Thin clients vs. Smart Displays

If you want to understand what happened to Sun, look no further than Sun Marketing’s insistence on selling the Sun Ray as a thin client - something it isn’t. Thin clients have few organizational consequences, smart displays have many - and that’s the distinction this is about.

This entry is about correcting a common misconception about the Sun Ray - and, to do it out of turn I’ve moved the Unix job description entry to next week.

Many of the comments, both those written to the blog and those received privately, came from people who clearly think of the Sun Ray as a thin client. It is not - and Sun marketing’s willingness to cater to this market misperception was, I think, diagnostic for its wider failure to move leading edge product.

A few years ago one of the wintel companies offered a PC architecture in which the graphics board was connected to the motherboard by cable. This enabled the customer to put the PC in a data center rack for easy access while putting the graphics controller, keyboard, mouse, and monitor on the user’s desktop.

Notice that the remote graphics board for the PC can’t be considered a “client” in a client-server sense because it does no application processing and is really just the normal display management component from the PC with the local motherboard connection stretched out as a cable.

Think of Sun Ray as the multi-user, multi-host version of this and you’ll understand the key to its simplicity of operation.

Thin clients, in contrast, attempt to do at least some local processing and run some local OS - even if, like Sun’s mid eighties diskless workstations, that OS is downloaded from the server at boot time. That can make them harder to abuse, but a client is a client and the complications arise from the architecture, not the nature of the client - thus both PC style thin clients and Linux desktops offer some benefits relative to the traditional wintel approach, but neither offers dramatic change and neither choice ultimately affects organizational structure and behavior.

As usual there’s history to the distinction: back in the eighties when Unix vendors like Sun and Apollo experimented with thin clients their actions were mostly motivated by the cost of local disk - then over $1,000 for 30MB devices - and ended when disk prices fell much faster than progress was made in reducing the operational complexity of the set-up.

In the alternative approach the Plan 9 people at AT&T invented the Gnot as the first real network display, Sun developed NeWS, and NCD started its first line of X-terminals. Gnot never went commercial and NeWS fell to Adobe’s licensing demands on PostScript, but NCD succeeded both technically and commercially until it eventually fell victim to a VP with a Microsoft driven NT infatuation and started making cheap thin clients instead of high end smart displays.

Specifically the original NCD network computer offered only an X-server that handled user interaction and absolutely nothing else - providing 24 bit color at 1600 x 1200 on 21″ screens at a time when the PC press was erupting enthusiasm over 13 inch greyscale screens at 640 x 480 - and despite the fundementally brain damaged nature of X, some NCD HMX terminals are still in use today.

In contrast NCD’s initial venture into the thin client world, software for an x-terminal capable of connecting to NT and running the Mosaic browser locally, ultimately led to what is now Citrix and various licensed thin client products, but also signalled the end of the company’s commitment to technical leadership, network computing - and profitability.

At Sun meanwhile, Bill Joy’s MAJC chip design was intended to power a new generation of super terminals but his ability to get the CPU made didn’t extend to getting a corporate commitment to the new desktop, and so we got the ill conceived and ill fated Java Station -a seriously overweight and under powered “thin” client- instead.

The Java Station was both a technical and a commercial disaster - and would have faded quietly into history if a few recidivist engineers hadn’t modified the Solaris X/Postscript display software for download, hung a Java Station with all the client code stripped out at the end of it, and called the result a Sun Ray.

To repeat: what they’d done was take a thin client and turn it into a smart display by taking out its ability to run anything locally. That’s what makes a smart display smart: lots of graphics power, no local code - basically a recreation of the NCD network computer but latterly with faster hardware, better software, better branding, and a more focussed security agenda.

Although Sun “sales” still calls the thing a thin client and there are always people trying to impose some local processing on it, the current Sun Ray 3 and its matching software is still very much like that early device: no local processing ensuring both no local hassles and complete portability, while better hardware and server software mean it can display almost anything - from real time Unix/HPC imagery to Wintel and MacOS applications.

Notice that the bottom line here is simplicity and the freedoms and reliability you get from that: no local processing means no ambiguity (and therefore no help desk), no software limits, and no desktop product churn: just load up the applications and trust Unix to run them, whether you have one user or thousands.

SiOx memory: How it Works

The new memory is non-volatile, offers fast sub-100 ns switching times, can be written 10,000 times and is fully compatible with current CMOS manufacturing processes. A 1,000 bit proof-of-concept chip has been built by a private company. But those aren’t the good parts.

The recent announcement that Rice University grad student Jun Yao has demonstrated a new memory device has created a stir. As well it should.

The new device is non-volatile, offers fast sub-100 ns switching times, can be written 10,000 times and is fully compatible with current CMOS manufacturing processes. A 1,000 bit proof-of-concept chip has been built by a private company.

