Ray's 'Rants'

A hyperbolically structured, pathological analysis of genetic mutations in the small form factor board market*

By: Ray Alderman

At the VITA Standards Organization meeting in May, two companies (PCI Systems and Themis) started new projects to define and standardize small form factor (SFF) boards and electronic packaging (“cubes”) specifically for Mil/Aero applications. You’re probably saying, “This isn’t news. There are over 100 SFF specs out there, and Mil/Aero has been using some of them for a long time!” Well, yes and no.

Any partially intelligent being [dl: uh, oh] can perceive that the whole menagerie of SFF specs is woefully, irreversibly and terminally flawed for mil apps. [dl: but, but, but…] What’s more, they are all incurably afflicted with significant and substantial negative characteristics that make them unfit for military duty.  [dl: but, but, but...] That means they were all developed by commodity thinkers who hold Intel’s PC reference designs close to their hearts and have no clue about Mil app requirements. [dl: oyyyyyyyyyyyyy!]

For the sake of argument here, my definition of “small form factor” is anything smaller (in surface area) than a traditional 3U board. There are a host of problems with the traditional SFF boards: they do NOT plug into a high-speed backplane [dl: that has some advantages]; they all use some horribly deficient, terribly flimsy, notoriously fragile and consistently unreliable cable interconnect method [dl: but what about stackthrough connectors?] ; and the present SFF cards out there use the cheapest commodity poor-Asian-metallurgy-based connectors in the industry [dl: there goes another potential advertiser!] , making them about as resilient to shock and vibration as a house of cards. That’s understandable, since they were all designed for brainless consumer applications.[dl: somebody put a leash on this guy!] 

Also, I have grave concerns that any of those cheap connectors can handle signals at 5G and 10G. Additionally, the authors of the present plethora of SFF specs must have been asleep in thermodynamics class: snapping-on one of those ridiculously low-MTBF chip-fans or putting a fan somewhere on the chassis is their idea of advanced cooling techniques. Moreover, most of the present mundane applications for existing SFF boards use the unproven and risky RoHS-compliant components and solders. In short, the present gaggle of SFF specs inflict severe gastric distress on anyone who tries to shoe-horn those boards into anything more demanding than purely pedestrian, mundane and anemic applications. [dl: nawwwww]

The PCI Systems folks call their new proposal “Micro VPX” (to designate a form factor smaller than the present popular 3U VPX board). The module is 77-mm high (3.00 inches) by 110-mm deep (4.33 inches), the same basic size as a 2.5 inch disk drive. The actual PCB is held in an aluminum X-frame to provide conduction cooling surfaces and rigidity. A six-slot “cube” chassis is 101-mm high (4 inches), 114-mm wide (4.5 inches) and 152- mm deep (6 inches). The boards plug into a small high-speed backplane with 10G-capable Erni gas-tight connectors.

Themis calls their proposal “Nano-ATR” (for the Air Transport Rack used in Mil/Aero applications). The board is 55-mm high (2.1 inches) and 77-mm deep (3.00 inches), smaller than the PCI Systems board, and it includes a thermal interface on the PCB for conduction cooling. The board is not mounted in a metal carrier frame. A five-slot cube is 90-mm high (3.5 inches), 95-mm wide (3.74 inches) and 90-mm deep (3.5 inches again). So the proposed Themis cubes are smaller than those used by PCI Systems. Similarly, the Themis boards plug into a backplane using the Samtec SEARAY high-speed serial connector.

Both companies have solved the cooling problems and eliminated all the nasty cable connections prevalent with the present SFF form factors. They both use high performance (10G) gas-tight rugged connectors, the cubes are hermetically sealed (eliminating entry of contaminates), and they are both highly resilient to severe shock and vibration. Each one uses high-density backplane connectors to get I/O connections to the boards through the backplane and eliminate cables. Both these proposals have solved the nefarious problems associated with the commodity nature of all the present SFF boards and packaging.

From the macro standpoint, what we are seeing here is a transition to distributed processing systems in military vehicles. As we all know, this started in the commercial automotive industry some years ago. A typical car today contains 60+ processors, all dedicated to some specific function (throttle-body control, ignition control, fuel control, anti-lock brakes, drive-by-wire, etc). All those processors are interconnected with some slow serial interface (like CAN) in today’s trucks and cars.

You may remember that earlier this year, two major programs were announced that will use this concept of small form factor “cubes”: (1) the NASA/USAF plug-and-play satellite bus, and (2) the Navy/Honeywell program to replace all aircraft cockpit instruments with standardized Line Replaceable Units (LRUs) . Both programs use small, simple cube-shaped boxes full of conduction-cooled electronics connected together with a high speed serial connection (like optical fiber).

Both the PCI Systems and Themis proposals are aimed at military UAVs and vetronics applications, where space, weight and power utilization (SWAP) are severely restricted. In addition, military applications are much more demanding than commercial vehicles in tolerating temperature extremes, shock, vibration and contamination (dust, dirt, salt spray, corrosive compounds, radiation, etc). And military electronics are expected to run reliability for 10 to 15 years, not just until the warranty expires. Car dealers make more money fixing those failed processor modules in cars and trucks than they do from selling the vehicles. Such consumer thinking has no place in the military environment where lives depend on the reliable operation of their critical embedded systems.

