Archive for the ‘ embedded ’ Category

Linaro – Open Source Glue

Last week ARM Ltd. and its licensees Freescale, Samsung, ST-Ericsson and TI, along with IBM, launched Linaro, a new organization to “foster innovation in the Linux® community through a common foundation of tools and software”.

My first reaction to the Linaro announcement was “O Joy, another Linux knitting circle”.  But I am happy to say that Linaro appears to be what the industry really needs – the glue between silicon and software (my apologies to fans of Linaro stallions). Instead of creating standards or aggregating yet another embedded Linux distribution, Linaro has the stated goal of enabling existing (and new) software to run on actual silicon in the marketplace.

Deliverables of such an effort include

  • device drivers
  • board support packages (configurations)
  • Linux kernel patches
  • tools to support integration of these and other contributions

Let’s take a stroll down .org memory lane to compare Linaro’s goals to the aspirations and accomplishments of other initiatives, past and present:

Embedded/Mobile Linux .org Roll Call

OSDL Mobile Linux InitiativeMLI put together requirements for a mobile Linux-based platform as a de facto soft standard (as OSDL did with Carrier Grade Linux).  Unlike CGL, MLI members were disappointed by the lack of actual software deliverables (that is, participants played chicken with contributions).  MLI did serve to help popularize Linux as a foundation of mobile telephony.

CELF – the Consumer Electronics Linux Forum worked to create standards for Linux in a range of consumer electronics, including mobile telephony; these efforts were very skewed toward particular member implementations and did not survive industry scrutiny.  CELF also instigated real  implementation by funding kernel contributions by maintainers (e.g., for flash and power management) and today survives as a sponsor of Embedded Linux Conferences.

LiPS – the Linux Phone Standards Forum had the explicit goal of creating Linux-based standards for mobile terminal devices.  Led by FT/Orange, ACCESS and VirtualLogix, they published several generations of specifications and were in the process of launching an open source TAPI project when the organization was absorbed by LiMo in 2008.

LiMo – the LiMo Foundation strives create the “first truly open, hardware-independent, Linux-based operating system for mobile device”, realized as a distribution shared by its members and deployed in member-built handsets.  Despite these lofty goals, LiMo is hampered by a highly stratified and expensive membership structure, tortuous IPR with limited out-licensing,  slow-to-market specifications, and most importantly incomplete validation suites and an MIA SDK.  While LiMo claims dozens of phones as compliant, the basis is a very rudimentary specification, with little or no visibility to applications developers (cp. Android).

Linux Foundation / MeeGo – In 2007, the Linux Foundation was born out of the merger of Free Standards Group (home of the Linux Standards Base) and OSDL.  They have been very successful in continuing work on LSD (fighting fragmentation) and in sponsoring a range of kernel engineering and other development activities.  They recently announced their acceptance of hosting MeeGo, the result of merging Nokia’s Maemo tablet platform with Intel’s netbook/MID Moblin project.  MeeGo targets a range of embedded/mobile applications, including mobile handsets.

So, Linaro is NOT a standards body, not a distribution supplier and not a mere cheerleader, as far as I can tell.  They seem to have a clear vision of what they want to do – enable Linux on real silicon.

To that end, they are not getting fancy, especially in terms of licensing.  Unlike LiPS, LiMo and others, they have pledged to adhere to existing licensing regimes and not indulge in license proliferation (beyond the profligate OSI corpus).  In particular, the Linaro IP Policy refreshingly stipulates

  • respect for and adherence to upstream licenses
  • commitment to use only existing, OSI-approved licenses

Cautious Optimism

Like embarking on a second (or third) marriage, launching a new .org for embedded Linux represents the triumph of optimism over experience. Despite (or because of?) my personal involvement with several of the .orgs above,  I believe that Linaro has achievable goals and the members and means to achieve them.  In particular, Linaro sets its sites on providing and improving infrastructure, an area where open source and .orgs have classically excelled.

So, keep your eyes on Linaro.  Not the horse, but much needed glue.


Cavium Acquires MontaVista – Embedded Linux Consolidation Continues

The original MontaVista Hardhat Linux penguinYesterday semiconductor supplier Cavium Networks announced its plans to acquire embedded Linux pioneer MontaVista Software.  MontaVista, founded in 1999 by Jim Ready (of Ready Systems / VRTX reknown) was among the first to commercialize and evangelize Linux for embedded designs.  It was one of the few remaining independent vendors in the embedded Linux business when Cavium snapped it up yesterday for $16M in cash and $34M in stock.

