X

Move your VMware and KVM applications to the cloud without making any changes

  • December 15, 2014

The curious case of cloud-based virtual training - increasing utilization and decreasing cost per training environment

Most major independent software vendors (ISVs) and other enterprises that develop and produce software, security appliances, networking appliances etc. provide their corporate customers training on their products, new releases, specific features, etc.. These enterprises often hold classroom training, or instructor led training sessions, where a fixed number of pre-provisioned environments in the datacenter can be used to train up to that number of students in each training session. This post shows how training managers can overcome under-utilization in slow times, excess demand in peaks and avoid provisioning each and every environment for each session. Training managers can, in fact, reach maximum utilization and provide as many training environments as required to meet the variable demand for training by using Ravello to run training environments in Oracle Cloud Infrastructure, while only provisioning the environment once.

Summary of financial and economic metrics of using the public cloud with Ravello

  Revenue Profit Margins Cost per student Utilization
With Ravello

Detailed comparison of economic/operational metrics of running training environments in the datacenter as opposed to using the public cloud with Ravello

Hosting

Month 1 2 3 4 Total
Demand 16 28 97 28 169
Capacity 40 40 40 40 160
Cost $5,652 $5,652 $5,652 $5,652 $22,609
# Trained 16 27 40 18 101
Utilization 40% 68% 100% 45% 63%
# Not trained 0 1 57 10 68

In the public cloud with Ravello

Month 1 2 3 4 Total
Demand 16 28 97 28 169
Capacity
Cost $3,360 $5,880 $20,370 $5,880 $35,490
# Trained 16 28 97 28 169
Utilization 100% 100% 100% 100% 100%
# Not trained 0 0 0 0 0

 

A real-life application for training

 

Let’s consider a training for a new release of a network security product. The application consists of 5 virtual machines - of which 2 are firewalls and one is a management VM - and 3 subnets. Each VM has 2 vCPU and 4 GB RAM, for a total application size of 10 vCPU / 20 GB RAM. The training lab existing infrastructure has a 10 student capacity. The servers compiling the lab sit in the enterprise hosted datacenter. For every class the training manager or instructor has to provision up to 10 of the application environments, depending on the actual class size. Provisioning these environments for each student before every training session is time-consuming and frustrating.

The cost of hardware (servers, storage, and network equipment), hosting and internet connectivity for these environments with a leading dedicated hosting service company will be ~$5,652/month.

The Challenge: Variable demand for training; Fixed capacity in the training lab

In the lifecycle of the enterprise’s product, there are lulls and peaks in the demand for training. Corporate customers will have certain availability and need for training in the steady state, and demand and requirements for training will hit a peak around a new version release of the product. This is a substantial business inhibitor: while the average utilization of the training lab is very low (since in the steady state there are fewer than 10 students per training session), still in peak times there are not enough environments to meet training demand requirements, and many end users are left not trained on the relevant set of features or the new release.

(Not) meeting demand for classes while under-utilizing your training lab

Quarterly demand for training and capacity of 10 environments


The graph above depicts the challenge of meeting the variable demand for classes with a fixed capacity of 10 environments. It illustrates two separate issues that result from the misaligned demand and capacity:

  1. Demand exceeds capacity: during peak times (weeks 9 through 14) the training manager faces more students than those he or she can support with the existing training infrastructure. We must remember that especially in these times the training team is under a lot of pressure to get as many end users up to speed with the new release. The training manager has two options: he or she can try to schedule students for week 15 onwards when demand is slow. This delay is a business inhibitor - end users don’t learn how to use the new product, and so they delay upgrading their systems to the new version, which translates into delayed revenue. On top of that, there are also administrative costs to these reschedules, to creating more classes, etc. The other option is also less than ideal - the training manager can decide to take as many students as there is demand and have the students share infrastructure. This inevitably means compromising on the quality of the classes since students are required to share environments.
  2. Underutilizing infrastructure - demand is lower than capacity: In most other weeks, as the graph illustrates, the number of students who attend the training classes is smaller than the capacity for a class. In this case, there isn’t a direct implication on the class - it will go on as scheduled and each student will have a designated environment. The implication of these scenarios is a sunk cost in the underutilized training capacity in the datacenter.

The economics of running the training sessions in a hosted datacenter as described are summarized in the table below:

Month 1 2 3 4 Total
Demand 16 28 97 28 169
Capacity* 40 40 40 40 160
Cost $5,652 $5,652 $5,652 $5,652 $22,609
# Trained 16 27 40 18 101
Utilization** 40% 68% 100% 45% 63%
# Not trained 0 1 57 10 68

* Demand, capacity. # trained and # not trained are aggregated for the month, but numbers were matched on a week by week basis with the constraint of capacity of 10 per week (thus 40 per month).

** Utilization is defined as # trained/capacity

The public cloud is better suited to meet variable training classes demand

By using Ravello instead of creating training labs in the hosted datacenter, enterprises can increase revenue (and through that profit), as well as increase margins by reducing operational and administrative costs, thus again increasing profit. With Ravello the training manager is no longer constrained by fixed capacity - the public cloud model makes it so resources consumed match the size of the class. This has several financial implications:

  1. No idle capacity (no sunk cost)
  2. No delay in training end users which means no delay in the corporate customers’ updates to new versions. The only constraint on the number of students becomes the number of instructors.
  3. This also means reduced administrative costs

Another major benefit in using Ravello and the public cloud is that the instructor needs to provision only one environment for all the classes combined: Ravello allows the instructor to take a blueprint of the entire application environment once the application VMs are uploaded to Ravello. This is effectively a snapshot of the entire application. Once the blueprint is saved, all the instructor and students have to do is create an application from the blueprint - every student will get an environment with just one click - and there is absolutely no need to provision each and every environment from scratch - the application is provisioned once and deployed as many times as needed. This doesn’t only save hassle, but also precious time and money - essentially substantially reducing the operational cost of the classroom-based training.

In Ravello, the cost of one training environment of the application described, deployed in "cost optimized mode" will be $1.05/hour. To make the comparison easier, the annual cost of 10 environments (which is the capacity of the datacenter described), is $25,200 (as opposed to $67,828 in the datacenter, or a 62% cost reduction).
However, because Ravello allows you to run as many training environments as demanded, this comparison is less important. While the total cost of the training might increase, that is because the number of students will increase. This means that profits will go up (as revenue will go up) and additionally, the cost per student will be lower than that cost when running in the datacenter, as mentioned above.

The economics of running the training sessions in the public cloud using Ravello are summarized in the table below:

Month 1 2 3 4 Total
Demand 16 28 97 28 169
Capacity
Cost $3,360 $5,880 $20,370 $5,880 $35,490
# Trained 16 28 97 28 169
Utilization 100% 100% 100% 100% 100%
# Not trained 0 0 0 0 0

It is very important to note that aside from meeting the training demand in its entirety, as well as gaining full utilization of the resources employed, using Ravello lowers the cost per training environment (that is, the cost per student) while increasing the number of students trained (thus accelerating acquiring more end users for the new release).

Learn more about Ravello’s solution for virtual training labs in the cloud here, or click here to get started for free. 

Be the first to comment

Comments ( 0 )
Please enter your name.Please provide a valid email address.Please enter a comment.CAPTCHA challenge response provided was incorrect. Please try again.