Those specs are better than or equal to current MLC NAND flash, but there are 3 other important advantages:

1. Size: devices are only 5 nm wide - 1/5th the feature size of the latest flash devices - which means much higher storage capacity.
2. Capacity: the architecture lends itself to stacking multiple dies - so-called 3D chips - to create even higher capacity devices.
3. Simplicity: it is a 2 terminal memory, not 3 as in most memories. This reduces device size and complexity.

The device
The device uses silicon oxide (SiO_x), a universal component of semiconductor devices for decades, in a novel way. The SiO_x is used to create a conductor - not an insulator.

From the Rice press release:

Applying a charge to the electrodes created a conductive pathway by stripping oxygen atoms from the silicon oxide and forming a chain of nano-sized silicon crystals. Once formed, the chain can be repeatedly broken and reconnected by applying a pulse of varying voltage.

I did not find a mention of the voltage needed to form the chain, but given the feature size and mechanism I’d expect it to be much less than the 20 volts required to pump NAND flash. If correct that should also reduce the chance of catastrophic die failure when the insulation shorts out.

Here’s a graphic that starts at the chip level and goes down to the nanocrystal level:

Why the limited write/erase lifespan? Presumably the local region runs out of nearby oxygen atoms, stopping the process, accounting for the ≈10,000 write limit. Clever materials or manufacturing process engineering might increase that limit.

The Storage Bits take
It’s great to see something novel found in such a common material as SiO_x. Professor James Tour, in whose lab Jun Yao works, says SiO_x is one of the most studied materials on earth.

Flash designers have been sounding alarms because they aren’t sure they can go below 20nm feature sizes - a fast approaching limit. Of course, storage and semiconductor engineers have been sounding alarms for decades: that’s how you keep the suits funding research.

But this development has great promise not only for its theoretical capabilities, but because it creates technological competition. We’ll all benefit from that.

Microsoft confirms 17-year-old Windows vulnerability

One day after a Google security researcher released code to expose a flaw that affects every release of the Windows NT kernel — from Windows NT 3.1 (1993) up to and including Windows 7 (2009) — Microsoft dropped a security advisory to acknowledge the issue and warn of the risk of privilege escalation attacks.

Microsoft warns that a malicious hacker could exploit this vulnerability to run arbitrary code in kernel mode. For an attack to be successful, the attacker must have valid logon credentials.

The flaw does not affect Windows operating systems for x64-based and Itanium-based computers, Microsoft said.
According to Tavis Ormandy, the Google researcher who released the flaw details, Microsoft was notified about the issue in June 2009. After waiting several months and not seeing a patch, he decided it was in the best interest of everyone to go public.

As an effective and easy to deploy workaround is available, I have concluded that it is in the best interest of users to go ahead with the publication of this document without an official patch. It should be noted that very few users rely on NT security, the primary audience of this advisory is expected to be domain administrators and security professionals.

Ormandy’s advisory includes instructions for temporarily disabling the MSDOS and WOWEXEC subsystems to prevent an attack from functioning. This can be done via Group Policy.
The mitigation in Microsoft’s advisory mirrors the advice from Ormandy.

IPv4 addresses: They are almost gone

The Number Resource Organization (NRO) represents the five Regional Internet Registries (RIRs) and is responsible for the unused IPv4 addresses as explained in their charter:

“The NRO exists to protect the unallocated Number Resource pool, to promote and protect the bottom-up policy development process, and to act as a focal point for Internet community input into the RIR system.”

That’s great, but why tell me, you may ask. Well, they feel the IPv4 address space has reached a critical juncture. That is, the number of remaining IPv4 addresses is less than 10 percent. If you are a clock watcher, you can keep track of the estimated days left before they are all gone at the Internet Society Web site.

What’s 10 percent?

The IPv4 addressing scheme consists of a 32-bit address space. According to RIPE that means IPv4 address space is 32-bits (2³²) in size and contains 4,294,967,296 addresses. At the time of this article, my iPhone app showed that 402,291,729 addresses (9.4 percent) remained. According to the app’s count-down meter, all the addresses will be gone in 593 days.

Is it a problem?

those that believe it is a problem and those that don’t. Let’s look at both viewpoints before deciding who’s right.

Will run out

NRO raised the alarm, so they definitely feel it’s a problem. Here is what Axel Pawlik, Chairman of the NRO says:

“With less than 10 percent of the entire IPv4 address range still available for allocation to RIRs, it is vital that the Internet community take considered and determined action to ensure the global adoption of IPv6.

The limited IPv4 addresses will not allow us enough resources to achieve the ambitions we all hold for global Internet access. The deployment of IPv6 is a key infrastructure development that will enable the network to support the billions of people and devices that will connect in the coming years.”

