BETA
This is a BETA experience. You may opt-out by clicking here

More From Forbes

Edit Story

IBM's NeXtScale: Early Thoughts On The New Scale-Out System (Updated)

Following
This article is more than 10 years old.

This week, IBM announced their NeXtScale System platform aimed at scale-out datacenters.  While the classic enterprise market is down, the scale-out datacenter market serving web giants like Google, Facebook, MSN, and Baidu is booming based on the benefits from cloud economics. Web giants are a real bright spot at least for unit volumes in the tech sector, but profitability for suppliers is another matter, as the web giants squeeze their suppliers hard.

NeXtScale is a follow-on to IBM’s iDataPlex and competes with Dell's DCS, and Hewlett-Packard's recent foray with Moonshot. While NeXtScale just launched, I'd like to provide some early thoughts on the platform. Let me start by providing a bit of market background.

At the peak of the modular x86 rack mounted IT server market just a few years ago, HP, Dell and IBM together sold roughly 3 out of every 4 new servers shipped into the market.  While HP and Dell have historically dominated volume x86 server shipments, IBM maintained a healthy business at the high end of the market with about half of the unit shipment share of the market leader. IDC publicly reported for 2Q/2013 that the “big three” share of server system shipments has declined over the last 12 months from 72% to 68% – a 4-point drop over only four quarters.   Dell has just about pulled even with HP for new server shipments (both have converged on ~30% share, HP from above and Dell from below), and IBM has fallen to 11% share.

What’s driving this trend?  In a word, scale.  Or more appropriately, hyper-scale. Moor Insights & Strategy (the firm I work for) defines scale-out datacenters as those for the web giants... ie Google, Facebook, Amazon, eBay, etc.  These datacenters are football-field sized buildings using modified versions of Linux and all open source or proprietary applications and middleware on top. Search, advertising, streaming media and caching, software-as-a-service (SaaS), social media, e-commerce, and cloud services giants now buy at such scale that it is becoming obvious that we are in the middle of the next generational shift in IT infrastructure.  Think Mainframe to Mini to RISC to Modular x86 to Hyperscale.

Dell recognized the web-giant trend and formed their Data Center Solutions (DCS) group with the goal of selling to the Web giants that were starting to purchase server infrastructure directly from Asian ODMs like Quanta.  Dell’s response has been relentlessly pragmatic – drive costs out of rack-level infrastructure, change architecture incrementally and only when there is a measureable advantage, and continue to build a solid IT quality product.  They are maintaining a cohesive technology strategy where hyperscale drives innovation and that innovation then migrates into traditional enterprise IT products.  Dell’s steady market growth mirrors their “hyperscale-first” technology approach and investment.  Dell is potentially exposed if a competitor creates advantage and differentiation with a new architectural approach.

HP took a different direction than Dell for hyperscale server hardware innovation and threw incrementalism out the door – they named their project “Moonshot” to acknowledge the big bet they placed in developing a differentiated rack-scale architecture.  Moonshot was launched in 2Q/2013 and so won’t really impact HP’s overall market standing for another few quarters, but while it was in development HP’s hyperscale server sales appeared to have suffered.  I expect that HP’s strategy here is that Moonshot Server’s architectural differentiation will enable them to maintain differentiated margins while they take back shipment share.  Looking forward, HP’s challenge is to translate their big bet on a new architecture into renewed hyperscale market share momentum.

IBM has continued to invest in their POWER-based systems at the high end of the market, successfully drove iDataPlex into HPC (50 of the top 500 are iDataPlex), but didn't find big success in the web-giant hyperscale servers.  With NeXtScale, IBM wants to change this.  From the information available, it is difficult to find significant architectural differentiation between IBM’s NeXtScale n1200 and other mainstream “cloud”, “microserver” or hyperscale rack-level architecture.  I’m not talking about fine-grain differences in processor choice, power supplies, fans, PCIe options, etc.  Those are implementation details and not architectural features.  At a basic level, it looks like IBM with NeXtScale is migrating to more of a pragmatic Dell approach than an HP approach with a different architecture.

Web-giant hyperscale customers aren't brand conscious, so IBM will need to amp up that technological innovation and find a way to leverage the billions of dollars invested in Linux. To attain their 50% market share in this space,  Dell DCS took years of customer interaction and incremental system R&D to differentiate itself from lowest cost solutions.

Hyperscale buyers are uniformly invested in open source software, outside of Microsoft and a couple of others who have designed their own proprietary cloud systems. IBM has invested billions of dollars into Linux and is a top 10 contributor to the Linux kernel.   IBM has also supported some open source software development, primarily Apache ’s Geronimo project (an open Java Platform, Enterprise Edition distribution) and Eclipse (as the basis for IBM’s Rational and Jazz products).  IBM’s open source page says that it also contributes to Apache’s Derby (Java relational database), Lucene (Java text search library), and Tuscany (service oriented architecture for Java) projects, as well as OpenAjax.

The short version of IBM’s legacy open source investment is that it is aimed at traditional enterprise IT.  IBM completed their acquisition of Softlayer in July, which gives IBM access to an integrated cloud software stack that appears like it has supplanted IBM’s internally developed SmartCloud.  But given how recent this acquisition is and that IBM announced NeXtScale yesterday, this is just the beginning of their integration cycle.  IBM is also a member of ONF (OpenFlow SDN) and their Programmable Network Controller is based on OpenFlow, but that also is a fairly new product.

IBM’s challenge will be to connect hardware and software silos to create an integrated hyperscale product offering that can compete on one side with Dell DCS’s intense focus on practical innovation and on the other side with HP’s Moonshot deep R&D-based innovation.  It is difficult now to see the IBM architectural differentiation or value-added yet in this very hardware-centric NeXtScale roll-out, but then again, maybe IBM is holding some of that information close to the vest.

There is no question that IBM has been successful in the HPC market with iDataPlex, but the web giant market has a different set of dynamics and architectures.  With NeXtScale, I expect IBM to continue their success in HPC, but the web giants have had a very different set of purchase criteria. IBM's challenge will be to leverage their experience and IP in the HPC market as well as their experience in Linux to win over the web giants.   Interestingly, in the future, we could see HPC technologies playing a bigger role in the cloud, but I will save that for another column.

You can find me on the webTwitterLinkedIn and Google+.

Disclosure: My firm, Moor Insights & Strategy, like all research and analyst firms, provides research, analysis, advising, and/or consulting to many high-tech companies,  including Dell, Hewlett-Packard, and Intel cited in this article and other scale-out technology providers like AMD SeaMicro, Applied Micro, and Calxeda.  No employees at the firm hold any equity positions with any companies cited in this column.