This week we’re announcing new systems at the upper end of the UCS server product line: some heavy-duty iron for heavy-duty times. These are important new tools for our UCS customers: the digital age is accelerating, IT needs more horsepower to keep up, and there is a lot at stake.
Consider this: less than 10 years ago, some of the largest mainframes scaled up to half a terabyte (TB) of main memory. What if I were to tell you that these latest generation UCS blade servers will scale to 3TB? Sound like a lot? It is. And that’s just the two-processor version. Connect two UCS B260 M4 blades with an expansion connector and they become a UCS B460 M4, a four socket server that will scale to 6TB. Putting that into perspective: a spiffy new laptop might ship today with 8GB of memory. Multiply that by 750 and you have 6TB.
Not too long ago, all the content Wikipedia would fit in this type of footprint (in 2010 it was just under 6TB with media.) Here is a fun illustration of what this scale of data would look like on paper (just the ~10GB of text, not the images.) Now remember, we’re not talking about fitting all that data on the local disks of the server – we’re talking about fitting it in main memory. This is becoming crucially important in the field of data analytics, where “in-memory” is the key to speed and competitiveness. Applications like SAP HANA are at the forefront of this trend. Today, at Intel’s launch event in San Francisco, Dan Morales (Vice President of Enabling Functions at eBay) joined us to talk about how they’re betting on this type of analytic technology to help them make the eBay Marketplace work better for buyers and sellers (and eBay shareholders.) I’ll post a video clip of that soon; his description of the challenges and opportunities, at eBay scale, is worth a watch.
We’ve talked about memory scaling, and Bruno Messina has a nice post that talks more about the scalability on these systems and UCS at large. But dominating performance is the name of the game: behemoth processing performance is what we look for at this end of the server spectrum and Intel has not disappointed on this round of new technology. The next generation of the Intel Xeon E7 family packs up to 15 cores per processor and delivers an average 2x performance increase compared to previous generation products. Performance will be even higher on specific workloads, for example up to 3X on database and even higher for virtualization. Cisco’s implementation of this technology has once again set the standard for system performance. In today’s launch, Intel cited Cisco with 6 industry-leading results on key workloads. As of this posting, the closest to come to that achievement that was Dell with 4. HP ProLiant posted 1. So hats off, once again, to the engineering team in Cisco’s Computing Systems Product Group. Girish Kulkarni has a great summary of the performance news here.
Our collaboration with Intel is one of the best technology combinations in the industry today. Consider what we both bring to the party. Intel: innovation in processor technology that drives Moore’s Law. Cisco: innovation in connecting things across the data center and around the world. UCS is an outcome of two blue-chip tech powerhouses investing in real innovation and the results have changed the industry.
In 1991, Stewart Alsop famously wrote: “I predict that the last mainframe will be unplugged on 15 March 1996.” He just as famously had to eat his words. He munched on those twelve years ago, and while mainframes and RISC-based systems remain, there is an inexorable trend as the heaviest analytic workloads continue to shift to the type of scale-up x86-based systems we’re talking about today. It only makes sense. So while this will garner me plenty of comments from the architectural purists out there, I say “go ahead and plug a mainframe back in.” It will fit right in your UCS B-Series blade chassis…