Mike Deliman

Remarkable

"Space... the final frontier" is how the pre-recorded message from NASA to the Crew of Discovery started this morning.

This afternoon while covering the final departure of Discovery from the ISS, the Newsman said "Remarkable - a remarkable bit of technology and a lasting mark on history" - or something to that effect. I thought "Remarkable" - that's a word Spock used to describe fantastic new technologies that had never been seen before. Things like Ion Drive in an autonomous space probe, or a life form that lives on something we'd considered impossible.

Some remarkable technologies demonstrated recently in the news include things like Northrop Gruman's X47B UCAS, and the comet flybys of Stardust NExT and DI Epoxi, and ongoing announcements of potentially hundreds of earth-like planets being found by Kepler Space Telescope.

One of the remarkable technologies just delivered to the ISS by Discovery is Robonaut R2. R2 is a telepresence operated robotic astronaut / assistant. It's the first robot designed to use the same tools as the astronauts, and to be able to assist them inside and outside on space missions.

Some other note worhty technology demonstrations/events coming up - later this month  MESSENGER will orbit Mercury.  In July  Dawn will go into orbit around asteroid Vesta.  Both of these orbital insertions represent "firsts" for mankind - the first time we've orbited and surveyed the sun-side of Mercury and the first time ever orbiting an asteroid.

All of these devices contain within them a bit of technology called VxWorks.  It's probably being used somewhere in the infrastructure that gets this posting to your monitor.  VxWorks has unparalled maturty and flexibility, going into everything from that internet modem in the closet to autonomous vehicles and robots.  I think that's also kind of remarkable...

Autonomy and Robotic Advancement

A few months ago on the Robotics Virtual Trade Show in
my presentation I spoke about the evolution of robotic systems.  One thing I predicted was as time goes on, robots will become more independent and autonomous, performing more of the work humans and mainframe-style systems have traditionally performed. 

Among the robots I talked about were robots similar to NASA's
Stardust Spacecraft  Stardust is currently on course for a Valentine's Day rendezvous with Comet Temple-1.  Stardust includes a package that performs autonomous navigation, d to help keep the craft on it's flight path with reduced intrusion from ground-based systems.  Since Stardust is so far out from earth, there is significant delay in round-trip communications.  Stardust is able to detect problems in trajectory and perform course correction maneuvers on its own.

Last week, Northrop Grumman had  first-flight for an evolution in Un-maned Aerial Vehicles.  The
X47B UCAS   is an autonomous unmanned flight system.  It is not a wireless-joystick system.  It flies itself, taking off, figuring a course to it's way points, landing.  Without human intervention.

This, in general, is the trend of the robotic future.  There will be more "robots" in all sectors, the robots will become more autonomous, and perform more of the work regarding their duties than in the past.

In general a robot is an artificial entity that performs work for humans.  I broaden this definition to include telepresence systems (telerobotic surgeons, astronauts, and systems like aercams on the ISS), and lifestyle assistants (products such as drobo, tivo, and roomba).  My thoughts are that these consumer systems are both cutting edge, and primitive.

They are cutting edge in that they do some pretty incredible things in comparison with competing products that were available at the times these were invented.  They also pushed conventional wisdom about engineering and marketing, and created some cool and popular products.  They are primitive when you consider what technologies are available and how they could be used to upgrade these devices, and how these technologies will evolve over the next 5 years.  If you think about DVR boxes, you can get new DVRs from many makers,  yet used TiVos sell for over $200!  It's because TiVo can use the programming you record to automagically select other programs to record for you.  It anticipates and adapts based on your habits.  This capability differentiates it from other DVRs.  It does more work for you than others do.

This trend of more intelligence and autonomy will eventually blend its way into everything from automotive products to tourism.  It could lead to smarter cars that can detect when the driver's attention is lapsing or alert that it is unsafe to pass due to oncoming traffic.  There may someday be a system of telepresence pods anywhere on earth or even in orbit, where you could schedule a short vacation and experience a different world just by putting on a special helmet and gloves.  With such systems you could control a telerobotic diver and explore a sunken wreck, then change connections and view the Earth and stars from orbit... Possibly even while driving home from work.

Such systems require many things to make real, among them is vision and technology.  With the proper technology your vision may be wrought into reality.  WindRiver is proud to be involved with programs like NASA/JPL's Stardust and Northrop Grumman's X47 UCAS.  Congratulations to X47B on successful first flight, and Stardust's upcoming Valentine's day rendezvous.

Bringing Home the Science Bacon

MRO recently passed a milestone of bringing back more data than all previous deep space missions combined.  To do this MRO was equipped with a large dish antenna and powerful radio, and is running a more powerful computer than it's companions on and orbiting Mars.  It gets a fair share of antenna time from the DSN antennae on the ground.

Science means collecting data, processing it for it's information, and analyzing the information.  All of these activities can be greatly facilitated by some degree of computer assistance. When it comes to remote-operated scientists (rovers, gliders, orbiters, diggers, swimmers, crawlers, hover-ers..) the computer has to do all the work.

