IT Management and Cloud Blog

« | Home | »

Thinking Out Loud – How will seat pricing work in the cloud?

By John | July 21, 2008

In a recent IT Management podcast I asked Tarus Balog of OpenNMS if he though his support only model (no software sales) would give him an advantage over his competitors. Unfortunately, he didn’t really respond to my question as Tarus is not a big fan of the clouds. However, it started me thinking about per seat vendor pricing and the cloud. I started thinking how current vendor pricing is going to translate in highly virtualized environments like clouds. If you’re a vendor that charges by the box, server, or server size how are you going to charge for your software in a world where servers are a pure commodity? Servers in a cloud are spun up or down based on demand and in may cases they are ephemeral.   As theses cloud environments mature customers will want to use commercially supported software in these environments. How will commercial software vendors charge for their software in a cloud?   Software usage in the cloud will eventually migrate from free unsupported software to more commercialized versions of software.  This is not only a proprietary software issue. There are a lot of open source vendors that have adopted enterprise pricing models that also charge by the hardware configuration.  Customer’s that use parts of their business critical services on the cloud will want “enterprise” grade support contracts with their vendors. Today most proprietary and even open source vendors charge by the customer’s hardware configuration. There is no hardware configuration in a cloud.   How are vendors going to adapt to customer environments that are elastic and hardware-less?  What does a vendor charge if a customer uses 1000 servers (instances) for only an hour?  Animoto is a startup company that had to spin up 5000 Amazon EC2 instances in one week. They started on a Monday with 50 instances and by the end of the week they had 5000. Then a few weeks later they were back down to about 100. If they were using using Oracle for their database or the enterprise versions of Hyperic or Zenoss for their monitoring how would they have been charged?   Other companies that do digital processing in the cloud, for example Digital Chalk and Smug Mug, might have 10 servers during the day then a couple hundred and maybe a 1000 that evening.

A few years ago IBM had to adjust their pricing model to deal with behind the firewall pricing for virtualized environments. Prior to thier change they were charging on a per CPU price. This model didn’t work in virtualized or blade server environments.  To adapt IBM changed their pricing model to a wacky service units scheme that is built on a very complex matrix of vendor/server types. The ideas was that IBM has a list of every type of hardware box and they assign service units based on the potential processing that could occur on that box.  Even their sales guys can’t figure it out how it works most of the time.  These types of adjustments are going to get harder and harder to keep track of as available CPU’s become a commodity.  Case and point, I had a customer last year that upgrade all of their hardware to the new IBM p590 series. Their vendor, a business partner, convinced them that they could upgrade to the new boxes and save money on their hardware by only using 25% of the capacity of the new boxes (an eat what you kill price model). Unfortunately that only turned out to be a hardware savings. About 6 months later Deloite and Touche did an audit and the customer was hit with a 3+ million true-up bill for their software.  Some large customers are building their own clusters of commodity CPU’s where the vendor might not be identifiable.  Cloud infrastructure providers like Amazon and Flexiscale are going to make this even more complicated with their hardware-less offerings and resource allocations as pay-as-you-go.

At a minimum some software vendors are going to have to change their pricing models. However, I wonder if this continued trend of the comodiization of CPU processing is not only going to change how a software is sold but it might also lower the cost of commercial software.

Topics: other | 6 Comments »

6 Responses to “Thinking Out Loud – How will seat pricing work in the cloud?”

  1. Benjamin Reed Says:
    July 21st, 2008 at 12:54 pm

    I can at least kind of answer for Tarus — the funny thing about our (OpenNMS Group’s) pricing model is we don’t *care* how many servers you run on. We only care about the bandwidth between us and you, the customer, and our pricing reflects it.

    If you’re a small shop and only have one or two admins who need to open tickets, you can get the basic support contract. If you have larger needs, you tend to have more staff who need to be able to open tickets (or if you are using the more complex, esoteric features of OpenNMS not covered under basic support) and you get enterprise support.

    It’s fair for the customer (there’s no artificial fiddling of resources on the customer’s part just to make the support contract happy), It’s self-regulating, and it keeps per-node, per-cloud, or per-whatever-processing-measurement out of the equation.

  2. Chris Sears Says:
    July 21st, 2008 at 9:57 pm

    Perhaps vendors will just move to a more cloud-friendly CPU and time-based licensing, like logical CPU hours. This would be easy enough to track since EC2 and similar services are already doing the time and CPU accounting.

    If necessary, a licensed component could “call home” or contact a local license tracking service when starting up and shutting down to provide an external metering system.

    There is some president for licensing moving in this direction. Both Microsoft and VMWare offer service provider licensing where companies track their own license usage and just pay for what they use each month. Monthly license charges make sense for service providers (like hosting companies) because they usually bill their customers on a monthly basis. For clouds, hours (or fractions of hours) are the new time quantum. It seems reasonable that they would just extend the existing SP licensing to hourly billing.

  3. ArcGIS Server Cloud Licensing « The Memory Leak Says:
    July 22nd, 2008 at 11:09 am

    [...] Out Loud has a good post about the difficulty of adapting per-cpu licensing models to the [...]

  4. John Says:
    July 22nd, 2008 at 2:28 pm


    The problem is that there are going to be so many variations of platforms (private, public, cloud, non cloud, …). It is going to be very difficult for vendors to keep track. It is already a nightmare for IBM with virtualization. Also not all clouds are going to be SP based.

    However, I do agree some kind of metering is the only thing that can make sense … unless .. everything becomes open source and free :)


  5. Reuben Swartz Says:
    July 23rd, 2008 at 9:15 pm

    Software vendors have long relied on pricing mechanisms that were “easy to count” instead of “reflecting value.” Cloud computing, SaaS, and even multicore CPUs have blown up these models.

    (We’re actually doing a webinar on SaaS pricing models on July 30– registration links available from our website for those interested.)

  6. Bookmarks about Vendor Says:
    January 21st, 2009 at 5:00 pm

    [...] – bookmarked by 3 members originally found by jaimexx14 on 2008-12-22 Thinking Out Loud – How will seat pricing work in the cloud? – [...]