Both Themis and PCI Systems have customers for their particular implementation of these new small form factor concepts, so the designs have been heavily reviewed and well tested. As you all know, the Pentagon terminated the Future Combat Systems Vehicle Program early this year. As that program is re-evaluated and reviewed, it could crank-up again in the next year or so. And if it’s funded again, we could see thousands of these little cubes from Themis and PCI Systems going into those combat vehicles. As we add more capabilities to the UAVs, these cubes are perfect for the confined space, weight limits and low power requirements of those platforms, too.

So, I think we can bid farewell and adieu to all those trashy [dl: oof!] SFF boards and specs that were designed for vending machines and telecom pico-cells. They just don’t have what it takes to serve our military.
______________________________

*Editorial disclaimer: The opinions expressed above reflect solely the opinions of the author and not of the editors, their families, clergymen, pets, college classmates or attorneys. Readers with differing opinions are invited to submit them to Open Architecture Review as guest editorials.

**************************************************

Why PCI Express stinks (for multiprocessing)

 

By: ra
December, 2009The crux of this problem is that Intel sees everything that connects to a processor (other than memory) as a peripheral. Consequently, it has only two stools of thought on multiprocessing architectures:1. use our single-chip multicore machines if you need multiple CPUs, or

2. use our incredibly dumb bridge chips to interconnect processors.

Bridges are how the parallel-bused PCI did multiprocessing, and the results were laughable. PCI Express just raises this historically retarded thinking to a higher level of architectural stupidity.

Intel wants to sell its homogeneous-multicore processors and its bridge chips. It doesn’t want other processors to be tightly coupled into the Intel domain. It believes that every application simply needs just one of their CPU’s and a bridge chip or two; and that, if you need multiple processors, then you should use their homogeneous-multicore machines.

That is how PCI Express looks at multiprocessing, and continuing to think this way carries the damaged DNA from the original PCI forward, infecting the next generation of machines with the froth of even worse feral hydrophobic architectural and performance horrors. If you should somehow need multiple discrete processors hooked together for a high-performance heterogeneous processing application like sigint, you must pay severe latency penalties if you adopt an Intel strategy and you must live with an incredibly lame programming model to implement it.

Pursuing the PCI Express model for microprocessing is the worst kind of design somnambulism. To begin with, PCI Express has no IPC (InterProcessor Communications) mechanisms like broadcast and synchronization. If you are hooking discrete processors together, PCI Express says you are engineering your system all wrong.

Out in the real world, however, there are applications where you need a generic CPU to execute some mundane applications code, plus a dedicated, unique CPU to handle communications protocols, or graphics, or streaming I/O from sensors, or DSP engines preprocessing signals in sigint, radar or sonar. If you do have one of these aberrant applications, then in the Intel model you must use bridge chips and consequently reduce the performance of the system to a crawl.

PCI Express is a peripheral interconnect that firmly suggests (i.e., dictates) that you use a dedicated, closed box for each segment of such an application. Then, you should hook those boxes together with something else (like Ethernet). In fact, there are PCI Express-to-Ethernet bridges out there that do just that. Sad to say, though, these introduce even more latency into the equation and kludge-up the programming model even worse.

Why is the PCI Express programming model so, shall we say, repulsive? The answer is the stupid root-tree structure, which further accentuates Intel’s trouble accepting multiprocessing architectures. At the heart of the issue is how PCI Express handles transparent silicon (basic I/O peripheral devices) vs non-transparent (another processor) silicon from a software standpoint.

There are plenty of alternatives to PCI Express for multiprocessing. The first interconnect actually capable of system-level multiprocessing functions, on a limited basis, was Infiniband (using RDMA, Remote Direct Memory Access). The best systems level multiprocessing interconnect available today is RapidIO (all the native IPC functions are in place) although it currently lacks high transmission rates. The 802.XX Ethernet folks are making noises about adding IPC mechanisms and RDMA to next generation 10G silicon to make it capable of supporting tightly coupled multiprocessing.

Whichever of these you choose, believe me: you do not want to design and maintain a multiprocessing system using PCI Express as your system-level interconnect. Your software guys will hunt you down and kill you.

**********************************************

The $2 billion ATCA market: a big lie

By: ra
November, 2009

Last month, I got some email invites to an ATCA conference held in October. In the hyped-up pitch for the show, there was a claim that the ATCA market is $2 billion and exploding. Well, I have a problem with that, considering that the U.S. GDP was down 6.2%, 5.7% and 1.0% in the past three quarters, and the EU GDP was down 1.7%, 4.9% and 4.6% in the same period.

I could be compelled to agree that there MIGHT be some ATCA business in the EU, since about 50% of their GDP comes from government spending (closer to 70% in the Scandinavian countries). I doubt there’s much telecom business in the U.S., where government spending is only about 22% of GDP (and rising). I have trouble believing that people in undeveloped nations are buying cellphones and surfing the net when so many of them are dying of hunger, disease and criminal violence. And in China, they BUILD telecom equipment; they don’t buy it from U.S. or European telecom companies. So, clearly, something is wrong here.