When Silicon Buys Software and Services

The acquisition is the latest in a string of M&A moves by silicon vendors hoping to gain an edge in filling sockets by providing software tools, platforms and services.  The most recent, most visible and most lucrative was Intel’s buy of Wind River earlier this year for a cool $884M, most palpably to support design-in of Intel’s mobile/embedded Atom architecture family.  Others include

  • Motorola Semiconductor (now Freescale) purchase of MetroWerks in 2003
  • Mentor Graphics acquisition of Embedded Alley earlier in 2009 (Mentor helps companies design semiconductors)

Independent embedded Linux companies remaining after Cavium’s move are much smaller players like TimeSys,  services provider Denx and Linux tools supplier Viosoft.

Why MontaVista, Why Now?

Cavium’s stated reasons for the buy are to “complement Cavium’s market leading processor portfolio” and to  “significantly increase Cavium’s software and services revenue”, which in 2009 could amount to a top-line bump of $30M according to Cavium’s investor call.  Unstated are likely concerns about sustaining Cavium’s traditionally close ties with Wind River for design wins with popular Cavium MIPS architecture CPUs for networking and other applications.

MontaVista, for its part, has reportedly been courting suitors for three years or more, after multiple funding rounds that totaled over $100M since the company’s founding in 1999.  The marriage with Cavium reflects the embedded Linux supplier’s long-standing ties to semiconductor suppliers, including  AMCC, Freescale, Intel, Marvell, and Texas Instruments, and processor IP providers ARM and MIPS Technologies.  Many were also strategic investors and sources of substantial historical enablement revenue.  It also probably reflects the state of the firm’s revenues and cash reserves.

Analysis – Acquisition a Bang or a Whimper?

The acquisition is certainly a bang for Cavium.  They get revenue growth, enabling technology, expertise and new ecosystem reach.  But for MontaVista?  They get financial security (for now) and a place in a strong and growing technology supplier.  However, this acquisition surely falls short of the “event” once envisioned by Ready and his many investors.

MontaVista made a strong start in 1999 and 2000, riding the wave of infrastructure build-out to support what later turned into the Internet bubble.  Even after that bubble burst, MontaVista continued to grow, complementing still-strong networking business with consumer electronics, mobile telephony and other intelligent device application segments.  They achieved an impressive series of “firsts” in bringing Linux and open source software (OSS) to the embedded space:

  • First commercial cross tools and fully embedded platform for Power, MIPS and ARM architectures
  • First support for redefinable CPU architectures
  • First to market with a Carrier Grade Linux platform
  • First mass deployments in dozens mobile handset designs and millions of handsets with MobiLinux


  • Key enhancements in real-time responsiveness of the Linux kernel
  • Investment in maintaining Linux kernel architecture trees, including especially PowerPC and multiple ARM family CPUs
  • Important advances in and contributions to  open source projects, including the Linux kernel and device drivers, threading libraries, power management, GDBserver and numerous others
  • Early support for embedded multicore architectures and designs
  • Industry leadership in evangelizing embedded Linux and providing assurances about the IP safety of embedded open source

So why did an acquisition occur not at first, but at last?  How was late starter Wind River able to enter the embedded Linux space in earnest after MontaVista’s five year head-start,  and eclispe MontaVista in Linux-based revenues and ultimately in valuation?

For all of the company’s “firsts”, MontaVista took numerous missteps, slowing its growth and causing it to miss multiple windows of opportunity:

Value-added:  for most of its history, MontaVista primarily acted as an integrator of OSS projects, treading water and often swimming below the ever-rising open source value line.  True innovation emerged from the company, but always so low in the stack (mostly in the kernel) that they were unsuccessful in commanding a premium for it.

Revenue Scaling: Because they packaged up and commercialized a broad array of existing projects, and marketed them as development seats to engineers, MontaVista revenue growth was always limited by their ability to capture development teams as customers. They resisted both developing deployment IP or reselling run-time technology from 3rd parties, limiting their opportunity to benefit from successful high-volume OEM customer products.  At one time the company did offer a per-unit licensing option for this aggregated open source content.  Primarily a response to requirements for risk-sharing from key customers, this short-lived selling model baffled many in the industry who assumed that OSS code could only accrue services revenues.

