WavebreakMediaMicro - Fotolia
With consolidation and colocation on the rise, data center operators need standardization for what used to be a once-in-a-lifetime event: the data center move.
"People will go out and wing it, but you end up on your third or fourth one and think, 'Why are we doing this alone?,'" said Danny Aldham, senior program manager at TELUS Communications Company, a Canadian telecom provider. "Other people are in the same boat."
With several data center moves behind him -- and several more on the horizon -- Aldham started compiling best practices and templates for data center migration (See the list).
The majority of processes occur in the same way -- no matter the data center in question -- he said: "How do you physically move servers, how do you track inventory, how do you power everything down?"
A dependency mapping exercise -- down to the hardcoded IP addresses -- will show the cascade effects of failure, influencing what should move when, and what shouldn't move with what else, said David J. Cappuccio, a research vice president at Gartner Inc. in Stamford, Conn.
Even elements that are unique and require specialized treatment have some lessons that carry over from one move to the next.
David J. CappuccioResearch VP, Gartner Inc.
"There are only so many ways to move a server, but for applications, the procedure will change every time," Aldham said.
That doesn't mean you can't devise a path that works generally: Shut down a line-of-business app at X time, and expect it back up by Y time. From here, each specific app team will develop a detailed move plan.
At some point, you'll have to freeze new installs or halt new app development while the back end moves, Cappuccio said. Application managers need you to plan around them, not vice versa. And once you're operational at the new site, your job isn't done.
"Benchmark user experience before and after the move," Cappuccio said. "This will stop the rumor mill that IT 'messed with' the application."
Expect to spend $1,200 to $6,000 per server (logical or physical) to move a data center, Cappuccio said, and timing is as critical as cost. He recommended bringing in a project manager that has an established methodology, whether to direct the whole move or at least to consult and offer training on their methods.
Experienced data center consolidation and move specialists approach the project from a different perspective than IT teams. For example, Cappuccio said, it might make sense to bunch together equipment on the hardware end, but it doesn't make sense from a service perspective to move those pieces together. Virtual environments seem easier to move, but often cause the most problems because of interdependencies and shared resources across hardware.
Expertise from a colocation
Credit bureau TransUnion had just gotten its data center migration project underway in 2013 when Josh Neyer joined as the global head of data centers. TransUnion built out a colocation space with data center provider Ascent LLC in Northlake, Ill., near TransUnion's Chicago headquarters. TransUnion considered data security and control of primary importance, but also wanted the abundance and redundancy of power, facility expertise, and costs sharing that colocation provides. And it wanted a fast build.
The benefit of partnering with a colocation provider was that Ascent handled everything "to the plug" -- building out the infrastructure as TransUnion planned its equipment refreshes and moves. The company collapsed disparate locations into the new data center within 12 months of signing its lease.
Neyer worked with the space provider to make the move happen quickly without surprises. It started with an experienced crew that has "been there, done that."
He designed aisle formations and pre-ordered cabinets to be available as soon as the space was ready.
"Pre-stage everything," he advised.
TransUnion chose pre-terminated power whips in steel pipe conduits; these and all its RPPs were installed prior to move-in, saving about six weeks of time.
Data center move checklist: Documents and procedures
- Business cases (why undertake the move, reference cases, profitability estimates) and cost models that take contingencies into account. Don't forget that this move affects the whole company.
- Discovery processes: Inventory of hardware, applications, databases, firewalls, load balancers, storage, etc. Don't expect this to be easy.
- Create contract templates for all third parties. Secure vendor agreements that warranties and service contracts are valid post-move.
- Project plan: This includes a high-level timeline and action plan, budget, communications strategy and change management plan. It should also cover the skills required to move the data center, architectural information on the new site, a risk management plan and quality assurance policies, and methods of procedure to migrate hardware and applications (swing applications from old to new equipment, or lift and shift the hardware?).
- Load tests and failure tests at the new site prior to full production operations.
- Plan for vacated space: Will it be used as additional capacity, renovated or left empty? Will it be sold?
- Tracking and reporting tools and dashboards. Application performance, network latency and other benchmarks met or improved.
Neyer said the key is to allow enough time for design and fabrication of components. The fiber and copper cabling, for example, took about two months to design and two more to make, but with advanced planning, the goods were ready in the first week of occupancy, and the team tackled the cable plant in a month.
"By having Ascent manage and maintain two hundred plus power whips for the servers, storage, mainframe, etc., we were rack-ready with cabling and power within eight months of signing the lease," Neyer said.
The move was also an opportunity for improvement. Everything connects back to core network areas and core storage areas to avoid rigid layouts. TransUnion upgraded its network architecture to spine-leaf technology both in the new location and in its main Chicago data center to pair the sites.
"Strategic refreshes were happening in certain instances of our portfolios," Neyer said. The team inventoried all the IT equipment: What they needed to buy, relocate and install.
Refreshes meant new purchases could go into place at Ascent while the apps ran on older hardware until it was operational, minimizing risk. Neyer moved existing hardware given the right circumstances, such as current-generation servers that didn't justify replacement. In this case, Neyer said, they replicated the entire supporting infrastructure and copied data over before the trucks left with the servers.
In some cases, a small expenditure will save time. Neyer purchased extra storage for his company's mainframe, for example, which minimized the disruption when actually transporting the mainframe into the new space.
Open lines of communication
Don't overlook communication. Shared calendars and dedicated discussion groups are easy to set up and they inform everyone on the team. A Web portal can share daily, even hourly, updates, displaying role-based levels of information. These come in handy when the project is months along and people aren't as actively engaged as they were at the start, Cappuccio said. This portal also acts as a record of work documenting the move.
"Everyone got involved and excited. ... This will make all the technical building and processes easier," Neyer said. "There are dozens, hundreds even, of people that are important to delivering this project at a huge scale on time and on budget."
A successful data center move tells the business that its applications and data are safe with IT, he added, while a badly executed move sends the opposite message.