Where it comes to an autonomous robot, or even partially autonomous, a balance must be struck between the amount of data that can be brought back (telemetry bandwidth), the amount of data collected by the science instruments, the amount of processing it takes to turn the data into information, and the processing ability of the computer(s) in the robot.  If the largest gating factor is the bandwidth, and you can gain 10x the science return by processing the data first, you may limit how much data you collect in favor of processing it into information.  If your biggest limitation is processing power, you may just concentrate on getting data off of science instruments and ready for transmit back.  The typical deep space mission has both challenges, and more.

There are terrestrial applications that have similar challenges.  Remote sensing networks, deep sea ROVs and robots, and some industrial applications, as well as military applications, may need to operate independently for periods of time, storing data for upload at the "next" connection.  There could be limitations on storage, processing ability, power, connection bandwidth. 

Some of the solution could be to run multicore.  You may be able to run a multicore chip at a reduced clock rate to save power and gain more processing power than a similar single-core chip at full clock speed.  By increasing the processing power, and using mixed methods (SMP,AMP) as in Mark's "Sea Of Cores...Now What?" discussion, you may be able to handle both steering data and processing it into information. This could increase the science return of the mission.  But with multicore, how do you balance the trade-offs and where do you focus your increase in processing power?

By simulating the whole affair you may be able to gauge how fully loaded your system is.  If you have processing power left, and also important, data steering capacity left, you may be able to share your telemetry stream with another project.  This is part of what NASA has done with the satellites orbiting Mars and the MER and Phoenix robots. The robots collect their data and do what processing they can, storing the results.  They are designed to send data directly to Earth, and do.  When possible, they can also transfer data to the satellites, borrowing some of their bandwidth.  By using a little slack-time in the orbiter's schedules, they're increasing the science return of the robots on the surface.

Increasing the processing power of one key system can increase the data-product returns of an entire group of systems. It can increase the returns both in terms of the quantity of data and quality of information returned.  With careful planning, the "more remote" systems can be coordinated to take advantage of data-stream offload, perhaps doing some data processing, or just collecting more raw data.  Prudent use of a virtualized / multicore system could be one great way of doing this.

None of the current Mars robots are empowered with multicore - when they were designed multicore wasn't as prevalent as today, and there are still very few multicore chips designed for the rigors of space.  There are plenty of off-the-shelf chips with multiple cores available.  Terrestrial based applications can start benefiting from SMP/AMP systems today.

Virtually... Not Yours

So with all this virtualization going on, one computer can run multiple copies of an OS (or multiple OSs), all acting like they own the whole computer. With only some fancy shim layer keeping them from corrupting themselves and everything else, what are the implications for security?  Security is an issue that has been raised more frequently as computers have become more powerful and the network more pervasive.  I mean, viruses spread fast enough when the World Wide Web was mostly single computers with modems dialing up to servers.  In this day of broadband everywhere and constant connections,  doesn't virtualization present a special problem? Doesn't a machine running 10 virtual servers just mean that a virus gets to invade 10 servers for the price of entering one machine? Let the SPAM flow!  NOT!

The problems mentioned are part of an ongoing concern for both the desktop world and the embedded world.  There are standards, and evaluation labs, and  certifications that can be obtained.  They mean "This computing platform can replace several physical computers with one physical computer providing several virtual computing environments, and keep them sufficiently separate so as to protect each one from all the others."  Most of these - such as ARINC 653 and MILS, set forth specifications for systems to satisfy in order to obtain a certification.  These certifications can be issued according to the importance of the system being certified; for example to something like "DO-178B level A" or "EAL 6+" for a system of high criticality (life-critical functions, "if it fails, people die").  Though these standards may not use the word "virtualization" outright, the services they require can be provided by a robust virtualized system.

Among the things these standards can do for us:  provide for protections that would prevent a problem in any one virtual environment from affecting any other.  A virus could corrupt and crash a virtual server, the supervising software could shut down and destroy, and rebuild that virtual environment, starting with a clean slate.  "No Virus, Bad Virus...  NOT YOURS!"  This oversimplifies the whole idea, but you get the picture.  If one virtual machine crashes, it's not going to affect the entire system, the other virtual machines keep running.  You might lose that powerpoint presentation you were "researching", but the feedback from the UAV being displayed, the real-time map updates, and the status of your deployed troops will still continue unaffected by the loss of that document or the restart of the OS that was affected by it.

Today we're announcing  VxWorks MILS, currently under  evaluation to EAL6+.  Now then... what about multicore?  :-)

Virtually Yours...

This morning we've made a couple of announcements that will hopefully surprise and delight the Real Time computing community.  Wind River is releasing two new products that have their base in  virtualization: Wind River's Hypervisor, and Wind River's MILS 2.0.  Both of these products introduce platform virtualization, used in different ways.   Wind River's implementations allow for a "floatilla" of OS's" to be run amongst a "sea of Cores" - it can be run as a one-to-many, many-to-many, or many-to one configuration (that is with SMP, one OS can be run by many cores, the Hypervisor is capable of running / scheduling many OS's among many cores, and both the Hypervisor and MILS 2.0 are capable of running many OS's over one shared core).

