We have been watching many variants on efficient packaging of servers for highly scalable workloads for years, including blades, modular servers, and dense HPC rack offerings from multiple vendors, most of the highly effective, and all highly proprietary. With the advent of Facebook’s Open Compute Project, the table was set for a wave of standardised rack servers and the prospect of very cost-effective rack-scale deployments of very standardised servers.
But the IP for intelligently shared and managed power and cooling at a rack level needed a serious R&D effort that the OCP community, by and large, was unwilling to make. Into this opportunity stepped Intel, which has been quietly working on its internal Rack Scale Architecture (RSA) program for the last couple of years, and whose first product wave was officially outed recently as part of an announcement by Intel and Ericsson.
While not officially announcing Intel’s product nomenclature, Ericsson announced their “HDS 8000” based on Intel’s RSA, and Intel representatives then went on to explain the fundamental of RSA, including a view of the enhancements coming this year.
RSA is a combination of very standardised x86 servers, a specialised rack enclosure with shared Ethernet switching and power/cooling, and layers of firmware to accomplish a set of tasks common to managing a rack of servers, including:
- Asset discovery
- Switch setup and management
- Power and cooling management across the servers with the rack
- Server node management
- Storage management
- Boot management
An open API set to allow external management and orchestration products to control the RSA environment
“But wait”, those with a long memory will exclaim, “Didn’t Intel try this once before to a thundering lack of success with blades?” Yes, they did, but I believe that RSA is an entirely different proposition than Intel’s dismal attempt to license and manufacture IBM blades in the early 2000’s. RSA has much more value-added IP around deployment and management, and Intel appears to be serious about creating an ecosystem of hardware partners who will license the RSA IP and designs and deliver rack-scale and large aggregates to the increasingly resource-hungry world of enterprises and service providers who are facing an avalanche of mobile engagement and big-data-driven requirements.
The successful promotion of RSA to a larger community of vendors who can manufacture hardware efficiently but maybe do not have the deep IP of the Tier-1 vendors will eventually translate into more options for any organisation that needs to deploy servers and storage against a large and scaling workload, which describes a large slice of the enterprise and ISP/CSP/MSP market today.
RSA systems will deploy more rapidly and more predictably than standard hardware, and the integration of Intel’s management IP (Intel’s attempts to establish a strong position in higher layers of management IP have not been as successful as they wished, but have left them with a cupboard stuffed with great IP, some of which looks like it is being redeployed in RSA) will provide a valuable assist to second-tier vendors in their quest to acquire footprint in Tier-1 customers. Ericsson’s deployment of its HDS 8000 is essentially a blueprint for CSP deployments – high value services and cloud software layered over a cost effective modular infrastructure based on RSA.
This is not rocket science. The big potential losers are the Tier-1 vendors, many of whom will end up competing with their own ODM suppliers armed with Intel RSA for the most rapidly growing (by unit volume) segment of the market - CSP/MSP and internal cloud in large enterprises. A large chuck of legacy server business will be somewhat immune for a while, but if RSA is even moderately successful, further encroachment is almost guaranteed. Consider, hypothetically, RSA systems bundled with a popular hypervisor environment, which would suddenly pressure some very profitable franchises among CI suppliers. RSA is also a natural substrate for the emerging hyperconverged software, some of which is already uncoupled from its associated hardware. Could RSA be an EVO:Rack platform?
For potential large-scale users, Intel RSA is an unalloyed positive. Worst case it sends a real message to the Tier-1 vendors to get even more efficient in delivering larger compute aggregates to the market, and best case it introduces some real new competition and brings additional alternatives to the table.
On a Final Note, Who Needs It?
A decade ago, the idea of building a rack-scale product would have been questionable. Even the physically largest blade offerings were at best about a quarter of a rack, and they were typically sold partially full into legacy enterprise environments and expected to fill up over the lifespan of the enclosures. Rack scale deployments were the purview of large supercomputers and the occasional financial services derivative farm or large media rendering farms.
The explosion of mobile data, analytics and general web-scale activity changed all that, and now multiple players, from the largest web operators to niche CSPs and MSPs along with large enterprises are finding themselves becoming de-facto rack-scale deployment engineers. This is the vacuum that RSA can fill, particularly as all these emerging rack-scale consumers strive for Googlesque economies of scale at a much smaller size.
Posted by Richard Fichera