DreamHost is a leading Web hosting and cloud services provider with over 375,000 customers worldwide that hosts over 1.3 million blogs, websites, and applications. The company offers a wide spectrum of Web hosting and cloud services, including Shared Hosting, Virtual Private Servers (VPS), Dedicated Server Hosting, Domain Name Registration, the cloud storage service DreamObjects, the cloud computing service DreamCompute, and the managed WordPress service DreamPress.
DreamHost sets itself apart from the cloud crowd by developing innovative and flexible web and cloud services using open source software. And the NVIDIA® Cumulus® open approach makes complete sense in this environment. DreamHost’s NVIDIA Cumulus Linux® deployments coincided with the launch of public cloud services. DreamCompute is engineered for scale and efficiency using best-of-breed open source solutions, including the OpenStack cloud platform, scalable Ceph™ block storage and NVIDIA Cumulus Linux, and VMware NSX’s network virtualization. DreamHost is well known for constantly striving to lower cost and deliver the best services by contributing to and leveraging world-class open source software.
For DreamHost, the need for speed and transparency across the data center was imperative. Equally as important was the need to successfully employ the same self-managing model — used in the server world for years — across the company’s networking operation. This meant that when DreamHost received a shipment of servers, they get installed and configured automatically and are never touched after that point except for minor automated updates/changes. Their objective was to apply a similar methodology to the network as well. Ideally, Network Architects would focus on architecting the network and bringing differentiated value to the network and a single Tech Ops team could manage the combined technical operations, resulting in significant savings in OpEx. This is what DreamHost strived for and what NVIDIA Cumulus helped them achieve:
- Performance and transparency in a single, easy-to-deploy,
- Ability to leverage existing investment in tools for network automation without reinventing the wheel — or in this case — the network!
- Incorporate a self-managing model proven to be effective in the server world across the network.
DreamHost’s environment consists of customer points of delivery (PODs) and command & control PODs for orchestration/management, running NVIDIA Cumulus Linux in a leaf/spine architecture, interconnected at 40G. Alternative platforms were considered in the new architecture, but NVIDIA Cumulus Linux provided far better performance and transparency at the right cost.
NVIDIA Cumulus opened the ecosystem to include a variety of networking gear, an open Linux operating system, and an open ecosystem of applications, thus lowering CAPEX and OPEX. In particular:
- Hardware cost roughly 1/3 of the incumbent vendors price
- Choice of optics, with no vendor lock-in, cost about 1/10th of the incumbent vendors cost. One license: Base license included everything. There were no extra licenses for routing, VXLAN, etc. which simplified license management
- Streamlined workflow: Configuration with automation shortened deployment and troubleshooting time
Traditional networking vendors tend to be complex. They have their own hardware and operating system, their own command line interface (CLI), their own API, and their own processes to integrate new functionality. Instead, Linux as an operating system the DreamHost team was familiar with, and that provides visibility, programmability, and a well known API. NVIDIA Cumulus Linux provided the transparency of Linux for networking gear.
Another important factor in deciding on NVIDIA Cumulus Linux was the use of familiar tools. Orchestration tools, in particular, was very important. DreamHost has been using Opscode’s Chef for server orchestration; using the same tools for automation of the network with just small changes to support Chef recipes saves a lot of time and manual operations.
DreamHost previously used the Quagga routing suite for DDoS mitigation, so it was easy to repurpose familiar technologies for switches. They used collectd for server monitoring that could be leveraged in network as well.
The IT team is knowledgeable about Linux, so they were able to leverage their in-house developers’ expertise to extend the system while utilizing existing NVIDIA Cumulus Linux tools such as PTMd to verify proper cabling operations against a topology graph.
Devices are bound to fail, and the best highly available designs work around failures. This means using resilient leaf/spine architectures with redundant fixed boxes, simplified configurations to remove all of the complicated HA protocols from the network, and having a simple automated solution with all familiar tool chains.
For DreamHost, the expectations around speed and transparency were met and exceeded with bottom line benefits. The company was able to successfully replicate the proven self-managing model of the compute side of their business and scale without the burden, weight and expense of vendor lock-in. DreamHost continues to leverage the support of NVIDIA Cumulus Linux operating system to expand its networking operation in the most effective and efficient way possible.