Total Cost of Ownership: AWS TCO vs OpenStack TCO Q&A
- October 17, 2016
- TCO
Last month, Amar Kapadia led a lively discussion about the Total Cost of Ownership of OpenStack clouds versus running infrastructure on Amazon Web Services. Here are some of the questions we got from the audience, along with the answers.
Q: Which AWS cost model do you use? Reserved? As you go? A: Both. We have a field that can say what % are reserved, and what discount you are getting on reserved instances. For the webinar, we assumed 30% reserved instances at 32% discount. The rest are pay-as-you-go.
Q: How does this comparison look when considering VMware's newly announced support for OpenStack? Is that OpenStack support with VMware only with regards to supporting OpenStack in a "Hybrid Cloud" model? Please touch on this additional comparison. Thanks. A: In general, a VMware Integrated OpenStack (VIO) comparison will look very different (and show a much higher cost) because they support only vSphere.
Q: Can Opex be detailed as per the needs of the customer? For example, if he doesn't want an IT/Ops team and datacenter fees included as the customer would provide their own? A: Yes, please contact us if you would like to customize the calculator for your needs.
Q: Do you have any data on how Opex changes with the scale of the system? A: It scales linearly. Most of the Opex costs are variable costs that grow with scale.
Q: What parameters were defined for this comparison, and were the results validated by any third party, or on just user/orgnaisatuon experience? A: Parameters are in the slide. Since there is so much variability in customers' environments, we don't think a formal third party validation makes sense. So the validation is really through 5-10 customers.
Q: How realistic is it to estimate IT costs? Size of company, size of deployment, existing IT staff (both firing and hiring), each of these will have an impact on the cost for IT/OPs teams. A: The calculator assumes a net new IT/OPS team. It's not linked to the company size, but rather the OpenStack cloud size. We assume a minimum team size of about 3.5 people and linear growth after that as your cloud scales.
Q: Should the Sparing not be adding more into the cost, as you will need more hardware for HA for high availability? A: Yes, sparing is included.
Q: AWS recommends using 90% utilization, and if you are using 60%, it's better to downgrade the VM to ensure 90% utilization. In the case of provisioning 2500 vms with autoscaling, this should help. A: Great point, however, we see a large number of customers who do not do this, or do not even know what percentage of their VMs are underutilized. Some customers even have zombie VMs that are not used at all, but they are still paying for them.
Q: With the hypothesis that all applications can be "containerized", will the comparison outcomes remain the same? A: We don't have this yet, but a private cloud will turn out to have a much better TCO. The reason is that we believe private clouds can run containers on bare-metal while public clouds have to run containers in VMs for security reasons. So a private cloud will be a lot more efficient.
Q: This is interesting. Can you please add replication cost? This is what AWS does free of cost within an availability zone. In the case of OpenStack, we need to take care of replication. A: I assume you mean for storage. Yes we already include a 3x factor to convert from raw storage to usable storage to factor in replication (3-way).
Q: Just wondering how secure is the solution as you have mentioned for a credit card company? AWS is PCI DSS certified. A: Yes this solution is PCI certified.
Q: Has this TCO calculator been validated against a real customer workload? A: Yes, 5-10 customers have validated this calculator.
Q: Do you think that these costs apply to another countries, or this is US based? A: These calculations are US based. Both AWS and private cloud costs could go up internationally.
Q: Hi, thank you for your time in this webinar. How many servers (computes, controllers, storage servers) are you using, and which model do you use for your calculations ? Thanks. A: The node count is variable. For this webinar, we assumed 54 compute nodes, 6 controllers, and 1080GB of block storage. We assumed commodity Intel and SuperMicro hardware with 3 year warranty.
Q: Can we compare different models, such as AWS vs VMware private cloud/public cloud with another vendor (not AWS)? A: These require customizations. Please contact us.
Q: Which AWS cost model do you use? Reserved? As you go? A: Both. We have a field that can say what % are reserved, and what discount you are getting on reserved instances. For the webinar, we assumed 30% reserved instances at 32% discount. The rest are pay-as-you-go.
Q: How does this comparison look when considering VMware's newly announced support for OpenStack? Is that OpenStack support with VMware only with regards to supporting OpenStack in a "Hybrid Cloud" model? Please touch on this additional comparison. Thanks. A: In general, a VMware Integrated OpenStack (VIO) comparison will look very different (and show a much higher cost) because they support only vSphere.
Q: Can Opex be detailed as per the needs of the customer? For example, if he doesn't want an IT/Ops team and datacenter fees included as the customer would provide their own? A: Yes, please contact us if you would like to customize the calculator for your needs.
Q: Do you have any data on how Opex changes with the scale of the system? A: It scales linearly. Most of the Opex costs are variable costs that grow with scale.
Q: What parameters were defined for this comparison, and were the results validated by any third party, or on just user/orgnaisatuon experience? A: Parameters are in the slide. Since there is so much variability in customers' environments, we don't think a formal third party validation makes sense. So the validation is really through 5-10 customers.
Q: How realistic is it to estimate IT costs? Size of company, size of deployment, existing IT staff (both firing and hiring), each of these will have an impact on the cost for IT/OPs teams. A: The calculator assumes a net new IT/OPS team. It's not linked to the company size, but rather the OpenStack cloud size. We assume a minimum team size of about 3.5 people and linear growth after that as your cloud scales.
Q: Should the Sparing not be adding more into the cost, as you will need more hardware for HA for high availability? A: Yes, sparing is included.
Q: AWS recommends using 90% utilization, and if you are using 60%, it's better to downgrade the VM to ensure 90% utilization. In the case of provisioning 2500 vms with autoscaling, this should help. A: Great point, however, we see a large number of customers who do not do this, or do not even know what percentage of their VMs are underutilized. Some customers even have zombie VMs that are not used at all, but they are still paying for them.
Q: With the hypothesis that all applications can be "containerized", will the comparison outcomes remain the same? A: We don't have this yet, but a private cloud will turn out to have a much better TCO. The reason is that we believe private clouds can run containers on bare-metal while public clouds have to run containers in VMs for security reasons. So a private cloud will be a lot more efficient.
Q: This is interesting. Can you please add replication cost? This is what AWS does free of cost within an availability zone. In the case of OpenStack, we need to take care of replication. A: I assume you mean for storage. Yes we already include a 3x factor to convert from raw storage to usable storage to factor in replication (3-way).
Q: Just wondering how secure is the solution as you have mentioned for a credit card company? AWS is PCI DSS certified. A: Yes this solution is PCI certified.
Q: Has this TCO calculator been validated against a real customer workload? A: Yes, 5-10 customers have validated this calculator.
Q: Do you think that these costs apply to another countries, or this is US based? A: These calculations are US based. Both AWS and private cloud costs could go up internationally.
Q: Hi, thank you for your time in this webinar. How many servers (computes, controllers, storage servers) are you using, and which model do you use for your calculations ? Thanks. A: The node count is variable. For this webinar, we assumed 54 compute nodes, 6 controllers, and 1080GB of block storage. We assumed commodity Intel and SuperMicro hardware with 3 year warranty.
Q: Can we compare different models, such as AWS vs VMware private cloud/public cloud with another vendor (not AWS)? A: These require customizations. Please contact us.