Sustaining vs. Enabling Revenue: MontaVista cultivated strong ties to semiconductor suppliers and other hardware vendors, and was successful for many years in charging a premium for hardware enablement and upstream migration of patches and other code to support CPUs, SoCs and embedded computers.  At various times, the company was more successful in extracting revenues for enabling reference hardware than for supporting OEMs in building product on those systems and silicon.  The result of such strong business development was a product line bloated with board support packages that never saw the light of day in shipping OEM products but added substantial time and costs to new releases and sustaining engineering.  This focus on enablement also served to alienate  partners over time when they could not justify ROI  for their NRE.

Business Model and Execution:  Many MontaVista watchers have argued that the company’s business model was essentially flawed. Certainly there is room for debate about the viability of going to market with a product built almost entirely from freely available OSS components (vs. complementing that platform with proprietary IP, etc.).  Such a model based on building with and for open source can devolve into less attractive high-overhead packaged service business in the face of a rising value line.

By contrast, I would argue that MontaVista insiders and its various detractors were in no position to critique the business model itself since that model (and its minor variations) was never really tested.  The model did not fail the company, but rather the company failed to execute on that model.

Failure to execute belies key assumptions about serving device OEMs with embedded Linux platforms and toolkits:

  • OEMs look to suppliers like MontaVista for productization of the latest Linux kernel technology, libaries, middleware and tools
  • OEMs expect frequent releases and deep expertise at many levels of the platform and tools
  • OEMs anticipate something “in the box” other than bits and bytes they can increasingly source directly from OSS project trees

While MontaVista made a strong start in all these areas, over time they reduced the  investments needed to meet these (not unreasonable) expectations. In the last five years, MontaVista Linux releases became fewer and farther between and did not closely track ongoing Linux kernel and other OSS project evolution.  The company lost most of its hallmark on-staff project maintainers, along with their insight and hands-on knowledge.  And the firm never made sorely needed investments in truly original differentiating technologies and products.

In closing, I remember my first encounter with the company shortly after its founding in the Spring of 1999.  I was doing a trade study of emerging embedded Linux with a colleague and we pondered the future of Jim Ready’s then-new company.  Based on the history of Ready Systems and its flagship VRTX RTOS (acquired by Microtec Research for a modest sum in 1994) we debated whether this new venture would rise to spectacular success or  ultimately stumble.

I guess we were both right.

Android Beyond Mobile. Way Beyond.

I recently published an article entitled “Android Beyond Mobile” in the venerable RTC Magazine, a war horse technical  publication focused on real-time and embedded computing:

Originally a niche platform for mobile handsets, Android is moving into a host of embedded applications building on Linux, Java and the desires of users to frequently bring new applications into existing embedded devices. [Read More]

In the mobile/wireless development circles I frequent, the piece has been well circulated, but I have to admit that  the concept has quickly gained currency, with or without my advocacy.  For example, at the recent ARM techcon3 developers conference, the Android workshops offered by Mentor Graphics for general embedded development were standing-room-only affairs.  I also just helped MIPS Technologies host their MIPS Linux Summit, where the focus was on Linux but much of of the buzz from IP licensees and developers alike came from Android on MIPS architecture designs in networking and multimedia.

This evening, I came across an article by my friend and colleague, Open IT Strategies blogger and professor at SJSU Joel West.  The article  highlights the next wave of Android disruption with a focus on the emerging eReader market, in particular on the Android-based Nook eBook from Barnes and Noble.

First phones, then netbooks, now eBooks, multimedia players, TVs, set-top boxes, DVRs — even industrial control and medical devices are turning to Android as an enabling platform, replete with applications from the Android Market and a growing ecosystem of developers and commercial players. I am almost ready to lay odds that within two years, Android will find its way into more intelligent devices outside of mobile telephony than it will in just mobile handsets.

Readers – am I crazy?  Please, tell me about your plans to deploy Android beyond mobile and how the Google/OHA platform is changing development and deployment of your intelligent device designs.

Intel to Acquire Wind River – Embedded Industry Realignment Coming

I woke up this morning to a train of email from friends and fellow pundits, intrigued and even aghast that Intel had announced its intentions of acquiring embedded industry leader Wind River Systems.  The purchase, while very much newsworthy, is not a huge surprise in an industry where low market caps and niche offerings are the rule.

Embedded Systems (or Device Software, as Wind River calls it), as a segment is defined not by its customers, but by the vendor community. Device OEMs self-identify as developers of networking equipment, consumer devices, automotive systems, medical devices, instrumentation, etc. — not “embedded systems” or “device software”.  This perennial identity crisis is part of the reason for my own saw – the embedded market is all tail and no body.

