Slashdot Mirror
Does Unisys Really Get It?
Joe Barr writes "There's an interesting story on NewsForge today about Unisys and its new-found love for Linux. In the story, Robin Miller interviews Unisys VP of engineering Anthony Gold and asks such delicate questions as how Unisys 'planned to make amends for its use of GIF patents against open source projects'? It's a good read, and in this day and age of software dinosaurs trying for peaceful co-existence with Linux, a very timely one."
Wow. This is the same Unisys as the "We have the way out" Unisys?...
Does the GPL protect against that? If Unisys contributes code with their own pending patents to a GPL'd work like Linux, would the GPL force them to give the Linux community the rights/license to those patents?
You could definately say that. I work for Unisys, and they are moving towards a more 'service-based' business model. Like IBM, they provide a 'solution', not just a product.
They find out what you need, they either make or buy a program, install it, support it. Buy the hardware, install it, support it, etc.
While it is trailing in IBM's shoes, its not a bad business model.
Now Unisys is still pretty big, but we did miss expectations for the first time in 4 years, and the stock price dumped, its now ~$9. And our stock is considered 'moderate gain, low-risk'.
Still, I'm new here, so I don't have much info on the "why's" of GIF.
I worked with some ES7000s when I was at Microsoft. They are a NUMA architecture, and there is very, very high latency across the crossbars: a 32-way is basically 4 8-ways with a very, very, VERY VERY high latency interconnect between them. You need to partition your app so that groups of threads execute on an individual group of 8 processors, NEVER cross the crossbar, or perf blows up.
Once you've done all that work to partition to only run on 8 CPUs, you might as well just scale out like Google does. You can't truly scale up.
Of course things get better all the time, and maybe Linux will be a better NUMA os, but scaling up with Unisys is really just easier and cheaper to do with scaling out.
I "inherited" a unisys ClearPath MCP mainframe admin job and hated everything about it. They never used any "open" standards, and tasks as simple as telneting into the box was made mode difficult since the environment only accepted the "unisys telnet", a proprietary version of telnet, which you need to buy a client license. The CANDIE environment was horrible, and the MCP environment actually existed as a VM within Windows NT 4.0. The MCP environment used NT for hardware recognition, drivers, etc., so not only did you have the overhead of the Windows environment on the system, you had a VM which prescribed to this model. Everything about Unisys was about making money for Unisys. Documentation needed a license to download, pay-for-use transactions, thier vewsion of Cobol (which is actually Sperry-RAND COBOL), their print servers, etc. I'm getting the willies just thinking about it all over again.
I wouldn't be surprised that Unisys would charge a transaction-based license for their Linux, or a Unisys-branded Linux licence (similar to SCO). Unfortunately, many East Coast (US) universities, especially in Pennsylvania, use their systems for accounting, grading, etc...
I have long years of experience with Unisys, back from the time they actually created fine technology (only way, way overpriced). Actually, I started working for them in the early 90s, about the time they started to devolve from a technology company into a "systems integrator", as they like to put it. I did a couple of small chores on series A equipment, but most of my work was done on the x86 BTOS, then CTOS, machines.
(Those were beautiful machines, by the way. You extended them by daisychaining "blocks" into a bus that started from the CPU and grew to the right. E.g., you snapped in a SCSI disk block, a video card block, etc., all without having to open the case or fiddle with cables. Your workstation just grew longer and longer with each hardware expansion---mine was just under a meter long. You created native "LANs" (called "clusters") daisychaining stations using the dual "cluster ports", zero configuration required. And every machine included high speed synchronous serial ports. Also, they had a really stable OS, I never saw a single crash. It was a bit weird, but in a creative, interesting way. Probably had to do with the mainframe background of the company, because it was completely different from UNIX inspired OSs. This thing was even more alien than the original MacOS. The text editor was lovely. One of my first Unisys workstations was a B28, which sported an Intel 286 @ 8MHz and a whole two megabytes of RAM, IIRC. I also remember using an Intel 186 processor on some BTOS box---yes, it really existed, it was sold, and used.)
Unisys management was never too bright, their pointy-hairness surpassed only by that of Commodore execs, IMO. And when they dropped their technology division, they fired almost every engineer. At least I can personally attest that every friend of mine that worked there, and knew two things about computers, was fired or left on his own. The execs and the salesmen stayed. Maybe they kept a couple of smart techs in Atlanta, but from what I've seen, from then on they've been hiring only inexperienced technical people (read: cheap interns) for when a project plan calls for that, and they fire them as soon as possible afterwards. And I've seen them blow projects, or lose money in penalizations, because they lacked competent techs to do the work, or even to timely warn their salespeople that they were being sold, and in turn selling, snake oil.
Today Unisys, the technology company, has been dead for over a decade. It is a company of salespeople, and they mostly trade with the Unisys name. You know, they deal with banks and airlines and the likes, customers who were buying Unisys stuff three decades ago, and still regard them as the company they used to do business with. They sell expensive software they buy from other companies (Microsoft being a major supplier, indeed), or consultancy projects where they hire experts in the market just for the event. And yes, the name is eroding, even as we speak. And there's not much left of it, again IMO.
That "ES7000" thing, well, it's a PC with 16/32 processors, large and ugly as a fridge, and it runs Windows. Which, by the way, does not seem to scale all that well on that many processors: I've seen several of those choke as application servers with just over 300 clients. To be fair, the application software was a hideous contraption that costs several hundred thousand bucks to license---so maybe the suckitude did not come from Windows entirely. Anyway, you may call the ES7K "innovative", but to me, having seen the stuff the old Unisys used to make, is only boring, and a bit sad.