The pressures on data center managers continue to mount. Applications need to scale for millions of potential Internet users. Data center consolidation efforts are accelerating. Multi-core servers are cranking out more data. And the information flowing across the wire is now often isochronous (voice and video) versus asynchronous (data).

In the words of Jeff Thermond, CEO of Woven Systems Inc. of Santa Clara, Calif., "It's a Budweiser-size problem. It's not a microbrew issue."

If your data center problems seem insurmountable, don't reach for the nearest six-pack. Consider, instead, upgrading your data center network to 10 Gbit/sec systems. And from Thermond's perspective, that would be Woven's EFX 1000 Ethernet Fabric Switch.

The device uses a proprietary ASIC processor to switch traffic (using Ethernet 802.3 standards, of course). The chip detects congestion in advance, cuts jitter and reduces latency enormously, according to Thermond.

You may think your 1 Gbit/sec Ethernet connections from your racks of servers to your network fabric will work just fine in the years ahead. Not so, says Thermond. He predicts that in 2010 more than half of the servers shipped will sport 10 Gbit/sec links. Then what will you do? Maybe give him a call. Prices are as low as $1,700 per port.

This week UPS vendor American Power Conversion rolled out new efficiency benchmarks for all of its products. Starting on Monday, all products now carry an efficiency rating in the tech specs that looks like this:

Buyers see a graph of predicted efficiency (y axis) based on load (x axis). Since many UPSes operate at very low load levels, efficiency levels on the low end of the load range are more important than what you see to the right.

By doing this, APC hopes to set the bar for competing products and start a trend toward providing efficiency ratings for IT equipment in general, says executive vice president and chief technology officer Neil Rasmussen. And he thinks APC has an advantage. "You'll see a factor of four reduction in power losses in power distribution units, UPSes and IT power supplies [used in servers] by 2010," he predicts.

"We’ve been working on this for a year and we are first in the industry to do this," Rasmussen said at a meeting last week. In fact, APC told me over a year ago it planned to release the specifications. What took so long, says Rasmussen, is that even APC didn't know what the efficiency numbers were for its products, nor did its component suppliers. That's a pretty sad state of affairs, given the intense focus on green computing and efficiency these days.

The effort to provide efficency specs is all fine and good as far as it goes, but I asked Rasmussen why we should trust their measurements, given that there's no standard. "It’s just like any other specification we publish. We’ll stand behind the numbers we publish," he said. Rasmussen didn't go into detail as to the methodology used.

The problem here is that efficiency numbers can be manipulated unless the specific test environment and measurement process is carefully detailed. For any kind of meaningful comparison to take place, a standard and some sort of certification process is needed to assure the public that such numbers are accurate.

Rasmussen thinks the numbers APC has come up with are good enough - at least for now. "We don’t intend to go out and certify the products in any third party way," he says, adding that it’s possible that standards will eventually emerge that every vendor be expected to test and certify to those standards.

For now, however, comparisons of such efficiency ratings, if and when other vendors issue their own variants, probably need to be taken with a grain of salt.

The momentum behind blade servers continues to build. IDC says blade server shipments in 2007 grew by more than 35%, while the overall server market increased by only 2.4%. And the $1 billion venture capitalists have invested into blade companies of late is resulting in a plethora of products hitting the market now and in the months to come.

So, it's no surprise that today in New York the Blade.org Technology Symposium 2008 will make its debut. Blade.org is an industry consortium backed by Intel, IBM and other heavyweights and includes user companies like MetLife and CBS Television as well.

For blade advocates like Doug Balog, chairman of Blade.org and the vice president of blade technology at IBM, the compact computers are more than mere servers and deserve their own conference, something the world never saw with pizza box or tower servers.

To him and others, blades are the loci of systems convergence and integration for the next generation data center. First, you can pack more blades into a standard server rack, more than doubling your computer performance for the same data center real estate. (Power issues need to be thought through, of course.) But it's the shared resources that make blades unique. The converged network fabric they share, in particular, is where the action is in the near term.

Balog says in early 2009 he expects to see blade systems with adapters running both fibre channel and Ethernet protocols, giving more flexibility to IT for configuring their SAN and server resources in a blade environment than with standard rack-based servers.

He suggests that when that happens not only will CIOs be able to make better deals on network gear, ultimately network management tools will converge and the skill sets people need to use them will converge, as it were, as well, making more efficient, less costly systems management possible.

We'll see.

Last week APC's Neil Rasmussen dismissed the idea of DC power as outdated. Today an EDS spokesperson said it's a worthy alternative in data centers - if plug standards can be developed. If UL-type standards were in place, he said, EDS's newest data center design would be using it.

EDS recently chose a high-voltage AC power distribution system for its new “sustainable” data center, under construction in the UK. But the company would have preferred to use DC power, said Dale Hoenshell, global environmental sustainability manager at the IDC IT Forum in Boston this afternoon.

“The reliability of [DC power] is way beyond AC data centers but you don’t have UL-rated plug designs. We ended up landing on high-voltage AC,” he said.

What about the efficiency gains we’ve seen in AC conversions in the last few years? Haven’t those reduced the overall efficiency benefit?

“The difference [in efficiency of DC over an AC design] may have been about 7%,” when they did their analysis, and that could be a bit lower today, he acknowledged. “But the reliability factor [of DC] is way higher. It’s worth continuing to pursue this pretty aggressively. But we as an industry need to get to together and get DC plugs that are approved.”

As you rush to create your virtualized data center, don't forget your disaster recovery plans. Do your failover systems have the same configuration settings as the virtual servers in your production environment?

Avi Stone, director of marketing for Continuity Software Inc. in New York, doubts it. Of course, Continuity sells disaster recovery tools, so you'd expect he'd be doubtful. But he does have a good point. That is, disaster recovery sites need near perfectly identical systems and configurations as your production data center. If not, should your main operations go down, your backup systems may not work right.

With traditional application servers, that can be trouble enough. But with many virtual machines running on a single physical device, "it can affect many more servers," Stone says.

By the end of June Continuity will begin shipping Recover Guard 3.0. The update includes support for servers running VMware. That means the software will be able to discover configuration gaps between your virtual production world and your very real backup data center before disaster strikes.

Also, Version 3.0 now works with trouble-ticket systems from BMC and Hewlett-Packard, so the problems it uncovers can be reported through established IT channels. In addition, the update includes integration with your configuration management database.

Pricing starts around $2,000 per server.

Forget about squeezing more and more servers into each rack; the new emphasis is on the number of virtual servers you can get per physical machine, says IDC analyst Michelle Bailey. Michelle spoke at the recent IDC IT Forum in Boston.

According to IDC:

The average number of VMs per physical server today = 4:1

The average ratio in "mature" sites = 10:1

The average ratio in leading-edge companies = 25 or 30:1

Going forward, IDC expects that the rest of the market will follow the leading edge in maximizing the number of VMs per physical server.

Adding virtual machines per physical server increases server density without increasing power density - and all of the power distribution and cooling issues that entails. But only up to a point. Adding more VMs does require additional on-board resources, including more processors and more memory. That takes more power.

How much?

In my column Memory: The New Power Hog, IBM talked about shipping systems with as many as 64 DIMMs. Depending on the DIMM, those can consume as much as 14 Watts each. In contrast, some AMD dual-core processors consume about 68 Watts.