The SOA Consortium hosted two end-user panels at Gartner’s Application Architecture, Development and Integration Summit this week. Notes taken during our Cloud Computing in the Real World panel follow.
Session Abstract: Is your organization considering cloud computing? Are you tired of hearing about simplistic (email and calendaring), enterprise irrelevant (Twitter, Facebook), and extreme (Google Datacenter) use cases? You aren’t alone. In this session, government and enterprise practitioners will discuss the cloud computing use cases they are considering, actively pursuing, and rejecting. In addition to the use case specifics, the panelists will share insights on financial benefits, true implementation costs, assessing and managing risk, governance, standards and their cloud computing wish list.
Moderators: David W. Cearley, Gartner and Richard Soley, OMG / SOA Consortium
Clark Dorman: Specific use case, analyst in DoD collects a lot of images, genetic algorithm to review images. Use cloud computing to optimize processing. Started with in 18k server, got output in 8-14 days. Now, use Amazon EC2, 100 images at a time, 12 hours, get run done for $130. There is an opportunity to parallelize further and reduce runtime in half again.
Picked EC2 because it is the biggest. Clark mentions Eucalyptus as way to hedge against lock-in. Use Appistry to orchestrate startup and runtime execution.
Andy Lapkin: Looking at cloud, on premise they pay for availability, not for use. Build KBB.com to peak. This is expensive, for off-hours, plus hedge against growth. Sized for extra 25%. Example of unexpected traffic was “cash for clunkers” program. A second use case is in data management. KBB has a lot of data not shown on website that is used to produce website. Also, data provider to car dealerships, etc. Data management work is done in batch.
KBB uses Azure and SQL Azure for their cloud computing environment. KBB is a Microsoft shop. Now, testing KBB.com failover site in Azure in SQL Azure. Biggest lesson learned: Transition to cloud is as good as your application maintenance.
Joseph Larizza: Customers are high net worth individuals, mostly from financial services industry. The IT budget is a shoestring, similar to community banking. Joseph has big banking background. Some providers Digital Insight, Rackspace. Full banking site, portal to members, all in the cloud. Core competency is banking, not data center management. Cost is critical factor.
Risk Question: Look at it as risk ledger. Responsiveness of provider vs. key employee risk. Compliance can be provided by cloud, more effectively than if done internally. Surprise, the amount of support you need to do, such as patching, backups, security.
Online banking application, could have built their own. But, not core competency. Could never keep up with feature/function of a Digital Insight. Instead, use Digital Insight, which is an Intuit company. Core banking platform is all in the cloud.
Implemented single sign-on across these applications; does have Fieldpoint branding.
Q: Security Concerns?
Joseph: Rackspace compliance and security, third party providers, learn their procedures. They are careful about what data is stored in the cloud. Didn’t elaborate further on this.
Andy: Security isn’t biggest concern, given the type of business application.
Clark: VPN on their side, SSH into Amazon. Data isn’t classified, but still don’t want to “leak all over”. Data is encrypted on the fly. No data resides in the cloud. Data is brought up at startup. Have very specific startup process. (Again, use Appistry)
Q: SLA with Amazon? Is the data really gone when the images are shut down?
Clark: Matter of trust, trust that policy and agreements will be adhered to. Security isn’t a thing, it’s a process.
Q: Cost of moving the data, latency?
Clark: For this use case, 6-8 hours of processing, 20 minutes worth of data movement isn’t an issue.
Q: What is your interest in standards in the cloud computing space? Cloud interoperability? Avoid lock-in?
Joseph: Very interested in the ability to move cloud to cloud, for risk management, and to move work closest to current customer base (time zone).
Clark: Want to develop application at highest level you can. Make the lower levels someone else’s problem.
Andy: Not moving everything to the cloud. Will always have a mixed model. Will always have datacenter to fall back on. The real issue is the operations work to move from running on one cloud to another.
Q: New job roles?
Andy: Monitor the same way as on-premise. Use Keynote for monitoring.
Joseph: Think about operations (backup and security) differently. Can’t just stick equipment in the mix. Don’t have control over the equipment.
Clark: Systems administration changes to cloud administration. Don’t worry about hardware layer, but still need to manage the environment (to deliver the application).
Q: Old wine in new bottles?
Clark – what’s new is dynamic nature, speed that you can do things, low costs. 8 cents for CPU hour.
Andy: Spin up new instance in 6 minutes. Scale, knowing you can serve 14 million visitors a month.
Q: What needs to improve in cloud computing?
Andy: Azure needs third party support for management tools. Doesn’t want to write his own. Needs tooling to sift through all the log data.
Q: [for Clark] Your project is a proof of concept, if DoD accepts this product, will it remain in the cloud, or move somewhere else?
Clark: It is up to the government to decide. The government does have a few clouds – NIST, DISA Race, and GSA Apps.gov. This application could land on existing government cloud.
Q: [for Andy] Considered running base load on premise, then burst peak to the cloud?
Andy: Something they are considering.
Q: [for Joseph] Said want to get to zero servers, does that mean Opex will outpace Capex? Is that ok?
Joseph: Not for their size. In general, economic model is service first, no related employee expense, if not appropriate, have inside.
Andy: KBB uses SaaS applications to run business, such as CRM (Salesforce.com)