Economics of public and private clouds
2 likes1,117 views
The document discusses the economics of different cloud models, highlighting that public cloud may not always be the most cost-effective option without careful negotiation and long-term contracts. It provides a detailed cost comparison between AWS and a DIY approach, emphasizing the additional hidden costs of management and infrastructure. The author advocates for automation tools to streamline operations in open source cloud environments.
Economics of public and private clouds
- 1. Private, Public, Hybrid: The Real Economics of Open Source Clouds Carlo Daffara
- 2. “Public cloud is better from any economic point of view” (or so the experts say)
- 10. “Public cloud allows for rapid provisioning, saving money by using only the resources you need.”
- 11. “Public cloud allows for rapid provisioning, saving money by using only the resources you need.”
- 14. “Public cloud providers pass Moore’s law hardware savings to you, because the public cloud is a commodity.” Fact: not really unless you negotiate (a lot) and use long terms contracts.
- 15. “Public cloud has far greater utilization rates” Fact: Absolutely not. Reported rates for Google (that has a scale comparable to AWS) are not far from those reported for private clouds.
- 16. Source: NRDC How VMware Virtualization Right-sizes IT Infrastructure to Reduce Power Consumption; VMware, 2008AWS
- 17. What happens with slow-varying, 24/7 loads?
- 18. ● “You have 100 physical servers, 6 virtual instances on each, for a total of 600 virtual instances. Fewer would make the scale of running it prohibitively worse. ● Your 100 physical servers will have four switches, two routers and two firewalls. In reality, you probably will have much more, adding to the expense. ● We will ignore your monitoring and other required services for the moment. ● You are virtualizing with Xen, (which AWS uses), available for free, as opposed to VMWare, which has significant licensing costs. ● The number of instance system admins (or “sysadmins”) is the same, since you have the same 600 instances, so we can consider them a wash. ● Your AWS costs use on-demand pricing. You could save double digit percentages by paying for reserved instances.” 3yr Cost AWS: $1,088,640 3yr Cost DIY: $2,288,000 (includes personnel for hw management, networking, cabling) source: http://blog.atomicinc.com/2015/03/03/does-amazon-web-services-pricing-follow-moores-law-2350/
- 19. 3yr Cost AWS: $1,088,640 3yr Cost DIY: $2,288,000 (includes personnel for hw management, networking, cabling) Equivalent cost by HaaS: 30 MG128 (216$/month) → 3yr cost: $233280 (roughly ⅕ of AWS). Fully serviced, including hardware replacement, networking and management.
- 20. ● “You have 100 physical servers, 6 virtual instances on each, for a total of 600 virtual instances. Fewer would make the scale of running it prohibitively worse. ● Your 100 physical servers will have four switches, two routers and two firewalls. In reality, you probably will have much more, adding to the expense. ● We will ignore your monitoring and other required services for the moment. ● You are virtualizing with Xen, (which AWS uses), available for free, as opposed to VMWare, which has significant licensing costs. ● The number of instance system admins (or “sysadmins”) is the same, since you have the same 600 instances, so we can consider them a wash. ● Your AWS costs use on-demand pricing. You could save double digit percentages by paying for reserved instances.” 3yr Cost AWS: $1,088,640 3yr Cost DIY: $2,288,000 (includes personnel for hw management, networking, cabling) source: http://blog.atomicinc.com/2015/03/03/does-amazon-web-services-pricing-follow-moores-law-2350/
- 21. Real cost is management.
- 23. Real cost is trying to do too much.
- 24. Real cost is trying to do too much.
- 31. Medieval Portrait of an OpenStack sysadmin
- 34. ● Use automated deployment tools the moment you plan to go over 5 servers. ● Among the tools: ○ Foreman ○ Razor ○ JuJu ○ Loom ○ Cobbler ○ Stacki…. ● All very good - use the one you like most ● OpenNebula is exceptional at automation of the machinery. Sometimes you need additional automation (but not always) ● This can be done by external VMs, that do have a privileged view of what’s inside ● Invest a little time in additional “helpers”: DCIM/CMDBs, NAC, Gateway VMs...