The data centre diet

How to slimline your data center – Revera discusses.A recent article in the Economist cited an observation made by Web 1.0 cheerleader Wired Magazine, which said: “As businesses, as well as most organisations outside the business world, begin to shift from hierarchical processes to networked ones, nearly every facet of human activity is transformed in some way by the emergent fabric of interconnection. This re-organisation leads to dramatic improvements in efficiency and productivity.”In many respects, the same hyperbole comfortably attaches to IT’s progression from device centricity to today’s information centricity. In the late 90s, ERP challenger J.D. Edwards exhorted “collaborate or die” (a strapline subsequently dropped following September 11), promising similar benefits of efficient connectedness, never mind that some thought adopting an ERP system was like pouring concrete into your business.Today, similar sounding promises flood the market, though more popularly bundled with the idea of unifying computing infrastructure and cloud services. Some even say that as the traditional computing stack disappears from view and abstracted IT resourcing takes over, business becomes more like technology. That is: faster, more efficient, and better focused on the ‘core’ business. The success of virtualisation and pay-as-you-go business model like SaaS, both anchored to Wired’s idea of ‘emergent fabric’, show that the hype can on occasion match reality. ‘New economy’ technology is available and at your service.And when underlying, largely virtualised technology defines so much of today’s business on top, success is only as good as your foundation.  So, back to the story headline: does your data center fulfil the promise of new, leaner efficiency, or is it still dishing up the same old computing services diet, simply from another location, which ends up weighing down business?In my mind, more than a few businesses are still being fed on the old diet, despite recent headlines heralding new data centers. Mostly it’s because these new facilities follow an 80s design format. Their best response to keeping blade servers cool is to build higher ceilings.The big problem is when physical control is poor, virtual performance is as subtle as a repeat Aitkens dieter. The first big question customers must ask their data center provider is: “Can the data center withstand a power outage?” The response you want is, “Yes, we’re okay, because we’ve got a generator.” But generators don’t kick in for at least 12 seconds. UPSs keep computers going, but air handlers stop. In a blade server centre that’s an awful lot of hot air looking for a meltdown, and just 12 seconds can take 12 hours to dissipate.  The other issue facing buyers of virtualised utility capacity is oversubscription, which leaves customers exposed. Providers must guarantee the absolute minimum level of on-demand virtual machine units available. Taking into account that you don’t run everything at 100% all the time, you will generally get more performance than guaranteed minimum levels. But at the least, every virtual machine unit should have another unit as a guaranteed minimum. Utility service providers who oversubscribe will arbitrarily specify a minimum, and when extra performance is required it will be robbed from someone else. For example, you might have 10GB of RAM spread across 10 servers (1GB of RAM per server). But if they then sell 20 virtual servers, without increasing available RAM, the service will be oversubscribed two-to-one. When all those servers run at full capacity, only half of the resources will be available to do the job. There must be a one-to-one relationship between the actual physical hardware and units of virtualisation, or performance will be impaired.But looming well ahead of these considerations is anticipating likely ripple effects when businesses first move from physical to virtual. This goes much deeper than hardware and operating systems, and includes environmental impacts, such as cooling and power. Server vendors will say, “Yes we’re very sustainable because we’re giving you the opportunity to consolidate.”Inadequate redundancy will also find you out. For virtual environments, that means clustering. All underlying infrastructure must have N+1 end-to-end redundancy, for full automatic failover. You must have a hot sphere, so if one physical server fails, all the virtual machines running there will automatically restart on another physical server. In some cases virtual server capacity may look cheap, in which case you might want to look a bit deeper and count the number of physical servers clustered underneath. I’d put money on a single physical server running in isolation, and when it fails, so too do the 20 virtual servers it supplies. They remain paralysed until the physical server has been replaced and the data restored. Virtual servers must run in a blade agency, connected to redundant switches for both storage and networks. So if any one component fails, server availability will remain stable. When you can safely close the bonnet, eyeballing the business heads controlling the buttons will provide useful clues about delivery style and nimbleness.Revera’s client the North Shore City Council (NSCC) said innovation was the new battleground for IT infrastructure providers. NSCC general manager of customer and information services, Tony Rogers said customers should be able to get a cost baseline that’s going to stay reasonably static through the life of the contract. “It’s about cost containment, with scope for the provider to be creative and innovate to make savings, without compromising business as usual,” he explained. Rogers said CIOs must demand new opportunities to contain costs, despite continuing organisational growth.“It was quite interesting talking to other vendors about the flat cost-line approach. The bigger ones usually shy away from it, preferring to stick to cost plus, whereas people like Revera have a better understanding of service provision and costs to be able to say with some certainty, ‘yes, we can do that’,” Rogers said.Though the underlying driver for utility  accelerated the council’s adoption of e-government and electronic services, Rogers said removing IT from the balance sheet has been a highlight. “Finance guys love it. Accounting for leases under the old regime was difficult and there’s always an argument over residual value. IT budget increases are notoriously hard to get. Part of the deal with Revera is disaster recovery and business continuity cover, including replication in a second data center. Not in my wildest dreams could we have justified doing this ourselves,” Rogers said.