The Open Compute Project (OCP) Summit (March 10-11, 2015) that took place in Santa Clara, CA showed that substantial progress is being made in the open-source hardware movement around OCP specifications, which began with the Summit in 2011.
What began as a consortium to reduce costs through the use of industry-standard interconnects and rack specifications has resulted in a wave of products that are designed to allow customers to take a plug-and-play approach to building out data center infrastructure. The aim of these designs is to create a situation in which data center operators will gain more choice, and more price-competitive products, by standardizing on OCP specifications, which are certified by OCP Lab tests.
This year, Neuralytix observed that more emphasis is being placed on bringing OCP-compliant systems to market. Vendors had a lot to say this time about why they have invested in building products around “open-hardware” specifications that can be easily adopted by other, competing vendors. Their central theme was customer-centricity. They’re betting that they will gain traction in the marketplace by making new products easier to consume, thereby making IT acquisitions more attractive.
We expect data center operators will gain more choice, and more price competitive designs, by standardizing on OCP specifications, which are certified by OCP Lab tests. Three main categories of customers are in sight for these vendors: cloud service providers; co-location sites or hosters; and enterprise customers making the leap to web-scale, scale-out infrastructure.
What’s Driving the Train for IT Consumption
Increasingly, it is the large IT organizations, and the service providers (including cloud service providers and telecoms companies) that are the primary consumers of IT products and services. Smaller firms often ask third-party providers to host their workloads, in an effort to reduce capex, and to contain operational costs (opex).
Case in point: Facebook, one of the OCP founding companies, which said it has already saved more than $2 billion in hardware acquisition costs by leveraging OCP technologies, over the last three years. Facebook said it is contributing two new technologies to OCP: Yosemite, a chassis that Facebook co-designed with Intel that holds up to four system-on-a-chip (SoC) processor cards, un a system that consumes less than 400 watts. Importantly, Facebook also announced a specification for a top-of-rack switch code-named Wedge, along with low-level board management software called OpenBMC for faster custom firmware development. Because Wedge is designed to be compatible with Open Rack OCP standard, it will hold up to 192 SoC server cards in a single rack.
A New Way to Shop for IT Hardware
IT flexibility is one key benefit cited by the large customers, such as Fidelity Investments, JP Morgan, Bloomberg, Capital One, and Goldman Sachs (an OCP co-founder), who have found open-source hardware is a new way to improve competition in the vendor world.
Speaking on a financial-services customer panel on Tuesday, March 10, IT executives from these firms said they are shopping for vendor products that fit into a scale-out infrastructure of servers, storage, networking switches and optimized special-purpose appliances. (See the video, here). Now, they see a lot of “price transparency” in working with open hardware and open-source software, which helps them plan their data center build-outs more effectively.
Here’s some background that Neuralytix notes, to place this in perspective. Historically, that the flavor of cost-competitive market dynamics was fueled by competition between multiple vendors in the IBM-compatible mainframe marketplace in the 1970s and 1980s, or in the Unix/RISC server marketplace in the 1990s and early 2000s. But as the world has moved to a scale-out architecture hat is mostly built on x86 processor architecture, the reduction in vendor-proprietary interfaces is making it easier to add capacity without doing a “forklift upgrade” or a “rip-and-replace” migration.
This pattern, combined with heavy virtualization of the platform, will enable the next-generation of software-defined infrastructure. With the underlying hardware no longer largely a collection of separate “silos” of architecture, customers will manage workloads (applications and databases) via software that automates provisioning and orchestrates where the workload is placed for processing.
In this world, a high degree of abstraction “lifts” the software workload above the hardware layers — allowing workloads to “flow” from hardware platform to hardware platform — and moving them to available hardware resources, as needed.
Product Announcements at OCP
Building on OCP technologies and specifications hammered out in recent years, many vendors chose this OCP Summit to announce products. Details are included here. Highlights of these announcements, among others, included the following (presented in alphabetical order):
- Broadcom announced its Broadcom Open Network Switch Library (OpenNSL), OpenNSL is a software interface with a set of APIs that enable the development of new applications on top of Broadcom StrataXGS switches.
- HP announced its new CloudLine product family of three OCP-based server designs optimized for cloud computing. HP announced that it is partnering with Foxconn, the large system-builder in China, to produce these systems on a new, and lower cost-basis. The CloudLine servers come in up to 5models, ranging from 1u servers to 2u servers—and a new narrow form-factor that allows customers to fit three servers side-by-side in a 21’’ rack. HP expects these systems to be adopted by cloud service providers and companies building out cloud-centric infrastructure.
- Hyve Solutions announced the Open Rack v2, Leopard, Honey Badger and Wedge products based on OCP specifications. Hyve also announced a partnership with Cavium, Inc. to bring 64-bit ARM-based volume server solutions to market; these solutions will be based on a dual-socket ARMv8 Open Compute processor.
- Intel showed several designs based on its Intel Xeon processor D-1500 SoC (system-on-a-chip) product family. Specifically, Intel worked with Facebook for 18 months to produce the “Yosemite” system, leveraging the Mono Lake server card with D-1500 processors.
- Mellanox showed its Mellanox Multi-Host, which links up to four system elements to a single, high-speed interconnect, supporting up to 100 Gbps of data transfer. The Multi-Host product will be leveraged for scalable Cloud, Web 2.0, and high-performance computing (HPC) workloads.
- Quanta showed new versions of its Rackgo X and Rackgo M rack systems, which support the latest generation of Intel Xeon E5-2600 v3 processors. The Rackgo X F06A was designed for compute density; the Rackgo XF06D is a converged infrastructure solution combining server and storage tiers; and the Rackgo M is a rack solution based on the Open CloudServer (OCS) version 2 standard, which Microsoft contributed to OCP last year.
Where Does It Go From Here?
Clearly, Neuralytix notes that the open-hardware ecosystem is expanding. Just last year, Microsoft and IBM joined OCP, and HP joined this year. Many ODMs have joined OCP in recent years (e.g. Quanta, Wiwynn), as well, realizing that open-hardware specs open up new sales opportunities to mid-sized and smaller vendors – as long as the products they make can fit into the data center infrastructure seamlessly.
We believe that the big themes of OCP certainly resonate with long-held IT “best practices.” Make sure there are enough choices in the marketplace to drive down costs in competitive bidding situations. Reduce the barriers to entry in the market to ensure greater competition. And improve interoperability, reducing IT’s migration, management, maintenance and service costs over time. As long as OCP keeps that “gestalt” of IT as its driving force, its future will be bright, fueled by customer upgrade, technology refresh and replacement cycles.
Neuralytix believes that cloud data centers and large financial services companies – both having data centers with thousands of servers under management – started leveraging OCP-certified hardware over the last two years, but those customers, all with large IT staffs, will need to be closely involved with end-to-end management of those cloud servers. However, it’s only a matter of time until the same style of computing becomes more strongly established in enterprise data centers, as they move more workloads onto low-cost cloud-style hardware in scale-out infrastructure.