Plumpest Part of a Long Tail

That being said, Wind River has for several decades occupied the most massive part of the tail (closest to the rump) and enjoyed a clear leadership position.  Its revenues for device software have only been rivaled by Microsoft, but only if you add up designs for Windows Mobile, CE, NT/XP Embedded and secondary use of WindowsXP and other desktop/server variants on blades and embedded motherboards.  Not only have they led the overall segment, but their product line revenues also put them at the forefront of the RTOS market (with VxWorks) and the embedded Linux sub-segment (with Wind River Linux).

Wind Historically Blew off x86

One issue with the acquisition, from the embedded software side, is that historically Wind has never garnered the greatest share of its design wins on Intel silicon.  Far more successful, first for VxWorks and later for the company’s Linux offering, have been architectures like ARM, PowerPC (Power Architecture) and MIPS.  In the last year, the major thrust of the company has been towards mobile handsets, with visible investment in Android and LiMo products and services offerings — all built around ARM CPUs.

Certainly the Intel acquisition reflects the investment that the two companies have made in their partnership to promote Atom over the last year, in particular for automotive designs.  To be fair and accurate, Wind River has also enjoyed a number of key design wins on single-board computers and blades with Intel Architecture CPUs, especially for its Carrier Grade Linux implementation, at companies like Nortel.

Intel – Off and On Again with Embedded Silicon

Intel’s acquisition of Wind River follows multiple forays into the embedded market on the silicon side.  Each time they have enjoyed reasonable success but ultimately never took their eyes off their most lucrative businesses, enterprise servers, desktops and notebooks.  A friend at Intel once despaired that their “embedded chips, revenue-wise, were a wart on the butt of desktop and data center”.

While much hype surrounds the Atom family of processors, it is actually the company’s sixth (and probably best) entrant into the embedded processor market:

  1. Of course, the 8008, 8080 and 8085, all originally embedded silicon
  2. 8031 and 8051 CPUs — long-lived 8-bit microcontrollers, in the “pre-software period” of embedded history (assembly only)
  3. Bigger/better 16/32 bit 80860 and 80960 RISC CPUs, which ended up in a mix of storage, graphics and aerospace applications
  4. Repositioning low-end 386 CPUs for SBCs and custom designs
  5. Intel’s expensive and short-lived licensing of ARM cores as XScale mobile applications (PXA), network (IXP) and storage processors (the XScale brand and technology were sold off to Marvell in 2006)

This evolutionary record alone, however, doesn’t not justify the current investment in all thing Atomic, including buying our friends in Alameda.  What does obviously motivate these investments is the slow death of the PC desktop form factor, already manifested as flat desktop silicon sales, the recession notwithstanding.

Metrowerks II?

A friend of mine pointed out a structural similarity in this acquisition to Freescale (then Motorola Semiconductor) buying compiler and tool vendor MetroWerks way back in 1999 (these historical references are showing my age!).  While the analogy is attractive and superficially accurate, key differences include

  • The market caps with Intel and Wind are greater by 10x or more
  • Neither MetroWerks nor Motorola Semiconductor held the clear leadership positions of Wind River and Intel
  • The emergence of Linux and open source (especially GNU tools) led to the decline of MetroWerks CodeWarrier as much as immersion within Motorola

Industry Impact

This type of cross-segment consolidation in theory results in a powerhouse successor company, but also causes rippling realignments among other players, in this case embedded software and silicon suppliers.  The remaining players, traditionally fragmented and balkanized, are sure to build new ties to replace those with Intel and Wind, and to construct bulwarks against them.

Obvious question arising from the acquisition include

  • To what degree will Intel management constrain Wind River, over time, to focus on Intel architectures?
  • How successful will Atom be in volume applications beyond netbooks?  How will acquiring Wind help with designs for in-car, in-home, in-hand and in-strumentation?
  • Can Wind River retain international mega-customers like NEC and Samsung, whose semiconductor subsidiaries compete head-on with Intel?
  • Will embedded software companies like Embedded Alley, GreenHills, LynuxWorks, MontaVista, QNXTimeSys, et al. even continue to support Intel silicon with RTOS and embedded Linux offerings?  Are they sufficiently agile and creative to take advantage of opportunities created by the acquisition, or will they be swamped by it?
  • Will ARM and MIPS and their licensees, as well as Freescale and AMCC with Power Architecture, win or lose designs?  Can Atom compete with 1B+ annual ARM shipments?
  • What other consolidation will follow on this heels of interesting turn of events?

The recession has put enormous pressure on the players in embedded and mobile ecosystem.  It’s not that large a club and not that well capitalized.  More shoes will shortly drop, heads will roll.  Stay tuned.