Tuesday, May 29, 2012

If I Could Do It Again...

Lately, we have been doing some concept work at UGA EITS around Amazon Web Services (AWS) and what its cloud computing, storage, and other services could mean for us. I am very intrigued by the possibility of moving from a model where we throw large one-time dollars ($) at capital needs, buying capacity for some hypothetical peak use that happens sparingly. This approach also drives our software licensing costs through the roof, because those licenses are also specified for the maximum peak usage scenario. I wonder, with most UGA students now gone for the summer, how many servers in the data center are sitting idle?

As we have been thinking about the possibilities of leveraging AWS, I have been reflecting on my experience bringing up the IT infrastructure for a new college of engineering in Qatar (2003 – 2007). Today, the IT infrastructure in place at the Texas A&M University branch campus there mirrors what you might find at any university. There is a data center, many racks of servers, a SAN for storage and backup needs, and a large high performance computing center to support faculty research. Had cloud-based services like AWS been viable in 2003, how might things have been done differently?
  • Our network design and implementation plan would remain unchanged. Getting the Qatar Foundation and Qtel to give us two OC3 equivalents to the United States, connecting to Internet2, was a brilliant move by Pierce Cantrell and one that continues to pay dividends. 
  • For authentication and identity services, we would have extended current LDAP/Kerberos/Active Directory services from the main campus into an AWS EC2 instance and then down to a local physical server in Qatar (as opposed to bringing up our own AD forest). This would allow employees to authenticate with the credentials they already know and would have allowed us to get better economies of scale around identity management.
  • For employee email, we would have adopted Live@edu because it works better with Microsoft Outlook calendaring. For students, we would have put them into the student email system from the main campus (instead of bringing up a separate physical Microsoft Exchange instance supporting everyone).
  • No way to get around the need for a local file server (because of network dependency and latency issues), but one physical server acting as an AWS S3 Storage Gateway would make sure that all files are automatically backed up to the U.S. For PCs and servers (virtual and physical) an AWS based solution like JungleDisk would also ensure that backups automatically reside in the U.S (as opposed to a tape backup system in Qatar).
  • All application servers, database services, Web servers, and the like would be delivered through AWS EC2 and RDS instances (as opposed to bringing up dozens of physical servers). In regards to PCs, the traditional desktop model would reign – network dependency and latency issues would make virtual desktop computing too risky. 
  • In regards to high performance computing, faculty researchers with the need for parallel processing capabilities would have been handed credits for AWS EC2 instances and S3 storage (as opposed to building large computing clusters physically). For faculty whose research required large shared-memory computing, such as that for visualization, a large shared memory machine physically on the ground would be required (as was implemented in Qatar).
Nine years of technological evolution, together with an identical amount of experience and maturity on my part, would lead me to take a fundamentally different approach to building out the IT infrastructure for this campus. In the end I would expect that services would be more flexible, require less upfront capital investment, and benefit from more efficient and productive disaster recovery / business continuity mechanisms. But the real benefit is this - instead of spending so much time on physical IT we might have been able to focus just a little bit more on the people side of IT. That’s what makes the adoption of cloud-based services like AWS truly compelling.

2 comments:

  1. Thanks for providing your thoughts. I would not be so quick to avoid virtual desktops. Traditional desktops are expensive to maintain and provide potential access points into the network for things you want to keep out. As VMware, Citrix, and Microsoft work to overcome latency and other potential barriers to their use, I believe you will see a much broader utilization of thin clients in educational environments.

    ReplyDelete
  2. I agree with your sentiment, but the only difference is the context for this need. The customers were in Qatar and the VDI's would have been on Amazon's infrastructure in the U.S. If we lost our link for a day now no one can access their desktop computer. That is why I would have done traditional desktop computing but storage and backups to the cloud.

    ReplyDelete