Building the Perfect High Performance Network
The primary task left is hyper-scaling to multiple datacenters
By: Rick Parker
Sep. 29, 2011 08:15 AM
Three years ago I began as IT Director for Fetch Technologies, where we started working on a new mission: to create the perfect network. Fetch wanted a network that would be cost effective, reliable and scalable, with a simple web interface to create and manage virtual servers. The innovation achieved is something that I have carried with me to my new position as a Cloud Architect at Activision.
Fetch’s business, which enables organizations to access real-time data from websites across the Internet, is growing quickly due to the explosion of public content across the Internet, including prices, news, product information, blogs, events, and more. Because Fetch provides data via a hosted service to its customers it was critical that we be able to scale effectively while being sensitive to controlling cost. After much research and experimentation, the system we developed became known as the “private cloud” computing model. Although, the “public cloud” model has garnered the most attention, cost and control questions made the thought of a public cloud problematic for a small company. A private cloud model, alternatively, can be implemented by small companies for less cost than most people realize.
Fetch’s technology and business model required rapid, massive scalability and high performance. We achieved this is by keeping the data center size small, so as more capacity was needed, additional data centers could be deployed very quickly via a new concept of a redundant array of inexpensive data centers (or “cloud array”). Individual data centers can also be rapidly scaled up by adding CPUs and memory to individual servers or adding capacity to the storage system, which allows for dynamic increases of virtual server drives as needed.
The vision for the high performance private cloud was rooted in the idea that the infrastructure could provide a pool of resources, CPU storage, memory, etc., for whatever number or configuration of servers was needed.
Starting with the understanding of the benefits of server virtualization we decided to research if we could virtualize the entire infrastructure. Virtualization requires some level of scalability, so we started with the smallest chassis-based system, like a router for example, and then read and compared router vendor documentation for the most “cloud” like functionality, i.e. multi-tenancy and scalability; we did this for every component. We then started building the infrastructure and found for some of the systems we had to work with up to the highest levels of vendor support because the “cloud” features were used so rarely or rarely in combination with the other systems selected. It took research and work but the results were well worth it.
Fetch’s data-as-a-service business model requires combinations of high-capacity virtual servers and many lower capacity virtual servers. Think of virtual servers as highly configurable software blades (as opposed to hardware blades). While VMware supports up to 8 CPUs and 255Gb memory virtual servers, Fetch requires the highest capacity Intel-based servers, as many CPU cores per chassis as available, 48Ghz of CPU and more. Using a combination of VMware and Platform Computing’s ISF cloud management product, Fetch can manage maximum resource utilization of these large servers, providing maximum cost effectiveness.
The foundation of the Fetch Cloud is VMware virtual infrastructure, fibre channel storage area networks, and virtualized networking systems (i.e. load balancers, firewalls, etc.). Virtual infrastructure (VI) provides the functionality of high availability and automatic load balancing of virtual servers across physical hosts, turning groups of physical servers into resource pools of CPU and memory. Fetch uses Platform ISF to provide a simple web interface for rapid scalability, the ability to create and schedule many virtual servers with a few mouse clicks, and the capability to distribute server administration to many departments in the company. This provides much faster product development and deployment because the other departments are not constrained by IT staff capacity.
The high-performance private cloud model has been critically important to Fetch, providing a means to quickly, flexibly and cost-effectively scale their business as needed. Today, the network effort, initially prototyped to support a single project, includes all servers for corporate infrastructure and products. Fetch is very close to a perfect network, having developed a very cost effective, extremely reliable, scalable and user-friendly system – usually contradictory attributes. The primary task left is hyper-scaling to multiple datacenters, a Redundant Array of Inexpensive Datacenters--also known as “IT Nirvana.”
I’ve taken these lessons in innovation with me on my latest endeavor as Cloud Architect at Activision and look forward to reaching IT Nirvana one day soon.
Reader Feedback: Page 1 of 1
Latest Cloud Developer Stories
Subscribe to the World's Most Powerful Newsletters
Subscribe to Our Rss Feeds & Get Your SYS-CON News Live!
SYS-CON Featured Whitepapers
Most Read This Week