Serkan Sahin is a Chief Architect for IBM Strategic Outsourcing Service Delivery in the Middle East and Africa. He has more than 18 years of professional experience in the IT Industry. He is an experienced Architect, System Engineer and IT Consultant who has been developing multi-component wall to wall complex IT infrastructure solutions. He has worked for IBM since 1999. He has been an IBM and Open Group Certified Architect as a Technology Architect since 2008. He is a certified IBM instructor for Architectural Thinking, Architecting for Performance Engineering and Technical Leadership College professional for internal classes in IBM for professional give back. He has both a computer science and industrial electronics degree.
It is very enjoyable to talk about System z and Cloud because System z itself was made for cloud 50 years ago. The name “Cloud” just wasn't invented until recent years. The System z machine itself is Cloud. Today System z is an important component for any company that wants to build an enterprise level cloud solution.
The IBM Enterprise Cloud System includes software, storage and server technologies in one simple, flexible, and secure factory integrated solution and there is no similar solution on the market. This helps IT organizations deliver cloud services with the ability to rapidly deploy a trusted, scaled OpenStack based Linux cloud environment with System z quality of services and you have the ability from the start to scale up to 6,000 virtual machines with 3TB of memory in a single footprint whether it is a public, private or hybrid cloud solution. Nowadays, everybody focuses on how a cloud solution can manage well and that means focus is actually on “Cloud Orchestration”. (Cloud orchestration is all about managing the interconnections and interactions among cloud-based and on-premises business units.)
- IBM System z provides dynamic provisioning and cloud orchestration for business critical workloads with added monitoring, performance and data backup, full virtualization and Cloud Management software with IBM Cloud Management Suite for System z.
- Additionally IBM z/VM is an industry leading and proven hypervisor and is the heart of the Enterprise Cloud Solution. z/VM provides impressive horizontal and vertical scalability, rapid server provisioning, rock solid workload isolation, and the ability to virtualize key system resource management
- IBM Wave for z/VM provides an intuitive virtualization management platform for managing the Enterprise Cloud Systems.
The management software layer (Cloud Orchestration) is the only visible part of the iceberg (the Cloud solution) cost structure as shown in the figure below. I know discussion is necessary and I do not say not required but I say it does not drive your main cost structure. Your cost is still hidden under the invisible iceberg part beneath your computing power infrastructure architecture.
What does it mean when we say that System z can handle 6,000 virtual machine workloads and 3TB memory requirement in a “single footprint”? It means you can “reduce your hidden costs” dramatically.
I can do just a very simple calculation here with a basic assumption: If you require 500 servers on your cloud system, and each virtual server will require at least 2 Virtual CPU (vCPU) and each server will need 6GB RAM, your requirement snapshot basically is:
- 1000 vCPU
- 3 TB RAM
If you try to size the same requirement in an x86 environment, lets see what happens: x86 baremetal hypervisor on a single x86 machine can give a maximum 64vCPU today with 1TB RAM on a single server. When you do the simple calculation again you find that you will need:
- 16 x86 servers
It looks very easy. But wait - before coming to any conclusion, see what the real result is when we consider a real architecture for a virtualization solution.
Your first barrier on x86 virtualization is “Constaint”. Ask yourself this before starting your solution: What is my limitation with x86 virtualization? First you need to consider best practices because this platform does not give the freedom of capacity planning like System z. You cannot overcommit your capacity and utilize 100%. You must understand best performance criteria.
For example, if your workload requires more CPU power (more than 6 vCPU) best practices dictates that you will not find this power on x86 virtualization because of performance and scalability limitations which should be a very important consideration in your solution. Ooops! Do I need to learn more? YES…
- If you require more CPU power, first you need to have a reserve of hardware for a dedicated virtual image or you need to consider a physical alternative to be set aside for your workload.
"Come on", you may say," how much hidden cost can there be in this infrastructure solution? It is really cheap when I want to buy a single x86 server." You’re right when you consider only a single machine but most likely, your infrastructure will grow incrementally ad-hoc . Let's change our calculation to be a little bit more complex and see what happens:
For reducing hardware, the first investment cost I choose is a “2 x socket” 2u x86 platform with 2.9GHz maximum core speed and 8 cores. Now your new calculation will be:
- 2.9 Ghz = 1vCPU
- 2 socket x 8 core = 32 core for each server
- 32 core x 2.9 GHz = 92 GHz cpu power
- 92 GHz / 2vPU= 46 vCPU can position each server (without any redundancy and overcommit calculation)
- Result is now 22 servers required …
Assume power consumption to be 900watts for each server with a single power supply and position redundancy. I now require two power supplies for each server and I need to take into consideration a 1.8 Kwh power load for each server. Another assumption based on my previous experience, this server can consume 10,000 Kwh/year power under full load which means 1.14 Kwh energy spend for each hour. And again I assume server is 30Kg with full hardware capacity futures. Let’s say the Data center has sqm/kg= 600 kg, 1 rack = 2 square meters and watt/rack= 15000 watts.
Your new calculation will be:
- 15 Kwh each rack / 1.8 Kwh = 8 server/rack
- 8 server x 1.14 Kwh = 9.12 Kwh per rack
- You need 3-rack space (6 square meter)
- Hourly you will spend 27.36 Kwh power (without cooling electricity consumption)
- Example = 0.25-cent Kwh means you need $6.80 per hour for only the server’s electricity. Your yearly electricity cost will be $60,000 USD for only 22 servers!
Why did I do this calculation?
A single system zEC12 is about 5.5 Kwh electric spent under the same load and only a 3.16 square meters footprint is needed. Which means your footprint cost is almost half and your electricity cost is 5 times less per year for only this basic example! In real life, it is more complex than this example and now I suggest that you think again about your Cloud Solution.