If you take the top eight board companies in the world (Kontron, GE Embedded, Curtiss-Wright, Mercury, Radisys, ADLink, Advantek, and Emerson), you MIGHT come up with $2 billion in board/system sales collectively. Consider for the moment that VME is still about 25 to 30% of the worldwide merchant board market and that many of the companies above are selling commodity motherboards and small form factor stuff at very low prices and margins, and it becomes clear that something is wrong with that $2 billion ATCA number.

If you look at the macro-level, recent Wall Street reports claim that AT&T is broke and about to become the next GM. In fact, AT&T has bigger pension and healthcare obligations that GM ever had! It has laid-off about 22,000 workers so far in 2009, and its free cash flow is declining at about a 0.5% rate.

According to the latest figures, landlines are being disconnected at a rate of about 700,000 per month now, and AT&T is losing all that “free” revenue from low-maintenance lines which were depreciated years ago. The cellphone market has already become a replacement market and this hindrance to growth will become even worse when the developed nations hit saturation, which is forecast to occur in 2013. As for cellular subscriptions, they have already saturated for the carriers, and cellphone services are themselves becoming a replacement market, in which carriers have to concentrate on targeting competitors’ customers by lowering prices.

Also consider that a few million iPhones on the AT&T network caused massive crashes and all kinds of technical havoc. One report I read said that AT&T has to buy $18 billion worth of gear to create the needed bandwidth for the new smart phones and to keep serving existing customers, but they could see only $6 billion in new revenue. Yes, AT&T has to buy some gear and create some bandwidth in a period when consumers are very stingy with their money. I don’t see them spending money like a drunken sailor in a Hong Kong bar on ATCA gear anytime soon. So, something is wrong with that ATCA number.

Verizon is also in bad shape and predicted to be the next Chrysler. They have laid-off about 16,000 people this year. Their free cash flows are declining at about a 0.4% rate. They also have huge pension and healthcare obligations and are subject to the same declining business environment as AT&T. The rest of the carriers are all in the same boat. So, something is definitely wrong here.

Now, it could be that whoever is claiming that $2 billion number for ATCA is accidentally double-counting a lot of components. That is, they count the connectors several times as they move from the connector company to the backplane maker, to the chassis maker, and then to the systems integrator. Perhaps the same thing is happening to the processor boards and line cards and power supplies and fans and backplanes and metalwork: they all get counted several times. That could account for some part of this terribly fictitious number.

Actually, I think the reason for this ridiculous number is more complex. I believe that if ANY company makes a card cage, backplane or daughtercard that is within plus or minus about 20% of the ATCA dimensions, then people intentionally count it as ATCA product, regardless of pin-out, connector used or end application. The explicit purpose is to overstate and hype-up the market for ATCA. This counting technique would also qualify a portion of the CEM (Contract Electronic Manufacturer) product output, even though it’s not ATCA or anything remotely close.

Now, if you look at the traditional board guys, you have a very hard time finding any more than $200 million in ATCA sales, and that is being very generous. I would say they make no more than about $150 million per year, collectively. So, in my opinion, the ATCA market is overstated by at least a factor of 10. And there’s an odd coincidence in that.

If look at Chapter “10″ of Mein Kampf, Adolf Hitler states: “All this was inspired by the principle—which is quite true in itself–that in the big lie there is always a certain force of credibility; because the broad masses of a nation are always more easily corrupted in the deeper strata of their emotional nature than consciously or voluntarily; and thus in the primitive simplicity of their minds they more readily fall victims to the big lie than the small lie, since they themselves often tell small lies in little matters but would be ashamed to resort to large-scale falsehoods. It would never come into their heads to fabricate colossal untruths, and they would not believe that others could have the impudence to distort the truth so infamously. Even though the facts which prove this to be so may be brought clearly to their minds, they will still doubt and waver and will continue to think that there may be some other explanation. For the grossly impudent lie always leaves traces behind it, even after it has been nailed down, a fact which is known to all expert liars in this world and to all who conspire together in the art of lying.”

Now, look at the top board vendors in the MIL/COTS segment, and you can see that there is a verifiable $600 million worth of true VME business. Most of those vendors tell me that about 80% of their sales are VME, with a small percentage in 3U cPCI and a small but growing segment in VPX. So, using the 10X “big lie” factor above, I would like to state for the record that the VME market, and anything using the IEEE 1101 mechanical and DIN connector standard (originally Chapter 7 of the VME spec), is at least $6 billion this year.

There. Whatever was wrong with the $2 billion ATCA number is now fixed and put in proper perspective by the $6 billion VME number. You got a problem with that?

OAR: Are you claiming that, unlike ATCA, there are no such variations on VITA specs?

RA: Yes, of course there are, but we do not count them in the merchant board market totals.

********************************

Check out Rays past ‘Rants’

Wind River becomes Intel. Watch out below!

Contact Ray at exec@vita.com