Lessons from the clouds

In my MBA studies, many classes touched on Herb Kelleher and Southwest Airlines. Mr. Kelleher was an excellent example of how leadership should be done, and he led Southwest to growth in a very difficult market. As I revisit my previous studies, I now see technology lessons that parallel the business lessons.

Southwest simplified operational costs by selecting a single model of airplane and focusing on high-density routes. They also rode out some short term spikes in oil prices by leveraging advanced purhcases of airline fuel.

To read how these lessons relate to IT, visit my article “Successful virtualization strategies learned from the airline industry” at SearchServerVirtualization, then come back here to leave comments.


Lessons from the clouds — 2 Comments

  1. Mark,

    You are correct that virtualization can learn many lessons from the airlines, but unfortunately there are far better lessons to learn than those you cited. I worked in the airlines building quantitative systems, fleet planning and airline purchasing. And I have been a CFO and CIO for large outsourcers. You are right, but below are the real important lessons:

    1) Virtualized servers are now pools or mathematically – sets with size. Pre-x86 virtualization, the set was a singleton (one server running its dedicated workload).

    An airline thinks of its planes as a certain set with size and quantities – how many 100 seaters of inventory exist. (Yes a seat is inventory to them.) So they think and model 100, 110, 150, 200, etc. set sizes of capacity. So the better analogy is server pools and the services they provide. IT now has a pooled server (a set) of some size that might be databases. IT might have a premium service for database service (log file shipping to move data every 2 hours) vs. simple database service (backup once each night).

    2) Pools of servers can now be upgraded or refreshed in place and higher performance capacity can be captured and brought online in managable set sizes. In fact, IT has to do this capture the savings. Most IT organizations just haven’t realized it because they still think in 3-5 year increments of refresh. That is dead wrong. The economics don’t work and industry standard TCO models fail for virtualized server pools. Plus IT thinks savings are realized by raising utilization. Not true. Savings are realized by not taking down utilization too far. This restart point defines how long the pool capacity will last. That is critical, but almost every client of ours misses that point. Airlines don’t change their fleet or schedule and bring their overall load factors (utilization) down when they do it. IT does – and they think that is perfectly acceptable to do it a lot.

    This is important, because upgrading servers more often for high software cost environments (i.e., capacity-licensed databases, middleware, OS, virtualization) is the key to reducing total compute costs. Server productivity is captured this way.

    3) Airlines are constantly moving different set sizes of airplanes around to better match to customer demand. That is too complicated to explain, but the equivalent for IT is upgrading a server pool where the improved perforamce has the biggest impact – and then cascading servers to other pools. This is where software cost come into play. Airlines do this and they refer to being long (too many) or short (too few) in different set sizes.

    In short, IT needs to perform the equivalent function of fleet planning and acquisition as the airlines. If you want to know more, check out our web site (www.ravelloanalytics.com). We don’t spell out a comparison to airlines, but it will detail the new economics of virtualization and server sets as services.

    Now to the ELA, which where you are a bit off-base. Airlines hedge fuel costs, they don’t pre-purchase fuel to secure a quantity discount. Hedging is used to take a risk neutral position – to lock in a price that secures a profit margin.

    LUV can do this because they can put up their strong balance sheet as a pledge (like a margin account) if their hedge goes bad on them. The counterparty to the hedge will want LUV to pay them. There is nothing controversial about it. A lot of the other airline’s would do IF their Balane Sheets allowed them. LUV’s BS is strong. The rest have been in and out of bankruptcy. In finance terms, they pose too high of a counter-party risk. Think of AIG.

    Most IT organizations far over-purchase required licenses in order to secure a price discount. Here is an example:

    If your new pool had MS licenses for SQL Server (unlimited vm’s) and you rented those licenses for $800/month and now throw in some OS, virtualization software, etc. and we are now up to $1000/momth per processor. A very good server costs $10k per processor — the breakeven is 10 months, not 3-5 years.

    How do think IT would add capacity to that pool? Most of my clients would add more servers and jack up their software bill. Instead they could upgrade some number of servers in the pool, gain the required incremental capacity, and avoid adding any new licenses. When server performance increases annually at 25-50% and software is 3-10 the cost of the server, knowing how to upgrade/refresh is the key.

    The airlines spend a boatload of money solving these set optimization, capacity/demand and queuing problems. IT doesn’t. That is the lesson IT will need to learn from the airlines.

    1) Planes do represent standardized platforms, but the better analogy is not one aircraft type (like LUV), but set size. Airlines think in terms of 100, 110, 120, 150 seat platforms. With virtualized servers, that is akin to pool of servers

  2. Dan, you have some great points. Thank you very much for adding to this discussion.

    I like your analogy of set sizes, which is very applicable in large organizations. This year’s model of server will not be available forever, so variations will have to be introduced over time. I was referring more to the goal of limiting variations to increase efficiency, not necessarily restricting options to a single model. My analogy to SWA’s single plane model may have been misleading.

    For the ELA, those are tricky. As a customer, I worked with vendors to (though the real credit goes to our highly aggressive purchasing dept) negoatiate ELAs that saved us A LOT of money. These ELAs were very advantageous to our organization, but we were a large company with some serious buying power in those negotiations.

    These ELAs would include a certain amount of purchasing commitments over a period of time, but they would also lock in a set discount for all items not included in the pre-purchase agreement. These were very lucrative for us, and the size of our purchases made these beneficial to our vendors as well. Like SWA and their competitors, not all companies have the ability to negotiate the deals I am referring to.