which ones needs to be replaced, and where opportunity lies for consolidation or even virtualization. Taking these steps leads to lower energy bills and cooling costs, as well as a chance to free up space if expansion is in your company's future. That certainly beats being a caretaker for the datacenter of the living dead.