Both of these products are very high-powered and sophisticated.  Both products are designed with an eye to the future.  The Hypervisor itself can be used in SMP, AMP, Supervised-AMP and cooperative multicore configurations, and can also be configured to paravirtualize several Operating Systems and run them all (time-share style) on a single core.  The MILS 2.0 product is designed specifically to paravirtualize OS's and run them on a single shared-core with a high degree of separation maintained by the MILS 2.0 separation kernel.

Many of our customers have built multiprocessor systems, sometimes because there wasn't enough horsepower on one board to handle the needs of their system, sometimes because of a need to keep data processing nodes separate (security).  With today's processors, you may be able to replace a multi-board system with either a single multicore board running Hypervisor or a single board running a modern (single) core with our MILS 2.0 separation kernel.  Both products allow a number of instances of VxWorks to run on paravirtualized platforms, or run without an OS at all (bare-metal or minimal executive, etc), and the Hypervisor also supports Wind River's Linux on paravirtualized platforms.

So what is virtualization?  I think of virtualization as a star-trek like copy machine, where you start off with a computer, and make "virtual" copies of it.  The copy can have anything that's already on the real machine, or a subset of those features.  The Hypervisor schedules those copies to run on the real machine, making them believe they are completely in control.  This can allow you to replace several older machines with one virtualized machine and get the same work done.

Some popular examples include VMware, Parallels, and QEMU.  These are all available for desktop machines, allowing users to create polymorphic environments.  These are mostly "type-2" hypervisors, meaning the hypervisor is a guest running under a main operating system and is subject to the whims of that OS.  Wind River's implmentations are "Type 1", meaning our Hypervusor runs as the only supervisory access code on the system and has absolute control over the system.

I'm pretty excited about this.  It brings supercomputing into the embedded realm!

Processing Paradigm: it's all about capacity

When I was a kid anything that was a computer, or had one in it, was pretty obvious. Computers weren't "just everywhere".  Now we've got multiple CPU chips on a board, multiple CPU cores on a chip, different kinds of cores on a single chip, multiple computer boards in a single chassis, ... it goes on.    These things are embedded in everything around us, all this hardware glued together to achieve something.
This brings us to software.  Software has to evolve to keep up with hardware, and with the needs of users.  With all these cores and chips and boards and systems running around, it gets a bit confusing.   Software is what enables everything from multi-processing to "poly-processing".  There's a lot of buzzwords about it - SMP, AMP, POS, VOS, Real time kernel, Separation Kernels... but what does it all mean?  And.. what's it for?

Eventually these things will all become even more blurry. As the realization that the "system" is a composite of the hardware and software all together spreads beyond the aerospace and defense industries, the various specifications will have to cover a fusion of SMP, AMP, and partitioned systems.

For now, though... what I'm talking about includes:
SMP - Symmetric  Multi-Processing - two or more cores running out of the same RAM, each processing parts of a shared task.   The cores may or may not run the same exact OS, but often do.
AMP - Asymmetric Multi-Processing - this is where several separate computers process data as parts of an overall larger system.  They may briefly share data, or have a limited amount of shared memory, but other than this each computer is a separate processing node.
SMP is comparatively  new, but AMP has been available through packages like vxMP for a long time.
SMP and AMP are multi-processor applications, where multiple computers are connected together to achieve a larger goal.  They both bring up issues of concurrency, and offer their own challenges for the implementer.

Way on the other end of the scale are what might be called "poly-processing computers" partitioned systems, where a single computer is virtually carved up and treated as if it's really several computers.   Each of those virtual computers can run a completely different OS - one for it's rich applications and ease of install, one for it's sophisticated communications abilities and firewall, perhaps even a third or forth to handle other kinds of processing. or devices.  You can already share a PC between Windows and Linux (for example) using virtual machine software technologies.  But did you know that similar things are being done with flight systems - using a single high-powered computer to replaces dozens of of single-purpose computers, to reduce weight, cost, and complexity? 

SMP and AMP are for systems where one CPU can not possibly handle the work load before it.   If the work can all be done in one "space" -simultaneously acquire data, process data in place, store data - SMP is a good fit, as all the processors run out of one pool of RAM.  One processor can be tasked with managing / storing the pools of data, another can perform the processing-in-place.

If the work should be handled in separate spaces - say the processing can't be done in-place for whatever reason - AMP would be a better fit. 

In the case of poly-processing (via vitalization, etc), this paradigm fits more where you want to take advantage of the extra processing power of a modern computer to have it replace a number of older computers (or other devices), to save on weight and expense of a newer system with similar function.  A Good examples of this could come from the shipbuilding and aircraft industries, where you have multiple separate mechanical and computational systems that may easily be replaced with smaller, lighter, more capable modern electronics.  Each of these separate systems might still need a small general purpose computer to control them, but that can be costly and add weight.   A single, modern, high-powered computer may be able to replace dozens of smaller, less capable systems.

With any of these paradigms it all comes down to: what is your application, how does it need to be accomplished, and does your system have enough processing capacity to achieve your goals.