According to my research, most agree with Pawlik. The IPv4 address space is rapidly depleting

Not running out

The people I’ve talked to, who feel we are not running out of addresses, tend to agree with what Steve Gibson talked about in this podcast. His argument is that NAT routing reduces the pressure to move to IPv6 and will continue to do so. So IPv4 addresses will not run out and the timeline for moving to IPv6 is unclear.

Since the podcast was two years ago. I tried (unsuccessfully) contacting Steve Gibson, to see if he still feels the same. Regardless, many businesses and organizations are hoping he is right. Switching everything to IPv6 is expensive and there is a steep learning curve.

Compare to IPv6

The replacement addressing system, IPv6, uses a 128-bit address space. With RIPE’s help again, the IPv6 address space is 128-bits (2¹²⁸) in size, containing 340,282,366,920,938,463,463,374,607,431,768,211,456 addresses. That seems like enough.

Final thoughts

Experts have differing opinions about what to expect when available IPv4 addresses become fewer and fewer. Some feel NAT will become commonplace at ISPs, large and small. Others say this is a wake-up call and IPv6 will gain momentum. Either way, it will be interesting.

My real concern is for the people who this directly affects. You know, the ones that have to make it work. As I mentioned earlier, IPv6 requires some effort to learn. For help in that regard, check out Charles Kozierok’s Web site. It helped me get up to speed.

LTE or WiMAX: Why not both?

The story starts with Beceem Communications, a well-known manufacturer of semiconductors. Not quite an Intel, but a force to be reckoned with when it comes to 4G. Their Web site mentions:

“Beceem was founded in October, 2003, as a fabless semiconductor company specializing in the emerging 4G-WiMAX marketplace. We offer baseband & RF chips as well as complete hardware & software solutions.”

That’s certainly true, considering Beceem was selected by Clearwire to provide chipsets for their mobile WiMAX devices. Needless to say, that’s a lot of chips.

Some history

There are two divergent ideologies vying to be the next cellular standard, — the technology allowing cell phones to communicate and exchange data. WiMAX is one. Albeit the new kid on the block, it has significant traction. Sprint is the major proponent for WiMAX and betting the house on it.

LTE is the other standard. It’s more in line with existing telecom technology and why providers like AT&T and Verizon are in favor of it. I personally feel that LTE will dominate, simply because most telecom service providers are behind it.

BCS500

Now let’s get back to Beceem. This year’s Mobile World Congress was held in Barcelona, Spain and that’s where Beceem introduced the BCS500. The BCS500 is a chipset that integrates LTE and WiMAX technology into one device.
Once that sunk in, I realized the significance. It completely redefines what is possible. In their news release, Beceem’s Vice President of Marketing Lars Johnsson mentions:
“Our BCS500 will end the 4G debate by connecting to any 4G LTE or WiMAX network with seamless roaming, and switching between TDD and FDD configuration as needed, freeing operators from concerns how best to utilize their available spectrum assets.”
Here are a few of the more prominent features built into the BCS500:

• The chip supports the 16e and 16m revisions of the IEEE 802.16 standard.

• BCS500 also supports the 3GPP-LTE standard, Release 8.

• It is the only device chip to support UE Class 4 capabilities.

The killer feature is how the BCS500 can enable real-time band/channel reconfiguration by using what Beceem calls “multi-mode autosense”. That means the 4G device can automatically detect available LTE and WiMAX networks and switch seamlessly between them.
The Beceem announcement quotes Caroline Gabriel, Research Director for Rethink Research Associates:
“This advancement of Beceem’s 4G chipset dramatically helps the entire ecosystem by adding flexibility to network planning, equipment deployment and utilization, device manufacturing, and much more. The BCS500 chip solution will be an industry ‘game changer’ in terms of the interoperability that it enables.”
Future plans

It was also announced at the Mobile World Congress that Beceem and Motorola are collaborating to pair the BCS500 with Motorola’s WiMAX 4G network. Bruce Brda, Senior Vice President and General Manager, Networks Motorola explains:
“The know-how gained during the development and deployment of our leading 4G WiMAX solution has contributed to our work with Beceem to offer this advanced chipset solution. We look forward to the opportunities provided by this capability to address market demand.”
It’s my understanding that Beceem and Motorola will start testing devices later this year, with plans to mass produce equipment in early 2011.
Final thoughts
Imagine the telecom service provider’s relief. They have more latitude as to which technology to choose, LTE or WiMAX, without having to worry if they made the wrong decision, ala Betamax versus VHS. Providers may now be willing to ramp up their deployment plans. This is good news for us consumers, since both LTE and WiMAX address the bandwidth problems we are encountering. BCS500 does appear to be a game changer.