Economic impact: Scale Out vs Scale Up, Part 2

This is the second part of this series, the first part can be found from here.

  • Starting point is very similar as with Scale Up sizing exercise
  • Collect some performance data from existing system in order to estimate growth rate
  • Select over how many years solutions should be amortized
    • Typically three to five years
    • With Scale Out you don’t initially size / buy solution for the whole three to five years, but it is good to know your upper limit, especially if there are limits in Scale Out cluster size
  • Based on collected data, plot required scaling over years
  • You might also want to plot different scenarios at different growth rates

Example sizing exercise with Scale Out model:

Screen Shot 2016-07-01 at 15.52.49

  • In this case
    • faster growth rate = yellow line, 40% annually
    • normal growth rate = green line, 25% annually
    • slower growth rate = blue line, 10% annually
  • Estimated required performance after five years is 9,2 ( of imaginary performance units)

Example: Smaller scaling unit with Scale Out

Screen Shot 2016-07-01 at 15.56.01

Typically with Scale Out model the scaling increments are much smaller than with Scale Up model. Some manufacturers have only few different sized “bricks” or “nodes” to build the cluster, some have more. With smaller steps to scale, the cost steps are also smaller. In this example, steps are in increments of one unit instead of increments of five units like with Scale Up model.

Some manufacturers have a limit in Scale Out cluster size and some do not. Cluster size limit only comes into play, if your scaling need is such that it would exceed cluster size limit and would require multiple clusters to serve your workload. For example Nutanix does not have maximum cluster size limit, where as NetApp has a limit in cluster size, but with NetApp you can combine Scale Up with Scale Out and build quite large setups.

In order to fully utilize the economic advantage of scale up model, just make sure that your chosen solution scales enough to meet your needs. This is where escpecially up-and-coming new Hyperconverged solutions are different. Most of the new players on this field have limited experience and qualification done, some might have architectual issues with larger scales, so they have decided to limit their cluster size, some quite dramatically, to only 4-8 nodes / cluster. Such small clusters are fine for many use cases, but not good candidates for large enterprise solutions where predicting future is difficult and/or large scale is required.

Example: Shorter timeframe for Scale Out prediction

Screen Shot 2016-07-01 at 15.57.54

One of the advantages Scale Out combined with enough room to scale, is that you can start small rather than sizing the solution for the whole lifespan of the solution like with Scale Up model (to avoid hitting Scale Up scaling limit). Since most scale out systems allow you to add capacity and performance without disruption to the existing workloads, it is easy to add more resources to the cluster as needed.

Like in this example above, maybe size your solution maybe a year a head. This will dramatically decrease the risks involved in predicting future. The shorter time frame – the more accurate your prediction typically is. Even if your sizing went wrong, you can react to it by using different hardware configuration for the additional nodes. At least with solutions that allow to mix different kind of nodes in the cluster, like Nutanix and NetApp.

Once again there are limitations with some solutions, some allow only one type of “node” or “brick” used in a cluster. With such systems some of the advantages of Scale Out model are lost as you cannot react to different needs when expanding your systems. Ability to mix nodes is an indication of product maturity and requires more work and qualification from the manufacturers, so the less established players usually lack in this field.

Example: React and Expand when required

Screen Shot 2016-07-01 at 15.59.30

With Scale Out model you can react to changes more easily and stagger your purchases over few years. In order to know how to react, having built-in analytics capabilities will help tremendously, otherwise you are making decisions blindly or have to use third-party analytics tools.

When staggering your buying over longer period of time, you can also benefit from tehnology advancements. While a similar but newer model node might cost about the same amount of money in two years, you will get more performance, capacity, etc with the same amount of money. Take benefit of Moore’s Law and ever decresing prices of SSD (and  HDD) capacity, get more bang for you buck.


Scale Out model is much more efficient way of scaling than Scale Up model

  • it eliminates some of the guesswork related to future
  • it can limit under and over provisioning
  • it can limit upfront costs, you can start small and increase your environment incrementally as needed
  • it allows to react changes in business needs
  • it allows to expand your systems without disruption
  • you can take advantage of evolving technolgy and not buy outdated tehnologies

Be aware that “Scale Out” is a new buzz word and different manufacturers have different view on what qualifies as “Scale Out”. If a “Scale Out” solution has too many limitations with scaling for your needs, it might not be a true “Scale Out” solution, it might actually be a “Scale Up” solution in disguise. Like slapping some lipstick on a pig does not make a Miss Universe, slapping some marketing buzz words to a solution does not magically turn a “Scale Up” solution into a “Scale Out” solution.

Thanks for reading


Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s