Global organisations often face the challenge of distributed IT environments and, consequently, duplicated effort when it comes to IT. And it was this scenario that Chris Haaker faced as principle architect at Reed Elsevier.
The global company encompasses four business units: Search information company LexisNexis, science and health information provider Elsevier, publishing arm Reed Business Information and events organiser, Reed Exhibitions. Until three years ago, each business unit ran its own information technology, but the company could see the opportunities of potential cost and efficiency gains of IT centralisation.
"Our company culture was such that we didn't adopt new technology very quickly - we took a long time to evaluate things," Haaker told attendees at VMworld 2011 in Las Vegas. "As we moved to a centralised function we ramped that velocity up a bit."
See more the VMworld action in the 'day 2' slideshow
The company had 102 physical servers spread over 12 countries and 32 staff who held responsibility, at least in part, as Exchange administrators. The result was a lot of duplication of effort. Many of its data centres were at near capacity and as such, virtualisation was an obvious path.
The other big issue, Haaker said, was the individual priorities and agendas of the business units. That can make application management and upgrade paths tricky, and Exchange was a shared application across the organisation.
The team decided to skip Exchange 2007 and go straight to Exchange 2010.
"When Exchange 2010 was in beta we decided to jump straight over," Haaker said.
By the end of the migration Haaker and his team had consolidated to four locations, with centralised and virtualised functions. The organisation could move IT staff away from commodity tasks and towards things that would add more value to the business.
It helped that virtualisation was already part of the IT skillet - the company was about 65 per cent virtualised. Although it had not virtualised Exchange before, it was a risk the organisation was willing to take.
Scale up or scale out?
Some organisations will want to put all their mailboxes on one server. Others want multiple servers with less mailboxes on each. It is a subjective decision, according to Haaker.
"It goes with the philosophy of your business," he said. "There are pros and cons for both. I'm not going to try to sell you on our decision; I'll tell you what we decided. We decided to scale out as we saw we would have more flexibility with more mailbox servers than with a one-basket approach."
DAS versus SAN
Storage was the next decision. Elsevier decided to go with SAN, again based on existing skillets and experience.
"There's a lot of literature about direct attached storage… but with a SAN you get a lot more flexibility," Haaker said, adding that the decision was also influenced by past experience with NTFS data corruption. A SAN, he said, means you turn around a new disk on the server in a matter of minutes.
In addition, while much is made about Exchange not being virtualisation-aware, the team has not seen any differences in overheads. By the end of the migration, they had consolidated data centres - two in the US and two in Oxford in UK - and went from 112 physical hosts to 12; three hosts per data centre.
The team initially decided to go with native Windows backup, but soon ran into problems with the environment when Windows would only backup against active databases.
"When we started out, it was normal performance and backup was running smoothly overnight with no issues. All of a sudden we started seeing disk contention," Haaker said.
"We found out that as mailbox sizes grew and database sizes increased in six to eight months of operation, backups were starting to spill over into the AM period when people were logging into the network."
A move away from Windows Backup to Microsoft Data Protection Manager eliminated the problem.
IT teams must also consider the number of hotfixes and client platforms, as well as the potential for free/busy issues. Internet Free/Busy is a feature of Microsoft Outlook that lets users see when others are free or busy when scheduling meetings. The lesson: When you migrate executives over, it's a good idea to migrate their assistants at the same time.
Further, lab testing before deployment is vital, particularly for replicating things that are unique to your corporate environment.
Once migration begins, Haaker said, don't be afraid to "let it sit".
"Migrate and wait," he said, "because stuff will pop up that you don't expect. You have to have some kind of coping strategy in place. Let it sit for a week or two to make sure everything is smooth and gremlins don't appear. Talk to all parts of the business. Do due diligence to ensure you have every base covered as best as possible.